Model data hybrid driven power system operation safety improvement method and device

Abstract

The invention discloses a method and device for improving operation safety of a model data hybrid driven power system, and the method comprises the steps: building a multi-scene optimization OPF model based on the output of a power system state sensing module, the calculation result of the model is used as the input of a data-driven scene similar category recommendation module and a model-driven scene similar category screening module; and performing linear equation solution on the multi-scene optimization OPF model by using the action constraint of the first k scene similar categories recommended by data driving, and screening out an optimal solution meeting an optimality condition. The device comprises a processor and a memory. According to the invention, the safety of the power system is improved, and the adaptability of the power system in practical application is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power system power flow optimization, and particularly to a method and device for improving the operation safety of a power system driven by a hybrid of model data. Background Art

[0002] The power source composition of the new power system is gradually evolving from being dominated by deterministic and adjustable conventional power sources to being dominated by new energy power generation with randomness, intermittency, and volatility. At the same time, new elements such as electric vehicles, virtual power plants, and distributed energy storage are emerging continuously at the load end, and the coupling of various energies such as electricity, gas, heat, cold, and hydrogen at the terminal is closer. [1] This makes both the source and load sides of the power system face greater uncertainties, and the operation modes are more diversified, decentralized, and differentiated. According to statistics, the maximum power fluctuation of a single new energy power station at the hourly level can reach 15% to 25% of the installed capacity; within 2 hours, its maximum fluctuation amplitude can even be as high as 40%. Considering that the maximum fluctuation amplitude of the new energy power fluctuation in the overall region within 2 hours can reach 20% to 35%. [2] It can be seen that with the large-scale access of new energy to the power grid, the power system faces an operation environment with strong uncertainties, and its impact on the power grid is changing from quantitative change to qualitative change, bringing new challenges to many aspects such as dispatching operation, planning and development, and reliability assessment.

[0003] Optimal Power Flow (OPF) is one of the most important basic analysis tools in power system planning and dispatching. With the increasing uncertainty of the new power system, traditional deterministic OPF methods are difficult to fully describe the safety dispatching requirements in a complex operation environment. Therefore, OPF is gradually evolving into a probability optimization problem with strong uncertainties. Among them, the multi-scenario method [3] , robust optimization [4] and chance-constrained [5] and other methods are widely used. Compared with methods such as robust optimization and chance-constrained, the advantage of multi-scenario optimal power flow is that it does not depend on specific uncertainty distribution assumptions, and the results are more intuitive and easier to understand. In addition, the multi-scenario method can also be used for simulation sampling of chance constraints to determine the set of scenarios that meet the constraint conditions at a given probability level. [6] However, the scheduling accuracy of the multi-scenario method is closely related to the number of scenarios, and the OPF calculation burden under a large number of scenarios is extremely large, seriously affecting its practical application in the field of power system safety dispatching.

[0004] In recent years, with the continuous expansion of the scale of the power system, the gradual opening of the power market, and the rapid development of the integrated energy system with deep coupling of multiple energies such as electricity, gas, heat, cold, and hydrogen, the secure dispatching of the power system faces unprecedented challenges. The superposition of multiple uncertainty factors makes the system operation mode more complex, and traditional deterministic dispatching methods are difficult to ensure security and reliability. Against this background, the optimization dispatching method based on multiple scenarios has become one of the effective means to improve the security of the power system. Multi-scenario OPF can evaluate the system state under different operating conditions, and then optimize the dispatching strategy to enhance the system's ability to resist uncertainties. However, the computational complexity of this method increases exponentially with the increase in the number of scenarios, making the solution efficiency the core bottleneck restricting its wide application. Therefore, on the premise of ensuring the optimization accuracy, how to improve the computational efficiency of multi-scenario OPF to support the secure and stable dispatching of the power system has become an important issue that needs to be solved urgently.

[0005] References

[0006] [1] Xin Bao'an. New Power System and New Energy System [M]. Beijing: China Electric Power Press, 2023: 8-11.

[0007] [2] Huang Yu. The Key to High Proportion of New Energy Integration Lies in Improving the Flexibility of the Power System [N]. China Energy News, 2021-07-12(2).

[0008] [3] Stai E, Liakopoulou A, Hug G. CoLinFlow: An iterative solution approach for the scenario-based AC OPF in active meshed distribution grids with batteries [J]. IEEE Transactions on Smart Grid, 2023, 14(6): 4282-4295.

