Renewable energy scene controllable generation method and device based on depth generation model

Through the controllable generation method of renewable energy scenarios based on the deep generation model, the problem of difficult to generate renewable energy scenarios in the prior art with specific features and patterns is solved, especially under extreme meteorological conditions, and higher quality and credibility are achieved scene generation.

CN119089949APending Publication Date: 2024-12-06XI AN JIAOTONG UNIV +2
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202410891594.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prior art has difficulty generating renewable energy scenarios with specific characteristics and patterns, especially in extreme meteorological conditions and severe fluctuations, resulting in the credibility and quality of scenario generation being questioned.

Method used

Using a controllable generation method for renewable energy scenarios based on deep generation model, a controllable generation method is used to obtain historical data for preprocessing, a conditional deep generation model is established, a category weight allocation method and classification coding rules are designed, and a conditional adversarial generation network is used to generate specified features and patterns. scene.

Benefits of technology

The quality and credibility of scene generation are improved, especially in extreme meteorological conditions and severe fluctuations, and the learning efficiency and generalization ability of the model are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089949A_ABST
    Figure CN119089949A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of renewable energy sources, in particular to a renewable energy source scene controllable generation method and device based on a depth generation model, and can solve the problem that a renewable energy source scene with specific characteristics and modes needs to be controllably generated in practical application to a certain extent. The method comprises the following steps: firstly, obtaining wind power, photovoltaic and other renewable energy output and corresponding meteorological historical data, establishing a mathematical model generated by a renewable energy controllable scene according to the historical data, and meanwhile, designing a classification coding rule of a control condition; secondly, establishing a condition depth generation model suitable for controllable generation of a renewable energy scene, and designing a category weight distribution method to improve the learning efficiency of the renewable energy scene under extreme meteorological conditions with fewer samples or violent fluctuation; and finally, training the conditional depth generation model, and inputting the category codes of the control conditions into the trained conditional depth generation model, thereby controllably generating a specific renewable energy scene.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of renewable energy technology, and more specifically, to a method and device for controllable generation of renewable energy scenarios based on a deep generative model. Background Art

[0002] The extensive development of renewable energy is considered an important way to achieve low-carbon and sustainable transformation of power systems. However, the inherent intermittent and random nature of renewable energy poses great challenges to the operation, scheduling and planning of power systems. Therefore, accurately characterizing the uncertainty of renewable energy is crucial to meet this challenge. A widely used method for uncertainty modeling is scenario generation, which characterizes uncertainty by generating a set of representative renewable energy scenarios.

[0003] Traditional scenario generation methods are usually based on model implementation, but such model-based methods usually require complex or implicit prior assumptions, which makes the credibility and quality of the generated scenarios questionable. With the significant progress of machine learning technology, the application of model-free methods represented by deep generative models in the field of renewable energy scenario generation has attracted widespread attention. Due to the powerful fitting ability of deep neural networks, such methods can directly learn the distribution of real scenarios from historical data without the need for prior models.

[0004] However, in practical applications, there is a greater need for controllable generation of renewable energy scenarios with specific characteristics and patterns, especially scenarios under extreme meteorological conditions and with drastic fluctuations. These scenarios are crucial to the supply and demand balance of the power system, but they usually only account for a small part of the historical data set, posing a huge challenge to the learning-based model-free approach. Summary of the invention

[0005] In order to solve the problem that in practical applications, there is a greater need for controllable generation of renewable energy scenarios with specific characteristics and patterns, especially scenarios under extreme meteorological conditions and with drastic fluctuations, but these usually only account for a small part of historical data sets, the present application provides a method and device for controllable generation of renewable energy scenarios based on a deep generative model.

