Pre-decision method and device for ensuring online voltage stability in power systems

Abstract

This application provides a pre-decision-making method and device for online voltage stability in power systems. The method includes: inputting the current operating state and current fault information of the power system into a pre-decision-making strategy model to obtain a current pre-decision-making strategy and a current pre-decision-making action; inputting the current operating state, current fault information, and current pre-decision-making action into a pre-decision-making risk assessment model to assess the risk of the current pre-decision-making strategy and obtain a risk probability; if the risk probability is greater than or equal to a risk threshold, combining the pre-decision-making strategy model and the pre-decision-making risk assessment model to perform a safety correction on the current pre-decision-making action and update the current pre-decision-making strategy to obtain a target pre-decision-making strategy; and making a pre-decision based on the target pre-decision-making strategy. This application solves the problem of low security in existing transient voltage stability pre-decision-making strategies.

Description

Technical Field

[0001] This invention relates to the field of power system pre-decision technology, and more specifically, to a pre-decision method, pre-decision device, computer-readable storage medium, and computer program product for ensuring online voltage stability in a power system. Background Technology

[0002] Power system pre-decision-making is the process of formulating stability control strategies in advance for a set of possible faults, playing a crucial role in the power system stability control defense line. Traditional voltage stability pre-decision-making methods rely on extensive simulations and traversals, which are highly dependent on accurate power system modeling and simulation techniques. With the continuous increase in new energy sources and power electronic devices, accurate modeling of power system dynamics has become difficult. Furthermore, traditional pre-decision-making methods involve excessively long traversal computation times, hindering online applications. Some reinforcement learning-based methods alleviate these two problems.

[0003] However, in recent years, with the continuous expansion of power system scale and the increasing proportion of new energy in power systems, the possible operating modes and dynamic characteristics of power systems have become more diverse and complex. Moreover, the security constraints for power system pre-decision are nonlinear and dynamic, and there is a lack of research on the security of power system transient voltage stability pre-decision. Currently, for the operating constraints of power systems, most methods adopt explicit Lagrange multiplier transformation constraints or construct Lyapunov functions, or use methods to linearize the constraints. These methods can indeed effectively utilize the constraints and make the decision safer and more reliable. However, in complex power systems, constructing Lagrange multipliers or Lyapunov functions is relatively difficult, and approximating nonlinear constraints as linear constraints carries the risk of information loss. Therefore, a method that can efficiently process constraints and quickly correct actions is needed. Summary of the Invention

[0004] The main objective of this application is to provide a pre-decision method, pre-decision device, computer-readable storage medium, and computer program product for online voltage stability in power systems, so as to at least solve the problem of low security in the prior art of transient voltage stability pre-decision strategies.

[0005] To achieve the above objectives, according to one aspect of this application, a pre-decision method for ensuring online voltage stability in a power system is provided, comprising: inputting the current operating state and current fault information of the power system into a pre-decision strategy model to obtain a current pre-decision strategy and a current pre-decision action, wherein the pre-decision strategy model is a model trained using multiple sets of first training data through a deep convolutional network, each set of first training data including sample operating state, sample fault information and corresponding sample pre-decision strategy, and the current pre-decision action being a variable generated during the execution of the pre-decision strategy model to ensure that the current pre-decision strategy meets set safety constraints; inputting the current operating state, current fault information and current fault information of the power system into a pre-decision strategy model to obtain a current pre-decision strategy and a current pre-decision action; and inputting the current operating state, current fault information and current fault information of the power system into a pre-decision strategy model to obtain a current pre-decision strategy and a current pre-decision action. The pre-decision action is input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain a risk probability. The pre-decision risk assessment model is a model trained using multiple sets of second training data through the deep convolutional network. Each set of second training data includes the sample's running state, the sample's fault information, the sample's pre-decision action, and the corresponding sample risk probability. If the risk probability is greater than or equal to a risk threshold, the pre-decision strategy model and the pre-decision risk assessment model are combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain a target pre-decision strategy. A pre-decision is then made based on the target pre-decision strategy.

[0006] Optionally, before inputting the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action, the method further includes: building an initial pre-decision strategy model based on the deep convolutional network, wherein the initial pre-decision strategy model is a pre-decision strategy model to be trained; an input step, wherein first target training data is input into the initial pre-decision strategy model, and preliminary training is performed using a deep deterministic policy gradient algorithm to generate corresponding training pre-decision actions, wherein the first target training data is any set of the first training data; a first calculation step, wherein the action gradient corresponding to the training pre-decision action is calculated using the pre-decision risk assessment model; a first correction step, wherein the training pre-decision action is corrected for safety based on the action gradient to obtain the sample pre-decision strategy; and repeating the input step, the calculation step, and the correction step at least once in sequence until all the first training data are input into the initial pre-decision strategy model to complete the training, thereby obtaining the pre-decision strategy model.

[0007] Optionally, calculating the action gradient corresponding to the training pre-decision action using the pre-decision risk assessment model includes: calculating the action gradient according to a first formula, wherein the first formula is... f θThe pre-decision risk assessment model, s t F represents the running status of the sample. i This indicates the sample fault information. Let a(n) represent the training pre-decision action after the k-th correction at time t, and let a(n) represent the n directions of the action safety space to which the training pre-decision action is projected. The action safety space is the space in which the sample pre-decision strategy satisfies the set safety constraints.

[0008] Optionally, before inputting the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and current pre-decision action, the method further includes: building an initial risk assessment model based on the deep convolutional network, wherein the initial risk assessment model is a pre-decision risk assessment model to be trained; inputting multiple sets of the second training data into the initial risk assessment model in sequence, and training it using a deep convolutional network algorithm to obtain the pre-decision risk assessment model.

[0009] Optionally, before combining the pre-decision strategy model and the pre-decision risk assessment model to make a safety correction to the current pre-decision action when the risk probability is greater than or equal to the risk threshold, the method further includes: determining the action safety space of the current pre-decision action according to a second formula, wherein the second formula is... Let f be the action safety space, used to represent the space that satisfies the stability constraint f. θ (s t a t s t+T All feasible pre-decision actions ≤ ε, f θ (s t a t s t+T )≤ε represents the stability constraint of the transient voltage, ε represents the risk threshold, and a t Let s be the pre-decision strategy at time t. t Let s be the operating state of the power system at time t. t+T The operating state of the power system at time t+T, where A represents the set of emergency control strategies corresponding to all faults; the boundary of the safety space of the current pre-decision action is determined according to the third formula, which is: It represents the nonlinear boundary of the feasible pre-decision action space.

[0010] Optionally, if the risk probability is greater than or equal to a risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to perform a safety correction on the current pre-decision action, including: a second calculation step, using the pre-decision risk assessment model to calculate the current action gradient corresponding to the current pre-decision action according to the first formula, and outputting the current risk probability; a second correction step, transmitting the current action gradient to the pre-decision strategy model, and using gradient descent to correct the current pre-decision action according to a fourth formula, wherein the fourth formula is: Indicates the current action gradient. This represents the current pre-decision action after the k-th correction at time t. This represents the current pre-decision action after the (k+1)th correction at time t, where λ is the correction step size, indicating the degree of each correction. The second calculation step and the second correction step are repeated at least once until the current risk probability is less than the risk threshold, thus completing the safety correction of the current pre-decision action.