[0009] [4] Roald L, Andersson G. Chance-constrained AC optimal power flow: reformulations and efficient algorithms [J]. IEEE Transactions on Power Systems, 2018, 33(3): 2906-2918.

[0010] [5]Brust J J, Anitescu M. Convergence analysis of fixed point chance-constrained optimal power flow problems[J]. IEEE Transactions on Power Systems, 2022, 37(6): 4191-4201.

[0011] [6]Vrakopoulou M, Li B, Mathieu J L. Chance constrained reserve scheduling using uncertain controllable loads part I: Formulation and scenario-based analysis[J]. IEEE Transactions on Smart Grid, 2019, 10(2): 1608-1617. Summary of the Invention

[0012] The present invention provides a method and device for improving the security of a power system driven by a hybrid of model data. By virtue of the powerful feature extraction and classification capabilities of data-driven algorithms, the present invention "priors" recommends possible scenario similarity categories through data training and learning, and then combines optimality conditions for "posterior" correction of the correctness of the optimization scheme, providing an efficient solution for improving the operating security of a power system considering multiple scenarios and enhancing the security of the power system; and can greatly improve the accuracy of the data-driven strategy through the model-driven strategy, as described in detail below:

[0013] In a first aspect, a method for improving the operating security of a power system driven by a hybrid of model data, the method includes:

[0014] Based on the output of the power system state perception module, a multi-scenario optimization OPF model is constructed, and the calculation results of the model are used as the input of the data-driven scenario similarity category recommendation module and the model-driven scenario similarity category screening module;

[0015] Using the active constraints of the top k scenario similarity categories recommended by the data-driven recommendation, the multi-scenario optimization OPF model is solved by linear equationization, and the optimal solutions that meet the optimality conditions are screened out.

[0016] Wherein, the model-driven scenario similarity category screening module is:

[0017] The data-driven scenario similarity category recommendation module obtains the prediction probabilities of each similar category that plays a role in the operation safety improvement model problem for a given multi-scenario, and selects the top k categories with the highest probabilities;

[0018] Introduce the features of these categories into the multi-scenario optimization OPF model, transform it into a simplified linear programming problem, and obtain k possible optimal solutions by solving this linear equation system;

[0019] Verify each optimal solution. If a certain solution meets the optimality conditions, then this solution is determined as the optimal solution.

[0020] Among them, the simplified linear programming problem is:

[0021] c s =[0c G 0]

[0022]

[0023] In the formula, c G represents the output cost vector of traditional units, renewable energy units, and virtual power plants; n g represents the total number of traditional units, renewable energy units, and virtual power plants; n l represents the total number of lines, I ng×ng is the n g ×n g identity matrix; I nl×nl is the n l ×n l identity matrix; c s , A s and b s represent the standard form of linear programming Y bus is the nodal admittance matrix; C G is the connection matrix; F max is the upper limit of line transmission power; are the upper and lower limits of the output of traditional units, renewable energy units, and virtual power plants; P d is the nodal power injection vector; Y f , Y t are the admittance matrices at the beginning and end of the line.

[0024] Among them, the optimal solution is:

[0025]

[0026] In the formula, x s (k) represents the k-th optimal solution of the optimal solution vector of scenario s.

[0027] Second aspect: An operation safety improvement device for a power system driven by a hybrid of model data, characterized in that the device includes: a processor and a memory, and program instructions are stored in the memory, and the processor calls the program instructions stored in the memory to enable the device to execute the method described in any one of the first aspect.

[0028] Third aspect: A computer-readable storage medium stores a computer program, the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the method described in any one of the first aspect.

[0029] The beneficial effects of the technical solution provided by the present invention are:

[0030] 1) High accuracy: The recommendation accuracy of the method proposed by the present invention in power systems of different scales exceeds 99%. This high accuracy benefits from the adoption of classification task learning. Compared with the traditional fitting learning method, the training of classification learning is easier and more stable; at the same time, the model-driven correction further improves the accuracy, excluding wrong solutions through optimality conditions to ensure the accuracy of the final solution.

[0031] 2) Significantly reduce load loss: When dealing with the security dispatching problem of large-scale power systems, the present invention significantly reduces the load loss caused by system uncertainties; compared with traditional optimization methods, this method effectively reduces the power outage duration through multi-scenario security optimization, resulting in a significant decrease in the annual load loss; for example, in the PEGASE 2869-node system, after adopting this method, the annual load loss is reduced from about 505 MW to about 214 MW, a reduction of more than 50%; greatly improving the security of the power system.