[0006] The embodiment of the present application is implemented as follows:

[0007] In a first aspect, the present application provides a method and device for controllable generation of renewable energy scenarios based on a deep generative model, including:

[0008] Obtain historical output data of renewable energy and corresponding historical meteorological data, and perform data preprocessing;

[0009] Generate mathematical models based on renewable energy controllable scenarios, where the control conditions are meteorological conditions or the degree of fluctuation of renewable energy output;

[0010] By analyzing the correlation between renewable energy output and various meteorological factors in historical data, the meteorological factors with the highest correlation are selected as control conditions, and classification and coding rules for control conditions are established based on whether the meteorological conditions are extreme or whether the renewable energy output fluctuates violently;

[0011] Establish a conditional deep generation model suitable for the controllable generation of renewable energy scenarios, and design a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output;

[0012] The conditional deep generation model is trained using the preprocessed historical data, and the category code of the control condition is input into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

[0013] In a possible implementation, the obtaining of historical output data of renewable energy and corresponding historical meteorological data and performing data preprocessing further includes:

[0014] Obtain historical data on wind power, photovoltaic and other renewable energy output corresponding to multiple renewable energy sites or power system nodes;

[0015] Obtain historical meteorological data corresponding to the historical data of renewable energy output. The scale, time interval and length of meteorological data are consistent with the renewable energy output data. Meteorological factors include solar radiation, temperature, wind speed, relative humidity and rainfall, etc.;

[0016] Historical data are preprocessed through outlier processing, missing value processing and data normalization methods.

[0017] In a possible implementation, the generating mathematical model based on the renewable energy controllable scenario, wherein the control condition is the meteorological condition or the fluctuation degree of renewable energy output, further includes:

[0018] Mathematical modeling of renewable energy scenarios:

[0019] x i ={x j,t |j∈N,t∈T}

[0020] Among them, x i represents a renewable energy output scenario; x j,t represents the active output value of renewable energy of the station or power system node j in time period t; N and Τ represent the set of stations or power system nodes and output time periods in a scenario respectively;

[0021] Establish the control condition vector c corresponding to the renewable energy output scenario i , which is mainly composed of meteorological conditions or the degree of fluctuation of renewable energy output;

[0022] The goal of renewable energy controllable scenario generation is to generate scenarios with similar characteristics and patterns to historical scenarios corresponding to the control conditions under given control conditions, and the data distribution of the generated scenarios is close to that of the historical scenarios.

[0023] In a possible implementation, the analysis of the correlation between the output of renewable energy and various meteorological factors in historical data selects the meteorological factor with the highest correlation as the control condition, and establishes a classification coding rule for the control condition according to whether the meteorological condition is extreme or whether the output of renewable energy fluctuates violently, further comprising:

[0024] According to the distribution of historical data of various meteorological factors, the upper and lower thresholds for determining extreme meteorological conditions are set. Meteorological conditions with meteorological factor values ​​exceeding the upper threshold or below the lower threshold are considered extreme meteorological conditions, and the corresponding renewable energy scenarios are considered extreme scenarios. Meteorological conditions and renewable energy scenarios corresponding to other meteorological factor values ​​are normal meteorological conditions and normal scenarios.

[0025] According to the distribution of historical data on renewable energy output fluctuations, threshold ranges for determining severe, moderate and small fluctuation conditions are set, corresponding to severe fluctuation scenarios, moderate fluctuation scenarios and small fluctuation scenarios respectively;

[0026] Extreme meteorological conditions and normal meteorological conditions as well as severe, moderate and small fluctuation conditions are classified and coded separately.

[0027] In a possible implementation, the establishment of a conditional deep generation model suitable for controllable generation of renewable energy scenarios and the design of a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output further include:

[0028] According to the mathematical model of controllable generation of renewable energy scenarios, the generator, discriminator, target value function and loss function of the conditional deep generative model based on the conditional adversarial generative network are constructed, and the Wasserstein distance and gradient penalty are introduced;

[0029] According to the classification encoding of the control conditions, a category weight allocation method is designed to alleviate the category imbalance phenomenon, and it is introduced into the loss function to improve the learning efficiency of small sample categories.

