Multi-tense power grid trajectory analysis method based on digital twinning
Through the multi-temporal grid trajectory analysis method based on digital twins, a digital twin model of the power grid and a multi-temporal grid trajectory deduction model are constructed, which solves the problem of difficulty in effectively utilizing distributed resources in multi-temporal and high-random scenarios in the existing technology, and realizes efficient grid trajectory deduction and analysis, which improves the stability and economics of the power grid.
Patent Information
- Application Number
- CN202411857515.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-03-21
AI Technical Summary
The prior art is difficult to effectively utilize distributed resources in multi-tenth and high-random scenarios to achieve the stability and economics of the power grid, and lacks an efficient multi-tenth grid trajectory deduction and analysis system.
Using a multi-temporal grid trajectory analysis method based on digital twins, a digital twin model of the power grid and a multi-temporal grid trajectory deduction model are constructed, and analyses and diagnosis are combined with historical and real-time data. Specific steps include: feature reduction, digital twin model construction, multi-temporal trajectory deduction model construction, trajectory data generation and analysis.
It improves the accuracy and real-time performance of the power grid, realizes multi-temporal dynamic deduction and analysis, supports grid operation data simulation, risk assessment and intelligent decision-making, and improves the adaptability and flexibility of the power grid in complex environments.
Smart Images
Figure CN119671414A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power grid analysis, and in particular to a multi-temporal power grid trajectory analysis method based on digital twins. Background Art
[0002] The energy system is a large-scale, complex, capital- and technology-intensive complex system in the world. It has the characteristics of nonlinearity, high dimensionality, layering, and distribution. These complexities are particularly prominent in the system's new energy access and flexible load management. The proportion of new energy in the energy system continues to increase. The uncertainty of wind power and photovoltaic power generation is significant, such as the scenarios of "extremely hot and no wind" and "late peak and no light", which makes the system balance more difficult and poses a severe challenge to the stability of power supply. With the widespread access of renewable energy and the increase in uncertainty in the power grid environment, traditional power grid monitoring and decision-making technologies have certain limitations in responding to the real-time complex power grid control needs. Especially in multi-temporal and high-random scenarios, how to make full use of the flexibility of distributed resources (source, network, load, storage) and improve the stability and economy of the system has become a problem that needs to be solved urgently. At present, there is still a lack of efficient multi-temporal power grid trajectory deduction and analysis system, which makes it difficult to realize cross-temporal power grid operation data simulation, risk assessment and intelligent decision support. Digital twin technology can reflect the operating status of the physical power grid in real time by establishing a virtual-real mapping power grid model, providing technical support for power grid control and optimization.
[0003] Although many policy measures have been introduced to support digital twin technology, it is still necessary to accelerate the construction of intelligent dispatching systems and digital twin platforms at this stage to support the stable operation and optimized dispatching of power grids in more complex scenarios. Therefore, it has important theoretical and practical significance. Summary of the invention
[0004] In view of this, the present invention aims at the deficiencies in the prior art and provides a multi-temporal power grid trajectory analysis method based on digital twins.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a multi-temporal power grid trajectory analysis method based on digital twins, comprising the following steps:
[0006] S1. Perform feature simplification on the original power grid sample set;
[0007] S2. Build a power grid digital twin model based on the original power grid sample set after feature reduction;
[0008] S3. Based on the original power grid sample set after feature reduction, a multi-temporal power grid trajectory deduction model is constructed;
[0009] S4. Generate trajectory data based on the digital twin model of the power grid, and input the hybrid data set consisting of trajectory data and real-time data into the multi-temporal power grid trajectory deduction model for analysis and diagnosis.
[0010] Furthermore, in S1, the original power grid sample set includes load demand, power generation, line loss and transformer temperature characteristic data.
[0011] Furthermore, in S1, the Relief algorithm is used to simplify the features of the original power grid sample set.
[0012] Furthermore, in S2, a generative adversarial network is used to construct a digital twin model of the power grid, which is used to simulate the operating status of the power grid and map the characteristics of the power grid in different scenarios, providing simulation support for subsequent multi-temporal power grid trajectory deduction.
[0013] Furthermore, in S3, the Stacking algorithm is used to construct a multi-temporal power grid trajectory deduction model.