[0032] 3) Strong adaptability: Although the number of similar features increases as the scale of the power system increases; although the model training time increases as the scale of the power system increases, this training process is a one-time offline process. Once the training is completed, the model can be quickly applied online in different scenarios, greatly reducing the time cost in the real-time calculation process and ensuring the adaptability of the power system in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flowchart of an operation safety improvement method for a power system driven by a hybrid of model data.

[0034] Figure 2 It is a schematic diagram of the calculation time distribution of the OPF optimization algorithm and the algorithm of the present invention (GOC 10000-node system). DETAILED DESCRIPTION OF THE INVENTION

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the following further describes the embodiments of the present invention in detail. The embodiments of the present invention focus on achieving the following key objectives:

[0036] (1) Construct a framework for improving operation safety for multiple scenarios. By establishing a multi-scenario OPF model, systematically characterize the optimization and improvement problems under various potential operating environments. Consider uncertainty factors such as source-load fluctuations and topological changes during the operation stage to form a safe operation strategy adaptable to multiple operation scenarios;

[0037] (2) Construct a data-driven scenario similarity category recommendation model. Different from traditional "end-to-end" fitting learning, this method will use a convolutional neural network to predict the probabilities of a large number of scenarios of operation safety improvement problems belonging to each scenario similarity category;

[0038] (3) Propose a model-driven optimal solution verification and screening method for safety improvement in multiple scenarios. According to the top k categories with the highest recommendation probabilities, use the corresponding active constraints to transform the operation safety improvement problem in multiple scenarios into a system of equations and solve it, and screen out the optimal solution from these solutions through the optimality conditions.

[0039] Embodiment 1

[0040] The embodiment of the present invention provides a method for improving the operation safety of a power system driven by a hybrid of model data, which includes: a power system state perception module, a module for constructing an operation safety improvement model for multiple scenarios, a data-driven scenario similarity category recommendation module, and a model-driven scenario similarity category screening module. The purpose is to significantly improve the operation safety of the power system by mining and reusing the similar features between scenarios and solving the safety scheduling problem for multiple scenarios equationally. Among them:

[0041] I. Power system state perception module

[0042] This module is responsible for perceiving the voltage and power of the power system, establishing an OPF model, and finally outputting the power flow state of the power system in real time. By monitoring the key parameters of the power system, quickly obtain the electrical state information of the power system, and provide accurate initial data for subsequent OPF calculations.

[0043] II. Module for constructing an operation safety improvement model for multiple scenarios

[0044] Based on the output of the power system state perception module, this module constructs a multi-scenario optimization OPF model. This model characterizes the power flow characteristics of the power system under different operating scenarios and considers the impact of various uncertain factors on the optimal power flow calculation. Its core objective is to improve the applicability and reliability of OPF in complex power systems by establishing a mathematical optimization framework adaptable to multi-scenarios. Finally, the calculation results of this module are provided as input to the subsequent data-driven scenario similarity class recommendation module and the model-driven scenario similarity class screening module to support efficient power flow optimization calculations.

[0045] III. Data-driven Scenario Similarity Class Recommendation Module

[0046] Based on the module for constructing the operation safety improvement model for multi-scenarios, this module constructs a data-driven scenario similarity class recommendation model based on a convolutional neural network (CNN). Different from traditional "end-to-end" fitting learning methods, this model uses a convolutional neural network to train the OPF problem and predict the probability distribution of its possible belonging to each scenario similarity class. In this way, the power system can effectively identify and classify power flow optimization problems under different scenarios, improving the accuracy and processing efficiency of recommendations.

[0047] IV. Model-driven Scenario Similarity Class Screening Module

[0048] The input of this module consists of the safety improvement model provided by the module for constructing the operation safety improvement model for multi-scenarios (Module 2), and the top k scenario similarity classes recommended by the data-driven scenario similarity class recommendation module (Module 3) and their corresponding constraints. Based on the possible similarity classes given by the data-driven module, the solution space range is narrowed. Subsequently, the recommended classes are screened and corrected based on an accurate mathematical model to ensure the accuracy and safety of the final optimization results.

