Power system scenario reduction method and system, electronic device and storage medium

Abstract

A power system scenario reduction method and system, electronic equipment and storage medium, the method comprises the following steps: using an initial scenario set, establishing a scenario reduction problem model, and using an information entropy regularization algorithm to convex the scenario reduction problem model to obtain a regularized scenario reduction problem model; input the initial scenario set, the typical scenario set and the regularization coefficient, construct the scenario reduction network model, combine the regularized scenario reduction problem model and the scenario reduction network model, and calculate the regularized transmission matrix and the distance matrix; using the regularization coefficient, the regularized transmission matrix and the distance matrix, the loss function of the Sinkhorn distance considering the transportation cost and the entropy regularization penalty is solved; based on the loss function, the typical scenario set is trained in reverse gradient by using the scenario reduction network model, and the trained typical scenario set and the edge probability distribution of the typical scenario set are obtained. The present application can improve the feasibility of large-scale scenario reduction and the stability of solution.

Description

Technical Field

[0001] This invention belongs to the field of power system scenario analysis technology, specifically relating to a power system scenario reduction method, system, electronic device, and storage medium. Background Technology

[0002] With the rapid development of smart grid information technology, massive amounts of data samples have emerged in the power system. To effectively extract the characteristics of this big data and serve the power system, data clustering and feature mining have become crucial methods for effectively processing power big data. Therefore, how to extract effective and accurate uncertainty features from large amounts of renewable energy data has become a research hotspot.

[0003] In the dispatching and planning research of power systems with a high proportion of renewable energy, it is necessary to fully consider the uncertainty of renewable energy output in order to ensure that dispatching strategies and planning schemes can adapt to the characteristics of renewable energy output and fully absorb it. Scenario analysis can effectively characterize the uncertainties in the power system through a set of discrete scenarios, thereby transforming the uncertain optimization problem into a stochastic optimization problem. Scenario analysis is an effective method for describing uncertainties, and stochastic optimization theory based on scenario analysis has been widely studied in the field of power system optimization dispatching and planning. Using multiple scenarios to describe uncertainties and combining them with stochastic optimization can lead to more stable optimization solutions. However, due to computational complexity limitations, stochastic optimization cannot directly analyze massive amounts of scenario data; it is necessary to reduce the massive amounts of scenarios to improve the efficiency of solving stochastic optimization problems. Summary of the Invention

[0004] The purpose of this invention is to address the problems in the prior art by providing a power system scenario reduction method, system, electronic device, and storage medium. The method uses Wasserstein distance to describe the scenario reduction problem and transforms it into an unconstrained optimization problem through an entropy regularization penalty objective function. This allows the scenario reduction problem to be solved using a neural network training method, resulting in high computational efficiency and effective application to large-scale scenario reduction problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, a method for simplifying power system scenarios is provided, including:

[0007] Using the initial scene set, a scene reduction problem model is established, and the scene reduction problem model is convexified using the information entropy regularization algorithm to obtain the regularized scene reduction problem model.

[0008] Input the initial scene set, typical scene set and regularization coefficients, construct the scene reduction network model, and combine the regularized scene reduction problem model and the scene reduction network model to calculate the regularization transfer matrix and distance matrix.

[0009] By using the regularization coefficient, regularized transfer matrix, and distance matrix, the loss function of Sinkhorn distance considering transportation cost and entropy regularization penalty is obtained.

[0010] Based on the loss function, the typical scene set is trained by back gradient using the scene reduction network model to obtain the trained typical scene set and the marginal probability distribution of the typical scene set.

[0011] As a preferred embodiment of the method of the present invention, the step of establishing a scene reduction problem model using an initial scene set includes:

[0012] Let the initial number of scenes be n, and the number of typical scenes be m, where n >> m; X is the initial scene set, and Y is the typical scene set; the goal of the scene reduction problem is to find m typical scenes that approximately match the n initial scenes while minimizing the Wasserstein distance; the number of typical scene sets m is known, and the initial scenes have equal probabilities; let the distance between any scenes be d(x... i ,y j )=d ij The elements of the transfer matrix Π are π(x) i ,y j ) = π ij Then the expression for the scenario reduction problem model is:

[0013]

[0014] The probability measure v of the typical scene set includes the typical scene set Y and the marginal probability distribution Q = Π. T 1 n 1 n Let n be a unit column vector of length n; in the discrete scene reduction problem, the typical scene set satisfies the non-empty subset condition. condition.