[0011] Optionally, after inputting the current operating state of the power system, the current fault information, and the current pre-decision action into the pre-decision risk assessment model to perform risk assessment on the current pre-decision strategy using the pre-decision risk assessment model and obtain the risk probability, the method further includes: if the risk probability is less than the risk threshold, determining the current pre-decision strategy as the target pre-decision strategy.

[0012] According to another aspect of this application, a pre-decision-making device for stabilizing online voltage in a power system is provided. The device includes: a strategy generation unit, which inputs the current operating state and current fault information of the power system into a pre-decision-making strategy model to obtain a current pre-decision-making strategy and a current pre-decision-making action. The pre-decision-making strategy model is a model trained using multiple sets of first training data through a deep convolutional network. Each set of first training data includes a sample operating state, sample fault information, and a corresponding sample pre-decision-making strategy. The current pre-decision-making action is a variable generated during the execution of the pre-decision-making strategy model, used to ensure that the current pre-decision-making strategy meets set safety constraints; and a risk assessment unit, used to assess the current operating state, the current fault information, and the current pre-decision-making action of the power system. The action is input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of second training data includes the sample running status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability. The correction unit is used to combine the pre-decision strategy model and the pre-decision risk assessment model to make a safety correction to the current pre-decision action and update the current pre-decision strategy when the risk probability is greater than or equal to the risk threshold, so as to obtain the target pre-decision strategy. The pre-decision unit is used to make a pre-decision based on the target pre-decision strategy.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any of the methods described.

[0014] According to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement any of the methods described.

[0015] Applying the technical solution of this application, in the pre-decision method for ensuring online voltage stability in a power system, firstly, the current operating state and current fault information of the power system are input into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained using multiple sets of first training data through a deep convolutional network. Each set of first training data includes a sample operating state, sample fault information, and a corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model, used to ensure that the current pre-decision strategy meets the set safety constraints. Then, the current operating state, current fault information, and current pre-decision action of the power system are input into... In the pre-decision risk assessment model, the risk of the current pre-decision strategy is assessed using the aforementioned pre-decision risk assessment model to obtain the risk probability. The aforementioned pre-decision risk assessment model is a model trained using multiple sets of second training data through the aforementioned deep convolutional network. Each set of the aforementioned second training data includes the aforementioned sample operating status, the aforementioned sample fault information, the sample pre-decision action, and the corresponding sample risk probability. Then, if the aforementioned risk probability is greater than or equal to the risk threshold, the aforementioned pre-decision strategy model and the aforementioned pre-decision risk assessment model are combined to make a safety correction to the current pre-decision action and update the aforementioned current pre-decision strategy to obtain the target pre-decision strategy. Finally, a pre-decision is made based on the aforementioned target pre-decision strategy. This application first utilizes a dataset containing power system operating status, fault information, and pre-decision actions to train a risk assessment model based on a deep convolutional network to predict the risk of pre-decision strategies. Second, it constructs a deep network decision model with operating status and fault information as input to generate a preliminary pre-decision strategy. Then, based on the aforementioned risk assessment model, it predicts the risk probability of the current pre-decision strategy and corrects the current pre-decision action to ensure the strategy's safety. Finally, it makes a pre-decision based on the final pre-decision strategy. This application solves the problem of low safety in existing transient voltage stability pre-decision strategies. Attached Figure Description

[0016] Figure 1 A hardware block diagram of a mobile terminal for performing a pre-decision method for stabilizing power system voltage online, according to an embodiment of this application, is shown.

[0017] Figure 2 A flowchart illustrating a pre-decision method for online voltage stabilization of a power system according to an embodiment of this application is shown.

[0018] Figure 3 A structural diagram of a pre-decision risk assessment model provided according to an embodiment of this application is shown;

[0019] Figure 4A structural diagram of a pre-decision strategy model provided according to an embodiment of this application is shown;

[0020] Figure 5 A safety transient voltage stability pre-decision framework diagram is provided according to an embodiment of this application;

[0021] Figure 6 A structural block diagram of a pre-decision device for stabilizing online voltage in a power system, according to an embodiment of this application, is shown.

[0022] The above figures include the following reference numerals:

[0023] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application 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 for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover 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.

[0027] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0028] Pre-decision: In actual power systems, emergency control strategies for possible faults are generally formulated in advance (either offline or updated regularly online), and then matched and executed in real time within the system. This process is called pre-decision.

[0029] As described in the background section, the possible operating modes and dynamic characteristics of power systems in the prior art have become more diverse and complex. Moreover, the security constraints of power system pre-decision are nonlinear and dynamic. In complex power systems, it is relatively difficult to construct Lagrange multipliers or Lyapunov functions. Furthermore, approximating nonlinear constraints as linear constraints carries the risk of information loss. To address the problem of low security in the pre-decision strategies for transient voltage stability in the prior art, embodiments of this application provide a pre-decision method, pre-decision device, computer-readable storage medium, and computer program product for online voltage stability of power systems.

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0031] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a pre-decision method for online voltage stability in a power system, according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0032] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the device information display method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0033] This embodiment provides a pre-decision method for stabilizing the voltage of a power system online, which runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than that shown here.

[0034] Figure 2 This is a flowchart of a pre-decision method for ensuring online voltage stability in a power system according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0035] Step S201: Input the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of the first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model and is used to ensure that the current pre-decision strategy meets the set safety constraints.

[0036] Specifically, the purpose of pre-decision is to calculate a series of control strategies that minimize the overall cost of maintaining the power system stable in the event of a predetermined fault. These control strategies are... t The calculation is based on the current running state s t and a pre-defined fault set F = {F1, F2, ... F} n Its mathematical description is as follows: Among them, s t Let a be the operating state of the system at time t. t Let J(s) be the pre-decision strategy at time t. t a t F i f(s) represents the cost of the pre-decision strategy, such as the voltage drop level or the total amount of generator / load shedding. Assuming the system reaches a quasi-steady state at some point t+T after the fault occurs, the stability of the system is determined based on this state. t+T F i The expression denoted by represents the transient stability constraint after fault recovery. In the studied scenario, f is defined as the difference between the lowest system voltage after a certain time and 0.8 pu. If f > 0, it means the system voltage has failed to recover to above the safe voltage, i.e., the system has collapsed. Clearly, power system stability pre-decision is a nonlinear, non-convex constraint optimization problem. Therefore, in the actual process of formulating control strategies, the current operating state and current fault information of the power system are input into the pre-decision strategy model to obtain the current pre-decision strategy and current pre-decision action. Each current operating state corresponds to a current pre-decision action.

[0037] Step S202: The current operating status of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of the second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability.