[0049] Example 2

[0050] The solution in Example 1 will be further introduced below in combination with specific calculation formulas and examples. See the following description for details:

[0051] I. Power System State Perception Module

[0052] The input and output of this module respectively correspond to the input and output of the OPF problem. This embodiment of the present invention focuses on the DC OPF model, and its objective function is:

[0053]

[0054] In the formula, N represents the set of nodes in the power system; and respectively represent the set of traditional generator sets and the set of renewable energy generator sets on node i of the power system; Vi PP Represents the virtual power plant set on power system node i; And respectively represent the output cost functions of the j-th generator set, renewable energy unit, and virtual power plant. In the embodiments of the present invention, a first-order linear cost function is adopted, and for more complex cost functions, the embodiments of the present invention are also applicable; And respectively represent the outputs of the j-th generator set, renewable energy unit, and virtual power plant.

[0055] Therefore, the output parameters of this module include: And

[0056] The constraint conditions are as follows:

[0057]

[0058] In the formula, L represents the power system line set; x ij represents the admittance of line ij; P L,i respectively represent the loads of power system node i; And respectively represent the upper and lower limits of the output of the j-th traditional generator set on node i; And respectively represent the upper and lower limits of the output of the j-th renewable energy unit on node i; And respectively represent the upper and lower limits of the output of the j-th virtual power plant on node i; represents the upper limit of the transmission power of line ij.

[0059] Therefore, the input parameters of this module include: P L,i , And

[0060] Writing the above DC OPF model in matrix form is as follows:

[0061]

[0062] In the formula, f LC is the objective function; C(·) represents various linearized cost functions; θ, And P L respectively are the node phase angle vector, traditional generator set output vector, renewable energy unit output vector, virtual power plant output vector, and node load vector; Y bus is the node admittance matrix; And respectively are the connection matrices of the traditional generator set, renewable energy unit, and virtual power plant; Fmax is the upper limit of the transmission power of the line; Y ft is the admittance matrix at the beginning and end of the line. and represent the upper and lower limits of the output of traditional generator sets respectively; and represent the upper and lower limits of the output of renewable energy generator sets respectively; and represent the upper and lower limits of the output of virtual power plants respectively; F max represents the upper limit of the transmission power of the line.

[0063] In the multi-scenario OPF calculation, since the topological structure of the power system often remains unchanged, the differences in the OPF problem are mainly manifested in the different values of the source-load parameters. Therefore, the input of the OPF model can be expressed as:

[0064]

[0065] Correspondingly, the output of the OPF model, that is, the optimal solution, can be expressed as:

[0066]

[0067] II. Construction Module of the Operating Safety Enhancement Model for Multi-Scenarios

[0068] In the embodiment of the present invention, the above-mentioned Optimal Power Flow (OPF) model is extended to a multi-scenario security dispatch model by introducing a variety of potential operating scenarios. This model not only considers the dynamic changes of the power system under different source-load states, but also covers the adjustment of the power system constraints caused by topological changes, which is represented by the scenario index s. Specifically, the multi-scenario modeling method characterizes the coordinated output of multiple sources such as traditional units, renewable energy, and virtual power plants, and constructs a multi-dimensional operating safety enhancement optimization framework in combination with load fluctuations, network topology adjustments, and external environmental disturbances. As shown in Equation (10):

[0069]

[0070] In the formula, π s is the probability of the occurrence of scenario s, and S represents the set of scenarios; θ s , and P L,s are the node phase angle vector, the output vector of traditional generator sets, the output vector of renewable energy generator sets, the output vector of virtual power plants, and the node load vector for scenario s respectively. and represent the upper and lower limits of the output of traditional generator sets for scenario s respectively; and represent the upper and lower limits of the output of renewable energy generator sets for scenario s respectively; and respectively represent the upper and lower limits of the virtual power plant output for scenario s.

[0071] Correspondingly, the output of the security dispatch model for multiple scenarios, i.e., the optimal solution, is corrected to:

[0072]

[0073] III. Data-driven Scenario Similarity Category Recommendation Module

[0074] The data-driven scenario similarity category recommendation module relies on deep learning algorithms to handle a "multi-classification" problem. Based on the input parameters of the security dispatch model for multiple scenarios, it uses a convolutional neural network to determine which category of scenario similarity the input state belongs to, and uses the softmax function to obtain the prediction probabilities of each category, so as to select the top k most likely categories for "recommendation".

[0075] Furthermore, based on the top k similar active constraints with the highest probabilities, the operation security improvement model for multiple scenarios is solved equationally, and the optimal solution is screened out from the obtained k solutions using the optimality conditions. If none of the k solutions pass the verification, it indicates that the data-driven method fails to successfully recommend the scenario similarity category of the current state. At this time, the traditional optimization algorithm needs to be used to re-solve the OPF problem to ensure the accuracy of the solution. Thanks to the design of model-driven screening, all solutions obtained through data-driven recommendation feature calculations will be subject to optimality verification, thus ensuring the accuracy of the results.