[0015] As a preferred embodiment of the method of the present invention, the scene reduction problem model is convexified using the information entropy regularization algorithm, and the regularized expression of the scene reduction problem model is as follows:

[0016]

[0017] In the formula, ε is the regularization coefficient; considering only the equality constraints, a Lagrangian function analysis is constructed on the regularized objective function to obtain the optimal regularized transport matrix elements. The calculation expression is:

[0018]

[0019] The probability measure corresponding to the typical scenario set is obtained by solving the following formula:

[0020]

[0021] As a preferred embodiment of the method of the present invention, the input layer of the scene reduction network model includes an initial scene set, a typical scene set, and regularization coefficients; the hidden layer of the scene reduction network model calculates distance measurement through an explicit distance metric, and the optimal regularized transfer matrix is ​​calculated by equation (3); the output layer of the scene reduction network model is a loss function of Sinkhorn distance that considers transportation cost and entropy regularization penalty.

[0022] The loss function for Sinkhorn distance, which considers transportation costs and entropy regularization penalties, is as follows:

[0023]

[0024] As a preferred embodiment of the method of the present invention, the step of performing back gradient training on a typical scene set using a scene reduction network model includes:

[0025] Initialize the scene set X, Y and the regularization coefficient ε;

[0026] Based on the scene reduction network model, Y is trained to convergence through gradient backpropagation.

[0027] If the scene reduction problem is discrete scene reduction, then a typical scene set is selected from the initial scene set, and the typical scene set is updated using equation (6) to satisfy the non-empty subset constraint:

[0028]

[0029] If the scene reduction problem is continuous scene reduction, then Equation (7) is used to update the optimal transfer matrix, and Equation (8) is used to update the marginal probability distribution of typical scenes:

[0030]

[0031] Q = Π' T 1 n (8).

[0032] Secondly, a power system scenario simplification system is provided, including:

[0033] The problem model building and regularization module is used to build a scenario reduction problem model using the initial scenario set, and to make the scenario reduction problem model convex using the information entropy regularization algorithm to obtain the regularized scenario reduction problem model.

[0034] The network model building and calculation module is used to input the initial scene set, typical scene set and regularization coefficients, construct the scene reduction network model, and combine the regularized scene reduction problem model and the scene reduction network model to calculate the regularized transmission matrix and distance matrix.

[0035] The loss function solving module is used to solve for the loss function of Sinkhorn distance, which takes into account transportation costs and entropy regularization penalties, using regularization coefficients, regularization transmission matrix and distance matrix.

[0036] The reduction result acquisition module, based on the loss function, uses the scene reduction network model to perform back gradient training on the typical scene set, and obtains the trained typical scene set and the marginal probability distribution of the typical scene set.

[0037] As a preferred embodiment of the system of the present invention, the steps of the problem model establishment and regularization module in establishing a scenario reduction problem model using an initial scenario set include:

[0038] Let the initial number of scenes be n, and the number of typical scenes be m, where n >> m; X is the initial scene set, and Y is the typical scene set; the goal of the scene reduction problem is to find m typical scenes that approximately match the n initial scenes while minimizing the Wasserstein distance; the number of typical scene sets m is known, and the initial scenes have equal probabilities; let the distance between any scenes be d(x... i ,y j )=d ij The elements of the transfer matrix П are π(x) i ,y j ) = π ij Then the expression for the scenario reduction problem model is:

[0039]

[0040] Among them, the probability measure ν of the typical scene set includes the typical scene set Y and the marginal probability distribution Q = Π. T 1 n 1 n Let n be a unit column vector of length n; in the discrete scene reduction problem, the typical scene set satisfies the non-empty subset condition. condition.