[0038] Specifically, after obtaining the current pre-decision strategy, it is necessary to determine whether the current pre-decision strategy can minimize the overall cost of maintaining stability of the power system after a predetermined fault occurs, and whether the current pre-decision action is within the pre-decision action space. Therefore, the current operating state of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to conduct a risk assessment on the current pre-decision strategy and obtain the risk probability. The risk probability is the risk level of the current pre-decision strategy.

[0039] Step S203: If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain the target pre-decision strategy.

[0040] Specifically, if the aforementioned risk probability is greater than or equal to the risk threshold, it indicates that the current pre-decision strategy does not meet the policy requirements of the power system and cannot minimize the overall cost of maintaining stability of the power system after a predetermined fault occurs. Therefore, it is necessary to regenerate the pre-decision strategy. This is done by combining the pre-decision strategy model with the aforementioned pre-decision risk assessment model to make a safety correction to the current pre-decision action. Gradient descent is used to reduce the risk of the pre-decision, and the pre-decision action can be projected onto the safety action space, thereby significantly improving the reliability and effectiveness of the pre-decision strategy.

[0041] Step S204: Make a preliminary decision based on the above-mentioned target preliminary decision-making strategy.

[0042] Specifically, the aforementioned pre-decision-making strategy is used as an emergency control strategy for potential failures, and is matched and executed in real time within the power system.

[0043] In this embodiment, firstly, the current operating state and current fault information of the power system are input into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained using multiple sets of first training data through a deep convolutional network. Each set of first training data includes a sample operating state, sample fault information, and a corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model, used to ensure that the current pre-decision strategy meets the set safety constraints. Then, the current operating state, current fault information, and current pre-decision action of the power system are input into the pre-decision risk assessment model. The aforementioned pre-decision risk assessment model is used to assess the risk of the current pre-decision strategy and obtain the risk probability. This model is trained using multiple sets of second training data through a deep convolutional network. Each set of second training data includes the sample's operating state, fault information, pre-decision action, and corresponding risk probability. Then, if the risk probability is greater than or equal to a risk threshold, the pre-decision strategy model and the pre-decision risk assessment model are combined to safely correct the current pre-decision action and update the current pre-decision strategy to obtain a target pre-decision strategy. Finally, a pre-decision is made based on the target pre-decision strategy. This application first uses a dataset containing power system operating states, fault information, and pre-decision actions to train a risk assessment model predicting the risk of the pre-decision strategy using a deep convolutional network. Second, a deep network decision model with operating state and fault information as input is constructed to generate a preliminary pre-decision strategy. Then, based on the aforementioned risk assessment model, the risk probability of the current pre-decision strategy is predicted, and the current pre-decision action is corrected to ensure the safety of the strategy. Finally, a pre-decision is made based on the final pre-decision strategy. This application addresses the problem of low security in existing transient voltage stability pre-decision strategies.

[0044] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the pre-decision method for online voltage stability of the power system of this application will be described in detail below with reference to specific embodiments.

[0045] To ensure the security of the pre-decision strategy, in an optional implementation, before step S201 above, the method further includes:

[0046] Step S30 1, Build an initial pre-decision strategy model based on the above deep convolutional network. The above initial pre-decision strategy model is the pre-decision strategy model to be trained.

[0047] Step S302, Input Step, input the first target training data into the above-mentioned initial pre-decision strategy model, use the deep deterministic policy gradient algorithm for preliminary training, and generate the corresponding training pre-decision action. The above-mentioned first target training data is any set of the above-mentioned first training data.

[0048] Step S303, First calculation step: Calculate the action gradient corresponding to the training pre-decision action using the above-mentioned pre-decision risk assessment model.

[0049] Step S304, First correction step: Perform a safety correction on the above training pre-decision action according to the above action gradient to obtain the above sample pre-decision strategy;

[0050] Step S305: Repeat the above input step, the above calculation step, and the above correction step at least once in sequence until all the above first training data are input into the above initial pre-decision strategy model to complete the training and obtain the above pre-decision strategy model.

[0051] In the above embodiments, the pre-decision strategy model includes a pre-decision initial strategy formulation network layer and a security action correction layer, as shown in the specific structure diagram below. Figure 3 As shown. The preliminary decision-making network layer: taking the operating state [P, Q, V, S] and fault information F as input, outputs preliminary pre-decision actions. The aforementioned preliminary decision-making network can be trained using a general reinforcement learning algorithm. Without loss of generality, in this embodiment, the preliminary decision-making network is implemented using Deep Deterministic Policy Gradient (DDPG). Appropriate reinforcement learning reward design can help the agent avoid some dangerous policies during training and application. The first objective training data is input into the preliminary decision-making network in the aforementioned preliminary decision-making policy model, and preliminary training is performed using the Deep Deterministic Policy Gradient algorithm to generate initial training preliminary decision actions. The superscript 0 indicates that corrections have not yet begun. Assuming the aforementioned pre-decision risk assessment model has learned relatively accurate knowledge, the actions will... As an initial solution to the following constrained optimization problem: Where r is the reward function for the preliminary decision-making of the network layer. Where V iLet be the per-unit value of the voltage amplitude at node i 10 seconds after fault clearance, and B be the set of system node numbers. If the transient stability constraint is met, the reward function comprehensively reflects the action cost and the degree of voltage recovery; if the transient stability constraint is not met, a large negative reward (-100) is given as a penalty. The above optimization problem has a nonlinear objective function and constraints, making it difficult to solve the analytical function expression. However, gradient descent can be used to reduce the risk of pre-decision and obtain an approximate solution that meets the conditions. Specifically, the action gradient corresponding to the above training pre-decision action is calculated using the above pre-decision risk assessment model; based on the action gradient, the action is continuously corrected by safety projection in the safety action correction layer until the final safe action is obtained. The final pre-decision strategy is then determined. Safe pre-decision actions are obtained through a pre-trained risk assessment network and a safety correction layer, ensuring the safety of the pre-decision strategy. After training, the trained pre-decision strategy model needs to be validated to ensure it meets the requirements of the power system. If the model does not meet these requirements, the parameters need to be adjusted and the model retrained.

[0052] In order to modify and optimize the pre-decision actions in generating the pre-decision strategy and improve the accuracy of the pre-decision strategy, in an optional implementation, step S303 above includes:

[0053] Step S3031: Calculate the motion gradient according to the first formula, whereby the first formula is: f θ This indicates the aforementioned pre-decision risk assessment model, s t F represents the operating status of the above sample. i This indicates the fault information of the above samples. Let a(n) represent the training pre-decision action after the kth correction, and let a(n) represent the projection of the training pre-decision action onto the n directions of the action safety space, where the action safety space is the space that makes the sample pre-decision strategy satisfy the set safety constraints.

[0054] In the above embodiments, considering the results of risk assessment and mathematical form, at the operating point of the power system The gradient of the pre-decision risk assessment network to the input action is calculated according to the first formula. The calculated gradient represents the direction of the fastest risk reduction, which is crucial for subsequent action adjustments.