[0076] Introduce a Convolutional Neural Network (CNN) to construct a recommendation model for similarity categories. The convolutional layer is the core of the CNN and is responsible for extracting features from the input data. This layer contains multiple feature maps, and each feature map is generated by applying a convolutional kernel (also known as a filter or weight matrix) to the input data or the output of the previous layer. Through the convolution operation, the convolutional layer can extract different input features. When the input data is X, the feature map C of the convolutional layer is:

[0077]

[0078] In the formula, represents the convolution operation; W represents the weight value of the convolutional kernel; h represents the bias; f(·) represents the activation function, such as ReLU, Sigmoid or Tanh. The role of the activation function is to increase the non-linear ability of the network, enabling the CNN to learn and simulate more complex functions.

[0079] The pooling layer is located after the convolutional layer, and its main function is to perform feature dimensionality reduction. This process not only reduces the spatial size of the feature map but also retains important feature information, which helps reduce the computational amount and prevent overfitting. Pooling operations include: max pooling, average pooling, and stochastic pooling, etc. The input of the pooling layer is the output of the convolutional layer. Each pooling operation acts on a specific feature map, sliding within a predefined-sized window and taking the maximum value (max pooling) or the average value (average pooling) within the window.

[0080] The fully connected layer is usually located at the end of the CNN. After a series of convolutional and pooling operations, the fully connected layer "flattens" the output of all feature maps from the previous layer into a vector and performs one or more fully connected operations, finally outputting the final classification or regression result.

[0081] To adapt to the two-dimensional input of the CNN, in the embodiments of the present invention, the source-load parameter vector in the input data of Equation (8) is two-dimensionalized, and the load input vector P L is:

[0082]

[0083] In the formula, n represents the number of system nodes, and N is the result of rounding up the square root of n.

[0084] To reflect the node correspondence relationship, the parameter vectors and in the input data are both constructed into input matrices with reference to Equation (13) according to their access nodes.

[0085] The CNN model converts the network output into probability values by using the softmax function, where each probability represents the possibility that the input data belongs to each scene similarity category.

[0086] The last fully connected layer of the CNN generates an N c -dimensional vector, as follows:

[0087]

[0088] In the formula, N c represents the total number of scene similarity categories; z i is the unnormalized score that the input sample belongs to the i-th category; then each score in this vector is converted through the softmax function, as follows:

[0089]

[0090] In the formula, P(X∈i) is the probability that the given input sample data X is predicted to belong to category i; is the exponent of z i and is used to ensure that the output is positive.

[0091] In summary, the output Y of the CNN is the probability that the input sample X belongs to various categories with similar scenarios, as follows:

[0092] Y = P(X ∈ i) i = [1, 2,..., N c (16)

[0093] IV. Model-driven Scenario Similarity Category Screening Module

[0094] Using the effective constraints of the top k scenario similarity categories recommended by data-driven, solve the multi-scenario operation safety improvement model by linear equation, and screen out the optimal solution that meets the optimality conditions. It can be expressed as the standard form of the following linear programming problem:

[0095]

[0096] In the formula, θ is the free variable, θ = θ′ - θ″; y is the slack variable; P G is the output variable of the traditional unit, renewable energy unit and virtual power plant; are the upper and lower limits of the output of the traditional unit, renewable energy unit and virtual power plant; P d is the node power injection vector. C G is the connection matrix. Specifically as follows:

[0097]

[0098] Therefore, corresponding to the linear programming standard form c s , A s and b s are respectively:

[0099] c s = [0c G 0] (21)

[0100]

[0101] In the formula, c G represents the output cost vector of the traditional unit, renewable energy unit and virtual power plant; n g represents the total number of traditional units, renewable energy units and virtual power plants; n l represents the total number of lines; I ng×ng is the n g ×n g identity matrix; I nl×nl is the n l ×n l identity matrix; c s , A s and b sRepresent the standard form of linear programming Y bus is the nodal admittance matrix; C G is the connection matrix; F max is the upper limit of line transmission power; Y f and Y t are the admittance matrices at the beginning and end of the line.