[0041] As a preferred embodiment of the system of the present invention, the problem model establishment and regularization module utilizes the information entropy regularization algorithm to convex the scene reduction problem model, resulting in the following expression for the regularized scene reduction problem model:

[0042]

[0043] In the formula, ε is the regularization coefficient; considering only the equality constraints, a Lagrangian function analysis is constructed on the regularized objective function to obtain the optimal regularized transport matrix elements. The calculation expression is:

[0044]

[0045] The probability measure corresponding to the typical scenario set is obtained by solving the following formula:

[0046]

[0047] As a preferred embodiment of the system of the present invention, the input layer of the scene reduction network model constructed by the network model establishment and calculation module includes an initial scene set, a typical scene set, and regularization coefficients; the hidden layer of the scene reduction network model calculates the distance measurement through an explicit distance metric, and the optimal regularized transfer matrix is ​​calculated by equation (3); the output layer of the scene reduction network model is a loss function of Sinkhorn distance that considers transportation costs and entropy regularization penalties;

[0048] The loss function for Sinkhorn distance, which considers transportation costs and entropy regularization penalties, is as follows:

[0049]

[0050] As a preferred embodiment of the system of the present invention, the step of the reduction result acquisition module performing back gradient training on a typical scene set using a scene reduction network model includes:

[0051] Initialize the scene set X, Y and the regularization coefficient ε;

[0052] Based on the scene reduction network model, Y is trained to convergence through gradient backpropagation.

[0053] If the scene reduction problem is discrete scene reduction, then a typical scene set is selected from the initial scene set, and the typical scene set is updated using equation (6) to satisfy the non-empty subset constraint:

[0054]

[0055] If the scene reduction problem is continuous scene reduction, then Equation (7) is used to update the optimal transfer matrix, and Equation (8) is used to update the marginal probability distribution of typical scenes:

[0056]

[0057] Q = Π' T 1 n (8).

[0058] Thirdly, an electronic device is provided, comprising:

[0059] Memory, storing at least one instruction; and

[0060] The processor executes the instructions stored in the memory to implement the power system scenario reduction method.

[0061] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the power system scenario reduction method.

[0062] Compared with the prior art, the first aspect of the present invention has at least the following beneficial effects:

[0063] This invention proposes a model based on Wasserstein distance regularization to address stability issues. Traditional clustering methods directly solve non-convex scene reduction problems, which are difficult to improve in terms of stability. In contrast, this invention utilizes an information entropy regularization algorithm to convexize the scene reduction problem model, effectively reducing the problem of typical scene sets getting trapped in local solutions during the solution process, thereby improving the stability of the scene reduction problem solution. This invention uses a neural network to update and learn the typical scene set globally. By convexizing the scene reduction problem, Wasserstein distance regularization can be transformed into Sinkhorn distance, thus converting the scene reduction problem into an unconstrained optimization problem. This invention obtains a loss function for Sinkhorn distance that considers transportation costs and entropy regularization penalties. The scene reduction network model is then used for backpropagation gradient training on the typical scene set to obtain the trained typical scene set and its marginal probability distribution. This invention uses a neural network training method to solve the scene reduction problem, which can synchronously iterate the typical scene set instead of eliminating or retaining scenes one by one, thus effectively applying it to large-scale scene reduction problems.

[0064] It is understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 Flowchart of the power system scenario reduction method according to an embodiment of the present invention;

[0067] Figure 2 Block diagram of the simplified network model structure in this embodiment of the invention;

[0068] Figure 3 A simplified system structure block diagram for a power system scenario according to an embodiment of the present invention. Detailed Implementation

[0069] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0070] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0071] To accurately and efficiently solve the scene reduction problem, this invention proposes using Wasserstein distance to describe the problem and transforming the model into an unconstrained optimization by applying entropy regularization to the objective function. This allows for a rapid solution to the scene reduction problem using a neural network approach, resulting in high computational efficiency and applicability to large-scale scene reduction problems.