[0055] To evaluate the pre-decision strategy and improve its security, in an optional implementation, before step S201 above, the method further includes:

[0056] Step S40 1: Build an initial risk assessment model based on the above deep convolutional network. The above initial risk assessment model is a pre-decision risk assessment model to be trained.

[0057] Step S402: Input multiple sets of the second training data into the initial risk assessment model in sequence, and train it using a deep convolutional network algorithm to obtain the pre-decision risk assessment model.

[0058] In the above embodiment, the pre-decision risk assessment is regarded as a binary classification problem, with stable samples labeled as 1 and unstable samples labeled as 0. Assume y(1) and y(2) represent the predicted probabilities of instability and stability of transient voltage stability, respectively. Clearly, y(1)-y(2) has already determined the assessment result; y(1)-y(2)>0 indicates instability, and y(1)-y(2)<0 indicates stability. Considering that the sum of the probabilities of stability and instability in assessing the system's operating state is 1, the pre-decision strategy risk assessment can be simplified to estimating the instability probability after the pre-decision and comparing it with 0.5. The improved pre-decision risk assessment network can be written as: Where f θ Representing a parameterized neural network, This represents the probability of instability after the pre-decision strategy is applied. The solution is the boundary of the feasible region, i.e., an approximation. The above method also achieves the feasible region. The nonlinear approximation, i.e., the structure of the proposed pre-decision risk assessment model is as follows: Figure 4 As shown, P, Q, V, and S represent the active power, reactive power, equivalent slip of node voltage, and equivalent slip of motor at load nodes of the power system, respectively. F represents fault information, and Risk(s, a) represents the risk probability.

[0059] To achieve a balance between the benefits and risks of exploring pre-decision strategies, in an optional implementation, prior to step S203 above, the method further includes:

[0060] Step S501: Determine the safety space of the current pre-decision action according to the second formula, wherein the second formula is: The safety space for the above actions is used to represent the space that satisfies the stability constraint f. θ (s t a t s t+T All feasible pre-decision actions ≤ ε, f θ (s t a t s t+T )≤ε represents the stability constraint of the transient voltage, ε represents the aforementioned risk threshold, and a tLet s be the pre-decision strategy at time t. t Let s be the operating state of the aforementioned power system at time t. t+T The above operating states of the power system at time t+T, where A represents the set of emergency control strategies corresponding to all faults;

[0061] Step S502: Determine the boundary of the safety space of the current pre-decision action according to the third formula, wherein the third formula is: It represents the nonlinear boundary of the feasible pre-decision action space.

[0062] In the above embodiments, the boundary of the feasible pre-decision action space of the power system under a certain operating state is determined. However, considering the volatility and uncertainty of new energy sources, in power systems with high uncertainty, the boundary of the feasible region of the voltage stability pre-decision strategy is not fixed but is affected by some random factors. Meanwhile, traditional deterministic assessment methods do not consider the uncertainty of new energy sources, making it difficult to quantify the risk probability of uncertain systems. Based on the above stability assessment model, adjusting the threshold for determining stability and instability can take into account the uncertainty of the power system. Specifically, the feasible region and its boundary of the voltage stability pre-decision strategy can be written as: By adjusting the threshold ε, a trade-off between the benefits and risks of exploring pre-decision strategies can be achieved.

[0063] To enhance the security of the pre-decision strategy, in an optional implementation, step S203 includes:

[0064] Step S203 1, Second calculation step: Calculate the current action gradient corresponding to the current pre-decision action according to the first formula using the above-mentioned pre-decision risk assessment model, and output the current risk probability;

[0065] Step S2032, the second correction step, involves transmitting the current action gradient to the pre-decision strategy model and correcting the current pre-decision action using gradient descent according to the fourth formula, whereby the fourth formula is... This indicates the gradient of the current action. This represents the current pre-decision action after the k-th correction at time t. This represents the current pre-decision action after the (k+1)th correction at time t, where λ is the correction step size, indicating the degree of each correction.

[0066] Step S2033: Repeat the above second calculation step and the above second correction step at least once until the above current risk probability is less than the above risk threshold, and complete the safety correction of the above current pre-decision action.

[0067] In the above embodiments, considering the difficulty of end-to-end agents in handling complex nonlinear constraints, as well as other reasons such as distribution offset and sample-based learning, actions that violate safety constraints may still occur at this stage. Therefore, the preliminary decision-making network does not directly output a policy that satisfies the constraints, but only provides a suboptimal initial action for the next stage to correct, thereby further improving safety. The above optimization problem has a nonlinear objective function and constraints, making it difficult to solve the analytical function expression, but gradient descent can be used to reduce the risk of the preliminary decision and obtain an approximate solution that satisfies the conditions. Specifically, considering the results of risk assessment and mathematical form, at the power system operating point... The gradient of the pre-decision risk assessment model on the pre-decision action in the input data is calculated. The calculated gradient represents the direction of the fastest risk decrease, which is crucial for correction. Then, based on the action gradient, the action is continuously corrected by safety projection. The above steps are repeated until the risk probability is less than the risk threshold ε. The logical process is as follows: By utilizing a pre-decision risk assessment model and its gradient, the pre-decision actions can be projected onto the safe action space, thereby enhancing the safety of the pre-decision strategy, based on the original use of only soft constraints or approximate constraints.

[0068] To maintain the consistency and effectiveness of the target pre-decision strategy, in an optional implementation, after step S203 above, the method further includes:

[0069] Step S601: If the above-mentioned risk probability is less than the above-mentioned risk threshold, the above-mentioned current pre-decision strategy is determined as the above-mentioned target pre-decision strategy.

[0070] In the above embodiments, when a risk assessment determines that the probability of risk faced by the current pre-decision strategy is very low and does not exceed a pre-set risk threshold, the current pre-decision strategy can be directly determined as the target pre-decision strategy. This maintains the consistency and effectiveness of the target pre-decision strategy while keeping the risk relatively low.

[0071] This embodiment relates to a pre-decision framework for safe transient voltage stability, such as... Figure 5 As shown, it includes the following steps:

[0072] Step S1: Train a pre-decision risk assessment model using the power system operating status and fault dataset generated by simulation, in order to assess the risk level of the pre-decision strategy;

[0073] Step S2: Based on the simulation environment, train the preliminary decision-making network layer of the safety pre-decision model, and combine it with the gradient of the pre-decision risk assessment model to make safety corrections to the obtained actions, and construct an online pre-decision model for power system voltage stability that integrates risks.

[0074] Step S3: Apply the online pre-decision model for power system voltage stability to the online scenario.

[0075] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0076] This application also provides a pre-decision device for ensuring online voltage stability in a power system. It should be noted that this pre-decision device can be used to execute the pre-decision method for ensuring online voltage stability in a power system provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0077] The following describes the pre-decision device for stabilizing online voltage in a power system provided in the embodiments of this application.