[0102] For the above multi-scenario security scheduling problem, based on the data-driven scenario similarity class recommendation module, the embodiments of the present invention can obtain the probability that the given parameters X of the multi-scenario security scheduling problem belong to each similarity class. Select the top k classes with the highest recommended probability, denoted as Top k (Y), and substitute its scenario similarity features (active variable x a ) into the multi-scenario security scheduling problem, then the multi-scenario security scheduling optimization problem can be simplified into the following linear equations:

[0103]

[0104] Based on Top k (Y), obtain the optimal solutions of k possible multi-scenario security scheduling models, that is, formula (25). Subsequently, check each obtained solution one by one. If a certain solution satisfies the optimality condition, then this solution can be recognized as the optimal solution. If none of the k possible optimal solutions satisfy the optimality condition, it indicates that the screening fails, indicating that Top k (Y) does not cover the correct scenario similarity class of the multi-scenario operation security improvement problem, and only an optimization algorithm can be used to solve this problem.

[0105]

[0106] In the formula, x s (k) represents the k-th optimal solution of the optimal solution vector of scenario s.

[0107] Embodiment 2

[0108] The best implementation manner of the embodiments of the present invention aims to achieve the rapid calculation of the massive scenario optimal power flow of the power system through a highly integrated modular design. This implementation manner includes the following key components:

[0109] I. Power system state perception module

[0110] It is mainly responsible for perceiving the operating state of the power system and providing input data, optimization models, and output results for the Optimal Power Flow (OPF) problem. In this module, first, a DC Optimal Power Flow (DC-OPF) model is established with the goal of minimizing the output cost in the power system. The input parameters include: load data of system nodes, upper and lower limits of output of various generator sets, etc. Through a linearized cost function, the module constructs a mathematical model of the power system based on variables such as the voltage phase angle of nodes and the output of generator sets, and solves the optimal solution.

[0111] The core of the module is to represent the network topology structure of the power system in matrix form and optimize the dispatching of the outputs of traditional generator sets, renewable energy units, and virtual power plants to meet the load demands of each node. At the same time, through the transmission power limit of power lines and the node admittance matrix, the stable and optimal operation of the entire power system is ensured.

[0112] In the multi-scenario optimal power flow calculation, this module generates corresponding optimal solutions according to different input of source-load parameters, adapts to the changes in different operating scenarios, and ensures the efficiency and accuracy of the optimization results in the actual power system.

[0113] II. Module for Constructing a Model to Improve Operational Safety for Multiple Scenarios

[0114] Based on the output data of the power system state perception module (Module I), this module constructs a security dispatching model considering various potential operating scenarios. This model comprehensively considers the changes in the power system state under different operating scenarios and introduces various uncertain factors, including: source-load fluctuations, topological changes, and establishes a multi-scenario security optimization dispatching framework.

[0115] Through the calculation of this module, a security optimization solution for the module for constructing a model to improve operational safety for multiple scenarios can be generated, and this solution is used as input and provided to the data-driven scenario similarity category recommendation module (Module III) and the model-driven scenario similarity category screening module (Module IV) to improve the subsequent calculation efficiency and optimization effect.

[0116] III. Data-Driven Scenario Similarity Category Recommendation Module

[0117] Based on a Convolutional Neural Network (CNN) to process the input parameters of the OPF model, the similarity category to which the input state belongs is determined through a multi-classification problem (i.e., the multi-scenario optimal power flow problem). The convolutional layer extracts features, the pooling layer reduces the dimension of the features, reduces the computational amount and prevents overfitting, while the fully connected layer converts the network output into the probability of each category. The softmax function is used to calculate the prediction probability of each category, and the top k most similar categories are recommended according to the highest probability.

[0118] IV. Model-Driven Scenario Similarity Category Screening Module

[0119] Using the constraints of the top-k similarity categories recommended by data-driven, the problem of the security improvement model for multiple scenarios is solved by linear equation, aiming to screen out the optimal solutions that meet the optimality conditions.

[0120] First, based on the data-driven scenario similarity category recommendation module, obtain the predicted probabilities of each similarity category that constraints the operation security improvement model problem of the given multiple scenarios, and select the top-k categories with the highest probabilities.

[0121] Next, introduce the features of these categories into the optimization problem of the security improvement model for multiple scenarios, and transform it into a simplified linear programming problem. By solving this linear equation system, k possible optimal solutions are obtained.

[0122] Subsequently, each solution is verified. If a solution meets the optimality conditions, then this solution is determined as the optimal solution. If all k solutions cannot meet the optimality conditions, it indicates that the data-driven method fails to successfully recommend the similarity categories of this problem, and the power system will fallback to the traditional optimization algorithm for solution to ensure the accuracy of the solution.