[0072] Technical solution of existing technology 1

[0073] The clustering-based scenario reduction method achieves scenario reduction through iterative updates of typical scenarios, as detailed below:

[0074] Clustering methods require dividing the initial scene set into multiple sets, where scenes within the same set have high similarity, and then selecting a representative scene from each set as the typical scene. Unlike selection reduction methods, clustering methods involve iteratively updating the selection of the typical scene set as a whole, rather than selecting scenes one by one.

[0075] The drawback of the existing technology is that the iterative process of the clustering method is affected by the initial point, and the convergence result of each iteration is random, which makes it easy to get trapped in local solutions. Therefore, the stability of the clustering method is poor.

[0076] Technical solution of existing technology 2

[0077] The reduction method is chosen, and scene reduction is achieved by iteratively removing / retaining scenes from the initial scene set, as detailed below:

[0078] Selection reduction methods are mainly divided into forward selection and backward reduction. Forward selection first places all scenes into the discard set. Under the condition of minimizing the probabilistic distance between the initial scene set and the typical scene set, typical scenes are selected one by one, and the process is repeated until the number of scenes reaches the number of typical scenes. Backward reduction is the dual method of forward selection. Backward reduction initially retains all scenes, and then selects scenes one by one and places them into the discard set until the number of retained scenes reaches the number of typical scenes.

[0079] The drawback of the existing technology 2 is that when the initial number of scenes is too large, the computational efficiency of the selection reduction method will decrease significantly due to the characteristics of traversing the selected scenes.

[0080] This invention addresses the shortcomings of technique one in its power system scenario reduction method by proposing a model based on Wasserstein distance regularization to solve the stability problem. By employing regularization to make the scenario reduction convex, the problem of typical scenario sets getting trapped in local solutions during the solution process can be effectively reduced. This improves the stability of the scenario reduction problem solution. Compared to traditional clustering methods, which directly solve non-convex scenario reduction problems, it is difficult to achieve effective stability improvements.

[0081] This invention addresses the shortcomings of technique two by utilizing a neural network to update and learn a set of typical scenarios holistically. By convexifying the scenario reduction problem, the Wasserstein distance can be regularized to a Sinkhorn distance, transforming the scenario reduction problem into an unconstrained optimization problem. The method employs neural network training to solve the scenario reduction problem, which can synchronously iterate over the set of typical scenarios, rather than eliminating / retaining scenarios one by one, thus effectively applying it to large-scale scenario reduction problems.

[0082] Example 1

[0083] This invention proposes a power system scenario reduction method, comprising three parts: establishing a regularized scenario reduction model, establishing a scenario reduction network model, and designing a scenario reduction algorithm. First, a scenario reduction model is constructed based on information entropy regularization. Second, an explicit scenario reduction network model is used to perform backpropagation gradient training on a typical scenario set. Finally, a scenario reduction algorithm is used to reduce the typical scenario set to probability information. Figure 1 As shown, the embodiments of the present invention specifically include the following steps:

[0084] S1. Using the initial scene set, establish a scene reduction problem model, and use the information entropy regularization algorithm to convex the scene reduction problem model to obtain the regularized scene reduction problem model.

[0085] S2. Input the initial scene set, typical scene set and regularization coefficients, construct the scene reduction network model, and combine the regularized scene reduction problem model and the scene reduction network model to calculate the regularization transfer matrix and distance matrix.

[0086] S3. Using the regularization coefficient, regularized transfer matrix, and distance matrix, the loss function of Sinkhorn distance considering transportation cost and entropy regularization penalty is obtained.

[0087] S4. Based on the loss function, the typical scene set is trained by back gradient using the scene reduction network model to obtain the trained typical scene set and the marginal probability distribution of the typical scene set.

[0088] 1. Regularized Scenario Reduction Model

[0089] (1) Problem of reducing the modeling scene

[0090] Assume there are n initial scenes and m typical scenes, where n >> m. Let X be the initial scene set and Y be the typical scene set. The goal of the scene reduction problem is to find m typical scenes that approximately match the n initial scenes while minimizing the Wasserstein distance. Assume the number of typical scenes in the m set is known, and the initial scenes have equal probabilities. Let the distance between any two scenes be d(x...). i ,y j )=d ij The elements of the transfer matrix П are π(x) i ,y j ) = π ij The scenario reduction problem is then defined as:

[0091]

[0092] Among them, the probability measure ν of the typical scene set includes the typical scene set Y and the marginal probability distribution Q = Π. T 1 n 1 n Let n be a unit column vector of length n. In the discrete scene reduction problem, the typical scene set needs to satisfy the non-empty subset condition. condition.