[0014] Furthermore, the Relief algorithm is used to simplify the features of the original power grid sample set. The method is as follows:
[0015] Step 1: Initialization and random case selection
[0016] Select the original power grid sample set after feature reduction (D = {(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x n ,y n )}), where x i represents the feature vector, y i Represents the corresponding category label; initialize the weight W[j] of each feature to zero, (j=1,2,…,m), and randomly select an instance x from the sample set as the current instance;
[0017] Step 2: Find the nearest neighbor and calculate the similarity
[0018] Calculate the nearest neighbor of the current instance x; use the distance metric to calculate the similarity; for feature j, the similarity calculation can be defined as:
[0019]
[0020] where (d(x i ,x j )) is an instance of x i and x j The distance between them, σ is an adjustment parameter that controls the range of similarity;
[0021] Step 3: Feature weight update
[0022] For each feature j, update the feature weight according to the found similar and heterogeneous neighbors: increase the similarity score of the same neighbors to the current instance feature value, and reduce the similarity score of the heterogeneous neighbors;
[0023] W[j]←W[j]+similarity(x,x hit )-similarity(x,x miss )
[0024] Similar neighbors (x hit ): the instance in the same category that is most similar to x; heterogeneous neighbors (x miss ): the instance most similar to x in different categories;
[0025] Repeat steps 1 to 3 multiple times until the preset number of iterations is reached or convergence is achieved.
[0026] Furthermore, the generative adversarial network consists of a generator and a discriminator. The task of the generator is to generate data that is as real as possible from random noise. It accepts a random noise vector z as input and outputs the generated data G(z). The task of the discriminator is to distinguish whether the input data is real data x or forged data z generated by the generator. The output of the discriminator is D(x) or D(G(z)), which represents the authenticity probability of the input data.
[0027] The training process of the power grid digital twin model is achieved by minimizing the adversarial loss between the generator and the discriminator. This process is described by the following optimization objective function:
[0028]
[0029] P data (x) is the real data distribution, P z (z) is the noise distribution of the generator input, and E represents the expected value. The generator captures the distribution of the real data set and generates data that is as real as possible. The discriminator determines whether the data is real or not. Through continuous iterative adversarial training, the performance of both is improved at the same time. The goal of training is to achieve a balance state in which the generator can generate sufficiently real data so that the discriminator cannot distinguish between real data and generated data.
[0030] Furthermore, the Stacking algorithm is used to construct a multi-temporal power grid trajectory deduction model, the method is as follows:
[0031] Step 1: Select multiple different machine learning algorithms as base models;
[0032] Step 2: Use the original grid sample set after feature reduction to train all base models; each model will learn independently and generate predictions; use the base model to predict the original training set to obtain the prediction results of each base model;
[0033] Step 3: Use the prediction results of each base model as meta-features, select a base model as the meta-model, and use the meta-features to train the meta-model to obtain the final prediction results.
[0034] Compared with the prior art, the present invention has the following beneficial effects:
[0035] 1) Improve modeling accuracy and real-time performance: This invention overcomes the problems of insufficient accuracy and low real-time performance in the construction of traditional device-level digital twins by adopting a model data fusion method, thereby achieving accurate dynamic modeling and real-time interaction of distributed energy devices. This method can more accurately reflect the operating status of the equipment and improve the dispatching and optimization capabilities of the power grid.
[0036] 2) Multi-temporal dynamic deduction and analysis: This method is based on the dynamic deduction of multiple scenarios of "history-parallel-future", which can cover the real-time fluctuations and long-term trend changes of the power grid, and support power grid operation data simulation, risk assessment and intelligent decision-making in different scenarios. This makes the power grid more adaptable and flexible in the face of complex environments.
[0037] 3) Comprehensive risk assessment and optimization decision support: By building a comprehensive digital twin model, combined with historical operation data and real-time data, accurate risk assessment and sensitivity analysis are provided. This system provides efficient data support for the dispatch and optimization decision-making of the power grid, and helps to realize a smart, efficient, and low-carbon modern power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 is a flow chart of the method of the present invention;
[0039] Figure 2 is a flow chart of a generative adversarial network according to an embodiment of the present invention;
[0040] Figure 3 It is a flowchart of the Stacking algorithm implementation of an embodiment of the present invention. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.
[0042] Example: Figure 1 As shown, a multi-temporal power grid trajectory analysis method based on digital twins includes the following steps:
[0043] S1. Perform feature simplification on the original power grid sample set;
[0044] S2. Build a power grid digital twin model based on the original power grid sample set after feature reduction;
[0045] S3. Based on the original power grid sample set after feature reduction, a multi-temporal power grid trajectory deduction model is constructed;
[0046] S4. Generate trajectory data based on the digital twin model of the power grid, and input the hybrid data set consisting of trajectory data and real-time data into the multi-temporal power grid trajectory deduction model for analysis and diagnosis.
[0047] Specifically, in S1, the original grid sample set includes characteristic data such as load demand, power generation, line loss, and transformer temperature to better represent the dynamic characteristics of the grid.