[0078] Figure 6 This is a structural block diagram of a pre-decision device for stabilizing online voltage in a power system according to an embodiment of this application. Figure 6 As shown, the device includes:

[0079] The strategy generation unit 10 is used to input the current operating state and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of the first training data includes sample operating state, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model and is used to ensure that the current pre-decision strategy meets the set safety constraints.

[0080] Specifically, the purpose of pre-decision is to calculate a series of control strategies that minimize the overall cost of maintaining the power system stable in the event of a predetermined fault. These control strategies are... t The calculation is based on the current running state s t and a pre-defined fault set F = {F1, F2, ... F} n Its mathematical description is as follows: Among them, s t Let a be the operating state of the system at time t. t Let J(s) be the pre-decision strategy at time t.t a t F i f(s) represents the cost of the pre-decision strategy, such as the voltage drop level or the total amount of generator / load shedding. Assuming the system reaches a quasi-steady state at some point t+T after the fault occurs, the stability of the system is determined based on this state. t+T F i The expression denoted by represents the transient stability constraint after fault recovery. In the studied scenario, f is defined as the difference between the lowest system voltage after a certain time and 0.8 pu. If f > 0, it means the system voltage has failed to recover to above the safe voltage, i.e., the system has collapsed. Clearly, power system stability pre-decision is a nonlinear, non-convex constraint optimization problem. Therefore, in the actual process of formulating control strategies, the current operating state and current fault information of the power system are input into the pre-decision strategy model to obtain the current pre-decision strategy and current pre-decision action. Each current operating state corresponds to a current pre-decision action.

[0081] The risk assessment unit 20 is used to input the current operating status of the power system, the current fault information, and the current pre-decision action into the pre-decision risk assessment model, so as to use the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of the second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability.

[0082] Specifically, after obtaining the current pre-decision strategy, it is necessary to determine whether the current pre-decision strategy can minimize the overall cost of maintaining stability of the power system after a predetermined fault occurs, and whether the current pre-decision action is within the pre-decision action space. Therefore, the current operating state of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to conduct a risk assessment on the current pre-decision strategy and obtain the risk probability. The risk probability is the risk level of the current pre-decision strategy.

[0083] The correction unit 30 is used to combine the above-mentioned pre-decision strategy model and the above-mentioned pre-decision risk assessment model to make a safety correction to the current pre-decision action and update the above-mentioned current pre-decision strategy when the above-mentioned risk probability is greater than or equal to the risk threshold, so as to obtain the target pre-decision strategy.

[0084] Specifically, if the aforementioned risk probability is greater than or equal to the risk threshold, it indicates that the current pre-decision strategy does not meet the policy requirements of the power system and cannot minimize the overall cost of maintaining stability of the power system after a predetermined fault occurs. Therefore, it is necessary to regenerate the pre-decision strategy. This is done by combining the pre-decision strategy model with the aforementioned pre-decision risk assessment model to make a safety correction to the current pre-decision action. Gradient descent is used to reduce the risk of the pre-decision, and the pre-decision action can be projected onto the safety action space, thereby significantly improving the reliability and effectiveness of the pre-decision strategy.

[0085] The pre-decision unit 40 is used to make pre-decision decisions based on the aforementioned target pre-decision strategy.

[0086] Specifically, the aforementioned pre-decision-making strategy is used as an emergency control strategy for potential failures, and is matched and executed in real time within the power system.

[0087] In this embodiment, the strategy generation unit inputs the current operating state and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained using multiple sets of first training data through a deep convolutional network. Each set of the first training data includes a sample operating state, sample fault information, and a corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model, used to ensure that the current pre-decision strategy meets the set safety constraints. The risk assessment unit inputs the current operating state, current fault information, and current pre-decision action of the power system into the pre-decision risk assessment model. The aforementioned pre-decision risk assessment model is used to assess the risk of the current pre-decision strategy and obtain the risk probability. The aforementioned pre-decision risk assessment model is a model trained using multiple sets of second training data through the aforementioned deep convolutional network. Each set of the aforementioned second training data includes the aforementioned sample running status, the aforementioned sample fault information, the sample pre-decision action, and the corresponding sample risk probability. The correction unit is used to combine the aforementioned pre-decision strategy model and the aforementioned pre-decision risk assessment model to make a safety correction to the current pre-decision action and update the aforementioned current pre-decision strategy when the aforementioned risk probability is greater than or equal to the risk threshold, thereby obtaining the target pre-decision strategy. The pre-decision unit is used to make a pre-decision based on the aforementioned target pre-decision strategy. This application first utilizes a dataset containing power system operating status, fault information, and pre-decision actions to train a risk assessment model based on a deep convolutional network to predict the risk of pre-decision strategies. Second, it constructs a deep network decision model with operating status and fault information as input to generate a preliminary pre-decision strategy. Then, based on the aforementioned risk assessment model, it predicts the risk probability of the current pre-decision strategy and corrects the current pre-decision action to ensure the strategy's safety. Finally, it makes a pre-decision based on the final pre-decision strategy. This application solves the problem of low safety in existing transient voltage stability pre-decision strategies.

[0088] To ensure the security of the pre-decision strategy, in one optional embodiment, the device further includes:

[0089] The first building unit is used to build an initial pre-decision strategy model based on the aforementioned deep convolutional network before inputting the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The aforementioned initial pre-decision strategy model is the pre-decision strategy model to be trained.

[0090] The input unit is used to perform the input step, input the first target training data into the above-mentioned initial pre-decision policy model, perform preliminary training using the deep deterministic policy gradient algorithm, and generate the corresponding training pre-decision action. The above-mentioned first target training data is any set of the above-mentioned first training data.

[0091] The first calculation unit is used to perform the first calculation step, and to calculate the action gradient corresponding to the training pre-decision action using the above-mentioned pre-decision risk assessment model.

[0092] The first correction unit is used to perform the first correction step, which safely corrects the training pre-decision action according to the action gradient to obtain the sample pre-decision strategy.

[0093] The repeating unit is used to repeat the above input step, the above calculation step and the above correction step at least once in sequence until all the above first training data are input into the above initial pre-decision strategy model to complete the training and obtain the above pre-decision strategy model.