[0093] (2) Reduction problem in regularization scenarios

[0094] Scene reduction is a variational problem of Wasserstein distance, belonging to non-convex linear programming problems. The transfer matrix needs to satisfy numerous inequality constraints. To accelerate the computation of the optimal transfer problem, this invention proposes an information entropy regularization method to convexize the scene reduction problem. The regularized scene reduction problem model expression is as follows:

[0095]

[0096] Where ε is the regularization coefficient. Considering only equality constraints, constructing a Lagrangian function for the regularized objective function reveals the optimal regularized transport matrix elements. The expression is:

[0097]

[0098] The traditional scene reduction problem can be transformed into an unconstrained optimization problem by introducing entropy regularization. Only this problem needs to be solved to obtain the probability measure corresponding to the typical scene set.

[0099]

[0100] 2. Scene Reduction Network Model

[0101] After entropy regularization, the scenario reduction problem is equivalent to an unconstrained optimization problem. However, its nonlinear characteristics make it difficult to solve, and the solution process is affected by the distance measure d. For the unconstrained optimization problem, this invention designs a scenario reduction network model for solving it. The loss function of the scenario reduction network model is:

[0102]

[0103] The scene reduction network is trained on a typical scene set using inverse gradient training.

[0104] like Figure 2 As shown, the input layer of the scene reduction network in this embodiment of the invention includes an initial scene set, a typical scene set, and regularization coefficients. The hidden layer calculates the distance measurement through an explicit distance metric, and the optimal regularized transfer matrix is ​​calculated by equation (3). The output layer of the scene reduction network is a loss function of Sinkhorn distance that considers transportation costs and entropy regularization penalties.

[0105] from Figure 2 As can be seen from the diagram, the scene reduction network in this embodiment of the invention does not pass parameters between layers because the functional mapping between the input information and the output loss function is known. However, in the input layer, the typical scene set consists of trainable parameters. The mapping relationship of this network can also be modified by changing the calculation method between the parameters. For example, the regularization coefficient and the calculation method of the distance metric can be adjusted to adapt to scene reduction under different requirements.

[0106] 3. Scene Reduction Algorithm

[0107] Scene reduction encompasses both discrete and continuous scene reduction problems. To effectively address these different issues, this invention proposes a scene reduction algorithm that effectively handles the non-empty subset constraint problem in discrete scene reduction and the optimal transmission deviation problem introduced by regularization. The algorithm includes the following steps:

[0108] Initialize the scene set X, Y and the regularization coefficient ε;

[0109] Based on the scene reduction network model, Y is trained to convergence through gradient backpropagation.

[0110] If the scene reduction problem is discrete scene reduction, then a typical scene set is selected from the initial scene set, and the typical scene set is updated using equation (6) to satisfy the non-empty subset constraint:

[0111]

[0112] If the scene reduction problem is continuous scene reduction, then Equation (7) is used to update the optimal transfer matrix, and Equation (8) is used to update the marginal probability distribution of typical scenes:

[0113]

[0114] Q = Π' T 1 n (8)

[0115] Example 2:

[0116] A numerical example is given for both continuous scene reduction and discrete scene reduction to illustrate the effectiveness of the proposed method:

[0117] (1) Reduction of continuous scenes

[0118] In short-term power system dispatching, the reduction of time-series scenarios is akin to the reduction of discrete scenarios. Three sets of day-ahead wind power scenarios (n = 500 / 2000 / 10000) are generated based on these scenarios.