[0123] In addition, the wide access of renewable energy makes the traditional operation optimization based on a single scenario lack security, and more potential scenarios need to be considered to account for security risks. However, multi-scenario optimization cannot meet the timeliness requirements of operation. And the embodiments of the present invention significantly improve the efficiency of multi-scenario security optimization from the method level (modules 3 and 4), thereby enhancing the operation reliability of the system during the operation stage.

[0124] Example 3

[0125] Table 1 shows the OPF calculation results based on 5000 samples of the experimental method (abbreviated as SSDMOPF method) proposed in the embodiments of the present invention under different examples. In terms of accuracy, the 10-time recommendation accuracies of the proposed data-driven model all exceed 99%, especially in the IEEE RTS-79, IEEE 118-bus, and IEEE 300-bus examples, the accuracy exceeds 99.9%. The achievement of this high accuracy benefits from the fact that the algorithm does not "end-to-end" fit the optimal solution of the security scheduling model for multiple scenarios, but conducts classification task learning on the problem of the security scheduling model for multiple scenarios. Compared with traditional fitting learning, the training of classification learning is easier. Through model-driven correction, the accuracy of the SSDMOPF algorithm is further improved because the algorithm excludes the wrongly recommended solutions through the optimality conditions and corrects them through the traditional optimization algorithm, thereby quickly and accurately obtaining the optimal solutions to the security scheduling model problems of 5000 potential scenarios.

[0126] Calculation time and speedup ratio are important indicators to measure the efficiency of the proposed algorithm, and the results are shown in Table 1. It can be seen that as the system scale of the test cases expands from hundreds of nodes to tens of thousands of nodes, the calculation time of the traditional optimization method increases significantly, while the proposed SSDMOPF algorithm can significantly improve the calculation speed. For example, in the PEGASE 2869-node and GOC 10000-node test cases, the speedup ratio reaches 85 times.

[0127] Under the traditional optimization method, the annual load loss of each test case is relatively high, and the load loss of large-scale systems such as IEEE 300, PEGASE 2869, and GOC 10000 is particularly significant, reaching up to several hundred megawatts. However, after applying this method, the load loss is reduced by more than 50%. Especially in large-scale power systems, for example, in the PEGASE 2869 test case, the annual load loss is reduced from about 505 MW to about 214 MW.

[0128] In Figure 2 In the analysis of the GOC 10000-node test case shown, there are significant differences in the calculation time distribution between the traditional optimization algorithm and the SSDMOPF algorithm. Specifically, the traditional optimization algorithm needs to be called 5000 times to solve the OPF optimization problem, with a total time consumption of 9848.89 s; while the SSDMOPF algorithm only needs to call the optimization algorithm 27 times, and the time consumption is only 41.36 s. It can be seen that the proposed method significantly reduces the number of calls to the optimization algorithm by adopting data-driven recommendation and fast solution of the OPF equation, and ensures the accuracy and efficiency of the solution results through the verification and screening of the optimality conditions.

[0129] Table 1 Analysis of OPF calculation results

[0130]

[0131] Example 4

[0132] An operation safety improvement device for a power system driven by a hybrid of model data, the device includes: a processor and a memory, and program instructions are stored in the memory. The processor calls the program instructions stored in the memory to make the device execute the following method steps in Example 1:

[0133] Based on the output of the power system state perception module, construct a multi-scenario optimization OPF model, and use the calculation results of the model as the input of the data-driven scenario similarity category recommendation module and the model-driven scenario similarity category screening module;

[0134] Use the active constraints of the top k scenario similarity categories recommended by data-driven to linearly solve the multi-scenario optimization OPF model, and screen out the optimal solutions that meet the optimality conditions.

[0135] Among them, the model-driven scenario similarity category screening module is as follows:

[0136] Based on the data-driven scenario similarity category recommendation module, obtain the prediction probabilities of each similar category that plays a role in the operation safety improvement model problem of the given multi-scenarios, and select the top k categories with the highest probabilities;

[0137] Introduce the features of these categories into the multi-scenario optimization OPF model, transform it into a simplified linear programming problem, and obtain k possible optimal solutions by solving this linear equation system;

[0138] Verify each optimal solution. If a certain solution meets the optimality condition, then this solution is determined as the optimal solution.