[0094] In the above embodiments, the pre-decision strategy model includes a pre-decision initial strategy formulation network layer and a security action correction layer, as shown in the specific structure diagram below. Figure 3 As shown. The preliminary decision-making network layer: taking the operating state [P, Q, V, S] and fault information F as input, outputs preliminary pre-decision actions. The aforementioned preliminary decision-making network can be trained using a general reinforcement learning algorithm. Without loss of generality, in this embodiment, the preliminary decision-making network is implemented using Deep Deterministic Policy Gradient (DDPG). Appropriate reinforcement learning reward design can help the agent avoid some dangerous policies during training and application. The first objective training data is input into the preliminary decision-making network in the aforementioned preliminary decision-making policy model, and preliminary training is performed using the Deep Deterministic Policy Gradient algorithm to generate initial training preliminary decision actions. The superscript 0 indicates that corrections have not yet begun. Assuming the aforementioned pre-decision risk assessment model has learned relatively accurate knowledge, the actions will... As an initial solution to the following constrained optimization problem: Where r is the reward function for the preliminary decision-making of the network layer. Where V iLet be the per-unit value of the voltage amplitude at node i 10 seconds after fault clearance, and B be the set of system node numbers. If the transient stability constraint is met, the reward function comprehensively reflects the action cost and the degree of voltage recovery; if the transient stability constraint is not met, a large negative reward (-100) is given as a penalty. The above optimization problem has a nonlinear objective function and constraints, making it difficult to solve the analytical function expression. However, gradient descent can be used to reduce the risk of pre-decision and obtain an approximate solution that meets the conditions. Specifically, the action gradient corresponding to the above training pre-decision action is calculated using the above pre-decision risk assessment model; based on the action gradient, the action is continuously corrected by safety projection in the safety action correction layer until the final safe action is obtained. The final pre-decision strategy is then determined. Safe pre-decision actions are obtained through a pre-trained risk assessment network and a safety correction layer, ensuring the safety of the pre-decision strategy. After training, the trained pre-decision strategy model needs to be validated to ensure it meets the requirements of the power system. If the model does not meet these requirements, the parameters need to be adjusted and the model retrained.

[0095] In order to modify and optimize the pre-decision actions in generating the pre-decision strategy and improve the accuracy of the pre-decision strategy, in an optional embodiment, the first computing unit includes:

[0096] The first calculation module calculates the gradient of the aforementioned action according to a first formula, wherein the first formula is: f θ This indicates the aforementioned pre-decision risk assessment model, s t F represents the operating status of the above sample. i This indicates the fault information of the above samples. Let a(n) represent the training pre-decision action after the kth correction, and let a(n) represent the projection of the training pre-decision action onto the n directions of the action safety space, where the action safety space is the space that makes the sample pre-decision strategy satisfy the set safety constraints.

[0097] In the above embodiments, considering the results of risk assessment and mathematical form, at the operating point of the power system The gradient of the pre-decision risk assessment network to the input action is calculated according to the first formula. The calculated gradient represents the direction of the fastest risk reduction, which is crucial for subsequent action adjustments.

[0098] To evaluate the pre-decision strategy and improve its security, in one optional embodiment, the device further includes:

[0099] The second building unit is used to build an initial risk assessment model based on the above-mentioned deep convolutional network before inputting the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The above-mentioned initial risk assessment model is the pre-decision risk assessment model to be trained.

[0100] The training unit is used to sequentially input multiple sets of the second training data into the initial risk assessment model and train it using a deep convolutional network algorithm to obtain the pre-decision risk assessment model.

[0101] In the above embodiment, the pre-decision risk assessment is regarded as a binary classification problem, with stable samples labeled as 1 and unstable samples labeled as 0. Assume y(1) and y(2) represent the predicted probabilities of instability and stability of transient voltage stability, respectively. Clearly, y(1)-y(2) has already determined the assessment result; y(1)-y(2)>0 indicates instability, and y(1)-y(2)<0 indicates stability. Considering that the sum of the probabilities of stability and instability in assessing the system's operating state is 1, the pre-decision strategy risk assessment can be simplified to estimating the instability probability after the pre-decision and comparing it with 0.5. The improved pre-decision risk assessment network can be written as: Where f θ Representing a parameterized neural network, This represents the probability of instability after the pre-decision strategy is applied. The solution is the boundary of the feasible region, i.e., an approximation. The above method also achieves the feasible region. The nonlinear approximation, i.e., the structure of the proposed pre-decision risk assessment model is as follows: Figure 4 As shown, P, Q, V, and S represent the active power, reactive power, equivalent slip of node voltage, and equivalent slip of motor at load nodes of the power system, respectively. F represents fault information, and Risk(s, a) represents the risk probability.

[0102] To balance the benefits and risks of exploring pre-decision strategies, in one optional embodiment, the device further includes:

[0103] The first determining unit, before inputting the current operating state and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and current pre-decision action, determines the safety space of the aforementioned current pre-decision action according to a second formula, wherein the second formula is: The safety space for the above actions is used to represent the space that satisfies the stability constraint f. θ (s t a t s t+T All feasible pre-decision actions ≤ ε, f θ(s t a t s t+T )≤ε represents the stability constraint of the transient voltage, ε represents the aforementioned risk threshold, and a t Let s be the pre-decision strategy at time t. t Let s be the operating state of the aforementioned power system at time t. t+T The above operating states of the power system at time t+T, where A represents the set of emergency control strategies corresponding to all faults;

[0104] The second determining unit is used to determine the boundary of the safety space of the current pre-decision action according to the third formula, wherein the third formula is: It represents the nonlinear boundary of the feasible pre-decision action space.

[0105] In the above embodiments, the boundary of the feasible pre-decision action space of the power system under a certain operating state is determined. However, considering the volatility and uncertainty of new energy sources, in power systems with high uncertainty, the boundary of the feasible region of the voltage stability pre-decision strategy is not fixed but is affected by some random factors. Meanwhile, traditional deterministic assessment methods do not consider the uncertainty of new energy sources, making it difficult to quantify the risk probability of uncertain systems. Based on the above stability assessment model, adjusting the threshold for determining stability and instability can take into account the uncertainty of the power system. Specifically, the feasible region and its boundary of the voltage stability pre-decision strategy can be written as: By adjusting the threshold ε, a trade-off between the benefits and risks of exploring pre-decision strategies can be achieved.

[0106] To enhance the security of the pre-decision strategy, in one optional implementation, the above-mentioned correction unit includes:

[0107] The second calculation module is used to execute the second calculation step, using the above-mentioned pre-decision risk assessment model to calculate the current action gradient corresponding to the above-mentioned current pre-decision action according to the above-mentioned first formula, and output the current risk probability;

[0108] The correction module is used to execute the second correction step, which transmits the gradient of the current action to the pre-decision strategy model and corrects the current pre-decision action using gradient descent according to the fourth formula, whereby the fourth formula is: This indicates the gradient of the current action. This represents the current pre-decision action after the k-th correction at time t. This represents the current pre-decision action after the (k+1)th correction at time t, where λ is the correction step size, representing the degree of each correction or the correction step size, i.e., the degree of each modification.

[0109] The repeating module repeats the second calculation step and the second correction step at least once until the current risk probability is less than the risk threshold, thus completing the safety correction of the current pre-decision action.

[0110] In the above embodiments, considering the difficulty of end-to-end agents in handling complex nonlinear constraints, as well as other reasons such as distribution offset and sample-based learning, actions that violate safety constraints may still occur at this stage. Therefore, the preliminary decision-making network does not directly output a policy that satisfies the constraints, but only provides a suboptimal initial action for the next stage to correct, thereby further improving safety. The above optimization problem has a nonlinear objective function and constraints, making it difficult to solve the analytical function expression, but gradient descent can be used to reduce the risk of the preliminary decision and obtain an approximate solution that satisfies the conditions. Specifically, considering the results of risk assessment and mathematical form, at the power system operating point... The gradient of the pre-decision risk assessment model on the pre-decision action in the input data is calculated. The calculated gradient represents the direction of the fastest risk decrease, which is crucial for correction. Then, based on the action gradient, the action is continuously corrected by safety projection. The above steps are repeated until the risk probability is less than the risk threshold ε. The logical process is as follows: By utilizing a pre-decision risk assessment model and its gradient, the pre-decision actions can be projected onto the safe action space, thereby enhancing the safety of the pre-decision strategy, based on the original use of only soft constraints or approximate constraints.