[0119] The accuracy comparison of Scene Reduction Network, Fast Forward Selection (FFS), and K-mediods clustering methods is shown in Table 1 below:

[0120] Table 1 Comparison of Reduction Accuracy in Discrete Scenarios

[0121]

[0122] (2) Discrete scene reduction

[0123] In medium- and long-term scheduling and planning, long-term wind power data from multiple wind farms (number of wind farms 1 = 2, 5, 10) are used as a continuous set of scenarios. Compared with time series scenarios, the scenarios at each stage do not have time characteristics. The data sampling interval is 1 hour, and each sample is used as one scenario, resulting in 8760 initial scenarios throughout 2019.

[0124] The accuracy comparison between the scene reduction network and three commonly used clustering methods is shown in Table 2 below:

[0125] Table 2 Comparison of Reduction Accuracy in Continuous Scenarios

[0126]

[0127] As can be seen from the results in Tables 1 and 2, the scene reduction method based on Wasserstein distance regularization proposed in this invention has good reduction accuracy and can be applied to large-scale scene reduction problems.

[0128] Example 3

[0129] Please see Figure 3 An embodiment of the present invention provides a power system scenario reduction system, comprising:

[0130] Problem model building and regularization module 1 is used to build a scenario reduction problem model using the initial scenario set, and to make the scenario reduction problem model convex using the information entropy regularization algorithm to obtain the regularized scenario reduction problem model.

[0131] The network model building and calculation module 2 is used to input the initial scene set, typical scene set and regularization coefficients, construct the scene reduction network model, and calculate the regularization transmission matrix and distance matrix by combining the regularized scene reduction problem model and the scene reduction network model.

[0132] Loss function solving module 3 is used to solve for the loss function of Sinkhorn distance that takes into account transportation costs and entropy regularization penalties by using regularization coefficients, regularization transmission matrix and distance matrix;

[0133] The reduction result acquisition module 4, based on the loss function, uses the scene reduction network model to perform back gradient training on the typical scene set, and obtains the trained typical scene set and the marginal probability distribution of the typical scene set.

[0134] In one possible implementation, the steps of the problem model building and regularization module 1 in building a scenario reduction problem model using an initial scenario set include:

[0135] Let the initial number of scenes be n, and the number of typical scenes be m, where n >> m; X is the initial scene set, and Y is the typical scene set; the goal of the scene reduction problem is to find m typical scenes that approximately match the n initial scenes while minimizing the Wasserstein distance; the number of typical scene sets m is known, and the initial scenes have equal probabilities; let the distance between any scenes be d(x... i ,y j )=d ij The elements of the transfer matrix Π are π(x) i ,y j ) = π ij Then the expression for the scenario reduction problem model is:

[0136]

[0137] Among them, the probability measure ν of the typical scene set includes the typical scene set Y and the marginal probability distribution Q = Π. T 1 n 1 n Let n be a unit column vector of length n; in the discrete scene reduction problem, the typical scene set satisfies the non-empty subset condition. condition.

[0138] In one possible implementation, the problem model building and regularization module 1 uses the information entropy regularization algorithm to convex the scene reduction problem model, resulting in the following expression for the regularized scene reduction problem model:

[0139]

[0140] In the formula, ε is the regularization coefficient; considering only the equality constraints, a Lagrangian function analysis is constructed on the regularized objective function to obtain the optimal regularized transport matrix elements. The calculation expression is:

[0141]

[0142] The probability measure corresponding to the typical scenario set is obtained by solving the following formula:

[0143]

[0144] In one possible implementation, the input layer of the scene reduction network model constructed by the network model building and calculation module 2 includes an initial scene set, a typical scene set, and regularization coefficients; the hidden layer of the scene reduction network model calculates distance measurement through an explicit distance metric, and the optimal regularized transfer matrix is ​​calculated by equation (3); the output layer of the scene reduction network model is a loss function of Sinkhorn distance that considers transportation costs and entropy regularization penalties.

[0145] The loss function for Sinkhorn distance, which considers transportation costs and entropy regularization penalties, is as follows:

[0146]

[0147] In one possible implementation, the step of the reduction result acquisition module 4 performing back gradient training on a typical scene set using a scene reduction network model includes:

[0148] Initialize the scene set X, Y and the regularization coefficient ε;

[0149] Based on the scene reduction network model, Y is trained to convergence through gradient backpropagation.