[0030] In a possible implementation, the conditional deep generative model is trained using the preprocessed historical data, and the category code of the control condition is input into the trained conditional deep generative model to controllably generate renewable energy scenarios with corresponding features and patterns, further comprising:

[0031] Set training parameters and initialize the network parameters of the model's generator and discriminator;

[0032] Sampling a real data sample containing a renewable energy scenario and an encoded control condition vector from a historical data distribution, sampling a random noise vector and a noise data sample of the encoded control condition vector from a noise distribution, and sampling a random variable from a uniform distribution of [0,1];

[0033] The generator generates renewable energy scenarios based on noise data samples, and the discriminator discriminates the scenarios generated by the generator based on real data samples;

[0034] The loss function is calculated based on the results of the generator and the discriminator, and the network parameters of the generator and the discriminator are gradient updated according to the loss function through the optimization algorithm until the generator network parameters converge.

[0035] In a possible implementation, the preprocessing of historical data by outlier processing, missing value processing and data normalization methods further includes:

[0036] According to the 3σ criterion, outliers in historical data are identified and judged. When the sample distance average is greater than 3σ, the data sample is considered to be an outlier. For the identified outliers, they can be regarded as missing values ​​and processed according to the missing value method, or directly deleted.

[0037] The missing values ​​in the sample are filled by the K nearest neighbor algorithm to ensure the integrity of historical data;

[0038] The maximum and minimum normalization method is used to map the original historical data into the range of [0,1].

[0039] In a possible implementation, the normalization formula is as follows:

[0040]

[0041] Among them, v and v* correspond to the original and normalized historical data set vectors respectively, and vmax and vmin correspond to the maximum and minimum values ​​in the historical data set respectively.

[0042] In a possible implementation, the method further includes:

[0043] The performance of the conditional deep generation model is tested using real renewable energy output data and corresponding meteorological condition data.

[0044] In a second aspect, the present application provides a controllable generation device for renewable energy scenarios based on a deep generation model, comprising:

[0045] A data acquisition module is used to obtain historical output data of renewable energy and corresponding historical meteorological data, and perform data preprocessing;

[0046] The first model building module is used to generate a mathematical model based on a renewable energy controllable scenario, wherein the control condition is a meteorological condition or a fluctuation degree of renewable energy output;

[0047] The rule establishment module is used to analyze the correlation between renewable energy output and various meteorological factors in historical data, select the meteorological factor with the highest correlation as the control condition, and establish classification and coding rules for the control condition according to whether the meteorological condition is extreme or whether the renewable energy output fluctuates violently;

[0048] The second model building module is used to establish a conditional deep generation model suitable for the controllable generation of renewable energy scenarios, and to design a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output;

[0049] The model training module is used to train the conditional deep generation model using the preprocessed historical data, and then input the category code of the control condition into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

[0050] The technical solution provided by this application can at least achieve the following beneficial effects:

[0051] The controllable generation method of renewable energy scenarios based on deep generative models provided in this application establishes a conditional deep generative model based on a conditional adversarial generative network. Compared with traditional methods, this model can directly learn the real scene distribution from historical data without the need for a priori models, thereby effectively improving the quality and credibility of scene generation.

[0052] By proposing a model for controllable generation of renewable energy scenarios whose control conditions are meteorological conditions or output fluctuation degree, and designing classification and coding rules for control conditions, renewable scenarios with specified characteristics and patterns can be generated.

[0053] The conditional adversarial generative network was improved by using Wasserstein distance and gradient penalty, and a category weight allocation method was designed to solve the imbalance problem in scene categories, thereby effectively improving the learning efficiency and generalization ability of the model, especially the generation quality of some small sample extreme weather scenes or drastic fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0055] Figure 1 It is a flowchart of a controllable generation method of renewable energy scenarios based on a deep generative model, shown in an exemplary embodiment of the present application;

[0056] Figure 2 It is a schematic diagram of a framework of a controllable generation method of renewable energy scenarios based on a deep generative model, shown in an exemplary embodiment of the present application;

[0057] Figure 3 It is a wind power output scenario with violent fluctuation characteristics generated by a controllable generation method of renewable energy scenarios based on a deep generation model shown in an exemplary embodiment of the present application;

[0058] Figure 4 It is a photovoltaic output scenario under extreme meteorological conditions generated by a controllable generation method of renewable energy scenarios based on a deep generation model shown in an exemplary embodiment of the present application;

[0059] Figure 5 It is a structural schematic diagram of a controllable generation device for renewable energy scenarios based on a deep generative model, shown as an exemplary embodiment of the present application.