[0111] To maintain the consistency and effectiveness of the target pre-decision strategy, in one optional embodiment, the device further includes:

[0112] The third determining unit is used to input the current operating state of the power system, the current fault information, and the current pre-decision action into the pre-decision risk assessment model, and then use the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. If the risk probability is less than the risk threshold, the current pre-decision strategy is determined as the target pre-decision strategy.

[0113] In the above embodiments, when a risk assessment determines that the probability of risk faced by the current pre-decision strategy is very low and does not exceed a pre-set risk threshold, the current pre-decision strategy can be directly determined as the target pre-decision strategy. This maintains the consistency and effectiveness of the target pre-decision strategy while keeping the risk relatively low.

[0114] The aforementioned pre-decision-making device for stabilizing online voltage in a power system includes a processor and a memory. The strategy generation unit and risk assessment unit, among others, are stored as program units in the memory. The processor executes these program units to achieve their respective functions. All of the above modules reside in the same processor; alternatively, the modules may be located in different processors in any combination.

[0115] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the low security of transient voltage stability pre-decision strategies in existing technologies.

[0116] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0117] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the pre-decision method for stabilizing the online voltage of the power system.

[0118] Specifically, the pre-decision-making methods for ensuring online voltage stability in power systems include:

[0119] Step S201: Input the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of the first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model and is used to ensure that the current pre-decision strategy meets the set safety constraints.

[0120] Step S202: The current operating status of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of the second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability.

[0121] Step S203: If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain the target pre-decision strategy.

[0122] Step S204: Make a preliminary decision based on the above-mentioned target preliminary decision-making strategy.

[0123] This invention provides a processor for running a program, wherein the program executes the pre-decision method for ensuring online voltage stability of a power system.

[0124] Specifically, the pre-decision-making methods for ensuring online voltage stability in power systems include:

[0125] Step S201: Input the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of the first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model and is used to ensure that the current pre-decision strategy meets the set safety constraints.

[0126] Step S202: The current operating status of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of the second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability.

[0127] Step S203: If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain the target pre-decision strategy.

[0128] Step S204: Make a preliminary decision based on the above-mentioned target preliminary decision-making strategy.

[0129] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0130] Step S201: Input the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of the first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model and is used to ensure that the current pre-decision strategy meets the set safety constraints.

[0131] Step S202: The current operating status of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of the second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability.

[0132] Step S203: If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain the target pre-decision strategy.

[0133] Step S204: Make a preliminary decision based on the above-mentioned target preliminary decision-making strategy.

[0134] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0135] Step S201: Input the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of the first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model and is used to ensure that the current pre-decision strategy meets the set safety constraints.

[0136] Step S202: The current operating status of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of the second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability.

[0137] Step S203: If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain the target pre-decision strategy.

[0138] Step S204: Make a preliminary decision based on the above-mentioned target preliminary decision-making strategy.

[0139] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0140] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0143] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0144] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0145] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0146] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0147] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0148] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0149] 1) The pre-decision method for ensuring online voltage stability in a power system according to this application firstly inputs the current operating state and current fault information of the power system into a pre-decision strategy model to obtain a current pre-decision strategy and a current pre-decision action. The pre-decision strategy model is a model trained using multiple sets of first training data through a deep convolutional network. Each set of first training data includes a sample operating state, sample fault information, and a corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model, used to ensure that the current pre-decision strategy meets the set safety constraints. Then, the current operating state, current fault information, and current pre-decision action of the power system are input into the pre-decision strategy model. In the policy risk assessment model, the risk of the current pre-decision strategy is assessed using the aforementioned pre-decision risk assessment model to obtain the risk probability. The aforementioned pre-decision risk assessment model is a model trained using multiple sets of second training data through the aforementioned deep convolutional network. Each set of the aforementioned second training data includes the aforementioned sample operating status, the aforementioned sample fault information, the sample pre-decision action, and the corresponding sample risk probability. Then, if the aforementioned risk probability is greater than or equal to the risk threshold, the aforementioned pre-decision strategy model and the aforementioned pre-decision risk assessment model are combined to make a safety correction to the current pre-decision action and update the aforementioned current pre-decision strategy to obtain the target pre-decision strategy. Finally, a pre-decision is made based on the aforementioned target pre-decision strategy. This application first utilizes a dataset containing power system operating status, fault information, and pre-decision actions to train a risk assessment model based on a deep convolutional network to predict the risk of pre-decision strategies. Second, it constructs a deep network decision model with operating status and fault information as input to generate a preliminary pre-decision strategy. Then, based on the aforementioned risk assessment model, it predicts the risk probability of the current pre-decision strategy and corrects the current pre-decision action to ensure the strategy's safety. Finally, it makes a pre-decision based on the final pre-decision strategy. This application solves the problem of low safety in existing transient voltage stability pre-decision strategies.

[0150] 2) The pre-decision-making device for stable online voltage in a power system according to this application includes a strategy generation unit that inputs the current operating state and current fault information of the power system into a pre-decision-making strategy model to obtain a current pre-decision-making strategy and a current pre-decision-making action. The pre-decision-making strategy model is a model trained using multiple sets of first training data through a deep convolutional network. Each set of first training data includes a sample operating state, sample fault information, and a corresponding sample pre-decision-making strategy. The current pre-decision-making action is a variable generated during the execution of the pre-decision-making strategy model, used to ensure that the current pre-decision-making strategy meets the set safety constraints. A risk assessment unit is used to input the current operating state, current fault information, and current pre-decision-making action of the power system into the pre-decision-making strategy model. In the policy risk assessment model, the risk of the current pre-decision strategy is assessed using the aforementioned pre-decision risk assessment model to obtain the risk probability. The aforementioned pre-decision risk assessment model is a model trained using multiple sets of second training data through the aforementioned deep convolutional network. Each set of the aforementioned second training data includes the aforementioned sample operating status, the aforementioned sample fault information, the sample pre-decision action, and the corresponding sample risk probability. The correction unit is used to combine the aforementioned pre-decision strategy model and the aforementioned pre-decision risk assessment model to perform a safety correction on the current pre-decision action and update the aforementioned current pre-decision strategy when the aforementioned risk probability is greater than or equal to the risk threshold, thereby obtaining the target pre-decision strategy. The pre-decision unit is used to make a pre-decision based on the aforementioned target pre-decision strategy. This application first utilizes a dataset containing power system operating status, fault information, and pre-decision actions to train a risk assessment model based on a deep convolutional network to predict the risk of pre-decision strategies. Second, it constructs a deep network decision model with operating status and fault information as input to generate a preliminary pre-decision strategy. Then, based on the aforementioned risk assessment model, it predicts the risk probability of the current pre-decision strategy and corrects the current pre-decision action to ensure the strategy's safety. Finally, it makes a pre-decision based on the final pre-decision strategy. This application solves the problem of low safety in existing transient voltage stability pre-decision strategies.