[0150] If the scene reduction problem is discrete scene reduction, then a typical scene set is selected from the initial scene set, and the typical scene set is updated using equation (6) to satisfy the non-empty subset constraint:

[0151]

[0152] If the scene reduction problem is continuous scene reduction, then Equation (7) is used to update the optimal transfer matrix, and Equation (8) is used to update the marginal probability distribution of typical scenes:

[0153]

[0154] Q = П' T 1 n (8)

[0155] This invention defines a scene reduction problem based on information entropy regularization, constructs an unconstrained model using Lagrange extrema, and solves it using a neural network model. This effectively improves the feasibility of large-scale scene reduction, while the regularization model improves the stability of the solution.

[0156] Example 4

[0157] Another embodiment of the present invention also provides an electronic device, comprising:

[0158] Memory, storing at least one instruction; and

[0159] The processor executes the instructions stored in the memory to implement the power system scenario reduction method.

[0160] Example 5

[0161] Another embodiment of the present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the power system scenario reduction method.

[0162] The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying the computer program code, a medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals. For ease of explanation, the above content only shows the parts related to the embodiments of the present invention; for specific technical details not disclosed, please refer to the method section of the embodiments of the present invention. This computer-readable storage medium is non-transitory and can be stored in storage devices formed by various electronic devices, enabling the execution process described in the method of the embodiments of the present invention.

[0163] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.

[0164] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0165] 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.