[0139] Among them, the simplified linear programming problem is:

[0140] c s =[0c G 0]

[0141]

[0142] In the formula, c G represents the output cost vector of traditional units, renewable energy units and virtual power plants; n g represents the total number of traditional units, renewable energy units and virtual power plants; n l represents the total number of lines, I ng×ng is the n g ×n g identity matrix; I nl×nl is the n l ×n l identity matrix; c s , A s and b s represent the standard form of linear programming Y bus is the nodal admittance matrix; C G is the connection matrix; F max is the upper limit of line transmission power; is the upper and lower limits of the output of traditional units, renewable energy units and virtual power plants; P d is the nodal power injection vector; Y f , Y t are the admittance matrices at the beginning and end of the line.

[0143] Among them, the optimal solution is:

[0144]

[0145] In the formula, x s (k) represents the k-th optimal solution of the optimal solution vector of scenario s.

[0146] It should be noted here that the device descriptions in the above embodiments correspond to the method descriptions in the embodiments, and the embodiments of the present invention will not be elaborated herein.

[0147] The execution subjects of the above-mentioned processor and memory can be devices with computing functions such as a computer, a single-chip microcomputer, a microcontroller, etc. In specific implementation, the embodiments of the present invention do not limit the execution subject, and it is selected according to the needs in actual applications.

[0148] Data signals are transmitted between the memory and the processor through a bus, and the embodiments of the present invention will not be elaborated herein.

[0149] Based on the same inventive concept, the embodiments of the present invention also provide a computer-readable storage medium. The storage medium includes a stored program, and when the program runs, it controls the device where the storage medium is located to execute the method steps in the above embodiments.

[0150] The computer-readable storage medium includes, but is not limited to, flash memory, hard disk, solid-state drive, etc.

[0151] It should be noted here that the description of the readable storage medium in the above embodiments corresponds to the method description in the embodiments, and the embodiments of the present invention will not be elaborated herein.

[0152] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present invention are generated in whole or in part.

[0153] The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or a data center that includes one or more integrated available media. The available medium can be a magnetic medium or a semiconductor medium, etc.

[0154] Except for special instructions, the embodiments of the present invention do not limit the models of each device, and any device that can perform the above functions can be used.

[0155] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred embodiment, and the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0156] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for improving the operation safety of an electric power system driven by hybrid model data, characterized in that: The method comprises: Based on the output of the power system state perception module, a multi-scenario optimization OPF model is constructed, and the calculation results of the model are used as the input of the data-driven scenario similarity category recommendation module and the model-driven scenario similarity category screening module; By using the constraints of similar categories of the first k scenarios in data-driven recommendation, the multi-scenario optimization OPF model is solved by linear equations, and the optimal solution that meets the optimality conditions is screened out.

2. The method for improving the operation safety of a power system driven by model data hybrid according to claim 1, characterized in that: The model-driven scene similarity category screening module is: Based on the data-driven scenario similarity category recommendation module, the predicted probability of each similarity category of the operational safety improvement model problem in a given multi-scenario is obtained, and the top k categories with the highest probability are selected; The features of these categories are introduced into the multi-scenario optimization OPF model and converted into a simplified linear programming problem. By solving the linear equations, k possible optimal solutions are obtained. Each optimal solution is verified, and if a solution meets the optimality condition, the solution is determined to be the optimal solution.

3. The method for improving the operation safety of a power system driven by model data hybrid according to claim 1, characterized in that: The simplified linear programming problem is: c s =[0c G 0] In the formula, c G Represents the output cost vector of traditional units, renewable energy units and virtual power plants; n g represents the total number of traditional units, renewable energy units and virtual power plants; n l Represents the total number of lines, I ng×ng n g ×n g The identity matrix of nl×nl n l ×n l The identity matrix of s , A s and b s Represents the standard linear programming form Y bus is the node admittance matrix; C G is the connection matrix; F max The upper limit of line transmission power; The upper and lower limits of the output of traditional units, renewable energy units and virtual power plants; P d Y is the node power injection vector; f , Y t is the admittance matrix at the beginning and end of the line.

4. The method for improving the operation safety of a power system driven by model data hybrid according to claim 1, characterized in that: The optimal solution is: In the formula, x s (k) represents the kth optimal solution of the optimal solution vector of scene s.

5. A device for improving the operation safety of an electric power system driven by model data hybrid, characterized in that: The device comprises: a processor and a memory, wherein program instructions are stored in the memory, and the processor calls the program instructions stored in the memory to enable the device to execute the method according to any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is enabled to perform the method according to any one of claims 1 to 4.

Images