[0151] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A pre-decision-making method for ensuring online voltage stability in a power system, characterized in that, include: The current operating status and current fault information of the power system are input into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model, used to ensure that the current pre-decision strategy meets the set safety constraints. The current operating state of the power system, the current fault information, and the current pre-decision action are input into the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of second training data includes the sample operating state, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability. If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to make a safety correction to the current pre-decision action and update the current pre-decision strategy to obtain the target pre-decision strategy. Make a pre-decision based on the aforementioned target pre-decision strategy. If the risk probability is greater than or equal to the risk threshold, the pre-decision strategy model and the pre-decision risk assessment model will be combined to perform a safety correction on the current pre-decision action, including: The second calculation step involves using the pre-decision risk assessment model to calculate the current action gradient corresponding to the current pre-decision action according to the first formula, and outputting the current risk probability. The first formula is: , This refers to the pre-decision risk assessment model. This indicates the running status of the sample. This indicates the sample fault information. This represents the training pre-decision action after the k-th correction at time t. This means projecting the training pre-decision action onto n directions of the action safety space, where the action safety space is the space that ensures the sample pre-decision strategy satisfies the set safety constraints. The second correction step involves transmitting the current action gradient to the pre-decision strategy model, and then using gradient descent to correct the current pre-decision action according to the fourth formula, whereby the fourth formula is: , Indicates the current action gradient. This represents the current pre-decision action after the k-th correction at time t. This represents the current pre-decision action after the (k+1)th correction at time t. The step size indicates the degree of each correction. Repeat the second calculation step and the second correction step at least once in sequence until the current risk probability is less than the risk threshold, thus completing the safety correction of the current pre-decision action.

2. The method according to claim 1, characterized in that, Before inputting the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and current pre-decision action, the method further includes: An initial pre-decision policy model is built based on the deep convolutional network, and the initial pre-decision policy model is the pre-decision policy model to be trained. The input step involves inputting the first target training data into the initial pre-decision strategy model, performing preliminary training using the deep deterministic policy gradient algorithm, and generating corresponding training pre-decision actions. The first target training data can be any set of the first training data. The first calculation step involves using the pre-decision risk assessment model to calculate the action gradient corresponding to the training pre-decision action. The first correction step involves performing a safety correction on the training pre-decision action based on the action gradient to obtain the sample pre-decision strategy. The input step, the calculation step, and the correction step are repeated at least once in sequence until all the first training data are input into the initial pre-decision strategy model to complete the training, thereby obtaining the pre-decision strategy model.

3. The method according to claim 2, characterized in that, The calculation of the action gradient corresponding to the training pre-decision action using the pre-decision risk assessment model includes: The action gradient is calculated according to the first formula.

4. The method according to claim 1, characterized in that, Before inputting the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and current pre-decision action, the method further includes: An initial risk assessment model is built based on the deep convolutional network, which is a pre-decision risk assessment model to be trained. Multiple sets of the second training data are sequentially input into the initial risk assessment model, and the model is trained using a deep convolutional network algorithm to obtain the pre-decision risk assessment model.

5. The method according to claim 3, characterized in that, Before combining the pre-decision strategy model and the pre-decision risk assessment model to make a safety correction to the current pre-decision action when the risk probability is greater than or equal to the risk threshold, the method further includes: The action safety space for the current pre-decision action is determined according to the second formula, whereby the second formula is: , The action safety space is used to represent the space that satisfies stability constraints. All feasible pre-decision actions, This represents the stability constraint of transient voltage. Indicates the risk threshold, a t The pre-decision strategy at time t, The power system is in its operating state at time t. The power system in The operating state at any given time, where A represents the set of emergency control strategies corresponding to all faults; The boundary of the safety space of the current pre-decision action is determined according to the third formula, wherein the third formula is: , It represents the nonlinear boundary of the feasible pre-decision action space.

6. The method according to claim 1, characterized in that, After inputting the current operating state of the power system, the current fault information, and the current pre-decision action into the pre-decision risk assessment model, and using the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability, the method further includes: If the risk probability is less than the risk threshold, the current pre-decision strategy is determined as the target pre-decision strategy.

7. A pre-decision-making device for stabilizing online voltage in a power system, characterized in that, The device includes: The strategy generation unit inputs the current operating status and current fault information of the power system into the pre-decision strategy model to obtain the current pre-decision strategy and the current pre-decision action. The pre-decision strategy model is a model trained by a deep convolutional network using multiple sets of first training data. Each set of first training data includes sample operating status, sample fault information and corresponding sample pre-decision strategy. The current pre-decision action is a variable generated during the execution of the pre-decision strategy model, used to ensure that the current pre-decision strategy meets the set safety constraints. The risk assessment unit is used to input the current operating status of the power system, the current fault information, and the current pre-decision action into the pre-decision risk assessment model, so as to use the pre-decision risk assessment model to assess the risk of the current pre-decision strategy and obtain the risk probability. The pre-decision risk assessment model is a model trained by the deep convolutional network using multiple sets of second training data. Each set of second training data includes the sample operating status, the sample fault information, the sample pre-decision action, and the corresponding sample risk probability. The correction unit is used to combine the pre-decision strategy model and the pre-decision risk assessment model to make a safety correction to the current pre-decision action when the risk probability is greater than or equal to the risk threshold, and update the current pre-decision strategy to obtain the target pre-decision strategy. The pre-decision unit is used to make pre-decision decisions based on the target pre-decision strategy. The correction unit includes: The second calculation module is used to execute the second calculation step, which uses the pre-decision risk assessment model to calculate the current action gradient corresponding to the current pre-decision action according to the first formula, and outputs the current risk probability. The first formula is: , This refers to the pre-decision risk assessment model. This indicates the running status of the sample. This indicates the sample fault information. This represents the training pre-decision action after the k-th correction at time t. This means projecting the training pre-decision action onto n directions of the action safety space, where the action safety space is the space that ensures the sample pre-decision strategy satisfies the set safety constraints. The correction module is used to execute the second correction step, which transmits the current action gradient to the pre-decision strategy model and corrects the current pre-decision action using gradient descent according to the fourth formula, wherein the fourth formula is: , Indicates the current action gradient. This represents the current pre-decision action after the k-th correction at time t. This represents the current pre-decision action after the (k+1)th correction at time t. The step size indicates the degree of each correction. The repeat module is used to repeat the second calculation step and the second correction step at least once in sequence until the current risk probability is less than the risk threshold, thereby completing the safety correction of the current pre-decision action.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method according to any one of claims 1 to 6.

9. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images