[0048] In S1, the Relief algorithm is used to simplify the features of the original power grid sample set. The Relief algorithm is used to simplify the features of the original sample set to reduce the dimension of the data, extract key features, and ensure the efficiency and accuracy of the deduced data. The specific method is:
[0049] Step 1: Initialization and random case selection
[0050] Select the original power grid sample set after feature reduction (D = {(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x n ,y n )}), where x i represents the feature vector, y i Indicates the corresponding category label; initialize the weight W[j] of each feature to zero, (j=1,2,A,m) (assuming there are m features), and randomly select an instance x from the sample set as the current instance;
[0051] Step 2: Find the nearest neighbor and calculate the similarity
[0052] Calculate the nearest neighbor of the current instance x (i.e. the sample with the highest similarity); use a distance metric (such as Euclidean distance) to calculate the similarity; for feature j, the calculation of similarity can be defined as:
[0053]
[0054] where (d(x i ,x j )) is an instance of x i and x j The distance between them, σ is an adjustment parameter that controls the range of similarity;
[0055] Step 3: Feature weight update
[0056] For each feature j, update the feature weight according to the found similar and heterogeneous neighbors: increase the similarity score of the same neighbors to the current instance feature value, and reduce the similarity score of the heterogeneous neighbors;
[0057] W[j]←W[j]+similarity(x,x hit )-similarity(x,x miss )
[0058] Similar neighbors (x hit ): the instance in the same category that is most similar to x; heterogeneous neighbors (x miss ): the instance most similar to x in different categories;
[0059] Repeat steps 1 to 3 multiple times until the preset number of iterations is reached or convergence is achieved.
[0060] In S2, a generative adversarial network (GAN) is used to build a digital twin model of the power grid, which is used to accurately simulate the operating status of the power grid and map the characteristics of the power grid in different scenarios, providing simulation support for subsequent multi-temporal power grid trajectory deduction. The model is based on simplified power grid characteristics, combining physical models with data-driven methods to achieve accurate simulation of the power grid. By using machine learning or deep learning technology, a dynamic behavior model of the power grid can be established, which can reflect the operating status of the power grid in real time.
[0061] like Figure 2 As shown in the figure, the generative adversarial network (GAN) consists of two networks: the generator and the discriminator. The task of the generator is to generate data that is as real as possible from random noise. It accepts a random noise vector z as input and outputs the generated data G(z); the task of the discriminator is to distinguish whether the input data is real data x or forged data z generated by the generator. The output of the discriminator is D(x) or D(G(z)), which represents the authenticity probability of the input data.
[0062] The training process of the power grid digital twin model is achieved by minimizing the adversarial loss between the generator and the discriminator. This process is described by the following optimization objective function:
[0063] min G max D V(D,G)=E x～Pdata(x) [logD(x)]+E z～Pz(z) [log(1-D(G(z)))]
[0064] P data (x) is the real data distribution, P z (z) is the noise distribution of the generator input, and E represents the expected value. The generator captures the distribution of the real data set and generates data that is as real as possible. The discriminator determines whether the data is real or not. Through continuous iterative adversarial training, the performance of both is improved at the same time. The goal of training is to achieve a balance state in which the generator can generate sufficiently real data so that the discriminator cannot distinguish between real data and generated data.
[0065] In S3, the Stacking algorithm is used to build a multi-temporal power grid trajectory deduction model to analyze the behavioral changes of the power grid at different time nodes. The model can deduce the operation trajectory of the power grid at future moments based on historical data, helping to identify potential risks and optimize decisions.
[0066] like Figure 3 As shown in Figure 1, Stacking is an integrated learning method used to improve the predictive performance of the model. It combines the prediction results of multiple basic models (usually called "base models") to achieve better performance. The basic idea of Stacking is to use multiple different learning algorithms to learn the same data set and combine their outputs to form a stronger comprehensive model.
[0067] The Stacking algorithm is used to construct a multi-temporal power grid trajectory deduction model. The specific method is as follows:
[0068] Step 1: Select multiple different machine learning algorithms as base models, such as linear regression, decision tree, support vector machine, neural network, etc.
[0069] Step 2: Use the original grid sample set after feature reduction to train all base models; each model will learn independently and generate predictions; use the base model to predict the original training set to obtain the prediction results of each base model;
[0070] Step 3: Use the prediction results of each base model as meta-features, select a base model as the meta-model, and use the meta-features to train the meta-model to obtain the final prediction results.