[0166] 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.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for reducing power system scenarios, characterized in that, include: Using the initial scene set, a scene reduction problem model is established, and the scene reduction problem model is convexified using the information entropy regularization algorithm to obtain the regularized scene reduction problem model. Input the initial scene set, typical scene set and regularization coefficients, construct the scene reduction network model, and combine the regularized scene reduction problem model and the scene reduction network model to calculate the regularization transfer matrix and distance matrix. By using the regularization coefficient, regularized transfer matrix, and distance matrix, the loss function of Sinkhorn distance considering transportation cost and entropy regularization penalty is obtained. Based on the loss function, the typical scene set is trained by back gradient using the scene reduction network model to obtain the trained typical scene set and the marginal probability distribution of the typical scene set. The steps for establishing a scene reduction problem model using the initial scene set include: Let the initial number of scenes be n, and the number of typical scenes be m, where n >> m; X is the initial scene set, and Y is the typical scene set; the goal of the scene reduction problem is to find m typical scenes that approximately match the n initial scenes while minimizing the Wasserstein distance; the number of typical scene sets m is known, and the initial scenes have equal probabilities; let the distance between any scenes be denoted as . Transmission Matrix The elements are Then the expression for the scenario reduction problem model is: （1） Among them, the probability measure of typical scenario sets Includes typical scene set Y and marginal probability distribution , Let n be a unit column vector of length n; in the discrete scene reduction problem, the typical scene set satisfies the non-empty subset condition. condition; The scene reduction problem model is convexified using the information entropy regularization algorithm, resulting in the following regularized expression for the scene reduction problem model: （2） In the formula, These are the regularization coefficients; considering only equality constraints, a Lagrangian function analysis is constructed on the regularized objective function to obtain the optimal regularized transport matrix elements. The calculation expression is: （3） The probability measure corresponding to the typical scenario set is obtained by solving the following formula: （4）。 2. The power system scenario reduction method according to claim 1, characterized in that, The input layer of the scene reduction network model includes an initial scene set, a typical scene set, and regularization coefficients; the hidden layer of the scene reduction network model calculates distance measurement through explicit distance metric, and the optimal regularized transfer matrix is ​​calculated by equation (3); the output layer of the scene reduction network model is a loss function of Sinkhorn distance that considers transportation cost and entropy regularization penalty. The loss function for Sinkhorn distance, which considers transportation costs and entropy regularization penalties, is as follows: （5）。 3. The power system scenario reduction method according to claim 1, characterized in that, The steps of using a scene reduction network model to perform backpropagation gradient training on a typical scene set include: Initialize scene set X, Y and regularization coefficients ; Based on the scene reduction network model, Y is trained to convergence through gradient backpropagation. If the scene reduction problem is discrete scene reduction, then a typical scene set is selected from the initial scene set, and the typical scene set is updated using equation (6) to satisfy the non-empty subset constraint: （6） If the scene reduction problem is continuous scene reduction, then Equation (7) is used to update the optimal transfer matrix, and Equation (8) is used to update the marginal probability distribution of typical scenes: （7） （8）。 4. A power system scenario reduction system, characterized in that, include: The problem model building and regularization module is used to build a scenario reduction problem model using the initial scenario set, and to make the scenario reduction problem model convex using the information entropy regularization algorithm to obtain the regularized scenario reduction problem model. The network model building and calculation module is used to input the initial scene set, typical scene set and regularization coefficients, construct the scene reduction network model, and combine the regularized scene reduction problem model and the scene reduction network model to calculate the regularized transmission matrix and distance matrix. The loss function solving module is used to solve for the loss function of Sinkhorn distance, which takes into account transportation costs and entropy regularization penalties, using regularization coefficients, regularization transmission matrix and distance matrix. The reduction result acquisition module, based on the loss function, uses the scene reduction network model to perform back gradient training on the typical scene set to obtain the trained typical scene set and the marginal probability distribution of the typical scene set. The problem model building and regularization module utilizes the initial scenario set to build a scenario reduction problem model, including the following steps: Let the initial number of scenes be n, and the number of typical scenes be m, where n >> m; X is the initial scene set, and Y is the typical scene set; the goal of the scene reduction problem is to find m typical scenes that approximately match the n initial scenes while minimizing the Wasserstein distance; the number of typical scene sets m is known, and the initial scenes have equal probabilities; let the distance between any scenes be denoted as . Transmission Matrix The elements are Then the expression for the scenario reduction problem model is: （1） Among them, the probability measure of typical scenario sets Includes typical scene set Y and marginal probability distribution , Let n be a unit column vector of length n; in the discrete scene reduction problem, the typical scene set satisfies the non-empty subset condition. condition; The problem model building and regularization module uses the information entropy regularization algorithm to convex the scene reduction problem model, resulting in the following expression for the regularized scene reduction problem model: （2） In the formula, These are the regularization coefficients; considering only equality constraints, a Lagrangian function analysis is constructed on the regularized objective function to obtain the optimal regularized transport matrix elements. The calculation expression is: （3） The probability measure corresponding to the typical scenario set is obtained by solving the following formula: （4）。 5. The power system scenario reduction system according to claim 4, characterized in that, The input layer of the scene reduction network model constructed by the network model establishment and calculation module includes an initial scene set, a typical scene set, and regularization coefficients; the hidden layer of the scene reduction network model calculates distance measurement through explicit distance metric, and the optimal regularized transfer matrix is ​​calculated by equation (3); the output layer of the scene reduction network model is a loss function of Sinkhorn distance that considers transportation cost and entropy regularization penalty. The loss function for Sinkhorn distance, which considers transportation costs and entropy regularization penalties, is as follows: （5）。 6. The power system scenario reduction system according to claim 4, characterized in that, The steps of the reduction result acquisition module in performing back gradient training on a typical scene set using a scene reduction network model include: Initialize scene set X, Y and regularization coefficients ; Based on the scene reduction network model, Y is trained to convergence through gradient backpropagation. If the scene reduction problem is discrete scene reduction, then a typical scene set is selected from the initial scene set, and the typical scene set is updated using equation (6) to satisfy the non-empty subset constraint: （6） If the scene reduction problem is continuous scene reduction, then Equation (7) is used to update the optimal transfer matrix, and Equation (8) is used to update the marginal probability distribution of typical scenes: （7） （8）。 7. An electronic device, characterized in that, include: Memory, storing at least one instruction; and The processor executes instructions stored in the memory to implement the power system scenario reduction method as described in any one of claims 1 to 3.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the power system scenario reduction method as described in any one of claims 1 to 3.

Images