[0071] In S4, trajectory data is generated based on the constructed power grid digital twin model, and the generated trajectory data is combined with real-time monitoring data to form a hybrid data set, which is input into the deduction model for analysis and diagnosis. Through in-depth analysis of the hybrid data set, the performance of the power grid under various operating conditions can be evaluated, anomalies and faults can be discovered in a timely manner, and accurate decision support can be provided.
[0072] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art can still modify or make equivalent substitutions to the specific implementations of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention are within the scope of protection of the claims of the present invention to be approved.
Claims
1. A multi-temporal power grid trajectory analysis method based on digital twins, characterized in that: The following steps are involved: S1. Perform feature simplification on the original power grid sample set; S2. Build a power grid digital twin model based on the original power grid sample set after feature reduction; S3. Based on the original power grid sample set after feature reduction, a multi-temporal power grid trajectory deduction model is constructed; S4. Generate trajectory data based on the digital twin model of the power grid, and input the hybrid data set consisting of trajectory data and real-time data into the multi-temporal power grid trajectory deduction model for analysis and diagnosis.
2. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 1 is characterized in that: In S1, the original grid sample set includes load demand, power generation, line loss and transformer temperature characteristic data.
3. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 1 is characterized in that: In S1, the Relief algorithm is used to simplify the features of the original power grid sample set.
4. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 1 is characterized in that: In S2, a generative adversarial network is used to construct a digital twin model of the power grid, which is used to simulate the operating status of the power grid and map the characteristics of the power grid in different scenarios, providing simulation support for subsequent multi-temporal power grid trajectory deduction.
5. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 1 is characterized in that: In S3, the Stacking algorithm is used to construct a multi-temporal power grid trajectory deduction model.
6. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 3 is characterized in that: The Relief algorithm is used to simplify the features of the original power grid sample set. The method is as follows: Step 1: Initialization and random case selection Select the original power grid sample set after feature reduction (D = {(x1, y1), (x2, y2), ..., (x n ,y n )}), where x i represents the feature vector, y i Indicates the corresponding category label; initialize the weight W[j] of each feature to zero, (j=1,2,A,m), and randomly select an instance x from the sample set as the current instance; Step 2: Find the nearest neighbor and calculate the similarity Calculate the nearest neighbor of the current instance x; use the distance metric to calculate the similarity; for feature j, the similarity calculation can be defined as: where (d(x i ,x j )) is an instance of x i and x j The distance between them, σ is an adjustment parameter that controls the range of similarity; Step 3: Feature weight update For each feature j, update the feature weight according to the found similar and heterogeneous neighbors: increase the similarity score of the same neighbors to the current instance feature value, and reduce the similarity score of the heterogeneous neighbors; W[j]←W[j]+similarity(x,x hit )-similarity(x,x miss ) Similar neighbors (x hit ): the instance in the same category that is most similar to x; heterogeneous neighbors (x miss ): the instance most similar to x in different categories; Repeat steps 1 to 3 multiple times until the preset number of iterations is reached or convergence is achieved.
7. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 4 is characterized in that: The Generative Adversarial Network consists of a generator and a discriminator. The task of the generator is to generate data that is as realistic as possible from random noise. It accepts a random noise vector z as input and outputs the generated data G(z). The task of the discriminator is to distinguish whether the input data is real data x or forged data z generated by the generator. The output of the discriminator is D(x) or D(G(z)), which represents the authenticity probability of the input data. The training process of the power grid digital twin model is achieved by minimizing the adversarial loss between the generator and the discriminator. This process is described by the following optimization objective function: min G max D V(D,G)=E x～Pdata(x) [logD(x)]+E z～Pz(z) [log(1-D(G(z)))] P data (x) is the real data distribution, P z (z) is the noise distribution of the generator input, and E represents the expected value; The generator captures the distribution of the real dataset and generates data that is as realistic as possible; The discriminator determines whether the data is real or not; through continuous iterative adversarial training, the performance of both is improved simultaneously. The goal of training is to achieve a state of equilibrium in which the generator can generate sufficiently real data so that the discriminator cannot distinguish between real data and generated data.
8. The multi-temporal power grid trajectory analysis method based on digital twin according to claim 5 is characterized in that: The Stacking algorithm is used to construct a multi-temporal power grid trajectory deduction model. The method is as follows: Step 1: Select multiple different machine learning algorithms as base models; Step 2: Use the original grid sample set after feature reduction to train all base models; each model will learn independently and generate predictions; use the base model to predict the original training set to obtain the prediction results of each base model; Step 3: Use the prediction results of each base model as meta-features, select a base model as the meta-model, and use the meta-features to train the meta-model to obtain the final prediction results.