[0060] Reference numerals:

[0061] 1. Data acquisition module; 2. First model building module; 3. Rule building module; 4. Second model building module; 5. Model training module. DETAILED DESCRIPTION

[0062] In order to make the purpose, implementation mode and advantages of the present application clearer, the exemplary implementation mode of the present application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only part of the embodiments of the present application, not all of the embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0063] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0064] The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.

[0065] The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.

[0066] Before explaining the controllable generation method of renewable energy scenarios based on a deep generative model provided in an embodiment of the present application, the application scenario and implementation environment of the embodiment of the present application are first introduced.

[0067] The extensive development of renewable energy is considered an important way to achieve low-carbon and sustainable transformation of power systems. However, the inherent intermittent and random nature of renewable energy poses great challenges to the operation, scheduling and planning of power systems. Therefore, accurately characterizing the uncertainty of renewable energy is crucial to meet this challenge. A widely used method for uncertainty modeling is scenario generation, which characterizes uncertainty by generating a set of representative renewable energy scenarios.

[0068] Traditional scenario generation methods are usually based on model implementation, but such model-based methods usually require complex or implicit prior assumptions, which makes the credibility and quality of the generated scenarios questionable. With the significant progress of machine learning technology, the application of model-free methods represented by deep generative models in the field of renewable energy scenario generation has attracted widespread attention. Due to the powerful fitting ability of deep neural networks, such methods can directly learn the distribution of real scenarios from historical data without the need for prior models.

[0069] However, in practical applications, there is a greater need for controllable generation of renewable energy scenarios with specific characteristics and patterns, especially scenarios under extreme meteorological conditions and with drastic fluctuations. These scenarios are crucial to the supply and demand balance of the power system, but they usually only account for a small part of the historical data set, posing a huge challenge to the learning-based model-free approach.

[0070] Based on this, the present application provides a controllable generation method and device for renewable energy scenarios based on a deep generation model. By improving the deep generation model and designing a category weight allocation method, it effectively realizes the generation of renewable energy scenarios with specified characteristics and patterns, and improves the efficiency and quality of scene generation.

[0071] Next, the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems will be described in detail through embodiments and in combination with the accompanying drawings. The embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all of them.

[0072] Figure 1 It is a flow chart of a controllable generation method of renewable energy scenarios based on a deep generative model, shown as an exemplary embodiment of the present application.

[0073] In an exemplary embodiment, Figure 1 As shown, a controllable generation method of renewable energy scenarios based on a deep generative model is provided. In this embodiment, the method may include the following steps:

[0074] Step 100: Obtain historical output data of renewable energy and corresponding historical meteorological data, and perform data preprocessing.

[0075] Step 200: Generate a mathematical model based on a renewable energy controllable scenario, wherein the control condition is a meteorological condition or a fluctuation degree of renewable energy output.

[0076] Step 300: By analyzing the correlation between renewable energy output and various meteorological factors in historical data, the meteorological factor with the highest correlation is selected as the control condition, and classification and coding rules for the control condition are established according to whether the meteorological condition is extreme or whether the renewable energy output fluctuates violently.

[0077] Step 400: Establish a conditional deep generation model suitable for controllable generation of renewable energy scenarios, and design a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output.

[0078] Step 500: Use the preprocessed historical data to train the conditional deep generation model, and then input the category code of the control condition into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

[0079] Figure 2 It is a framework diagram of a controllable generation method of renewable energy scenarios based on a deep generative model, shown as an exemplary embodiment of the present application.

[0080] In one possible implementation, Figure 2 As shown, the controllable generation method is implemented by the following steps:

[0081] (1) Obtain historical data on renewable energy output and corresponding meteorological information and perform data preprocessing

[0082] Obtain historical data on the output of renewable energy such as wind power and photovoltaic power corresponding to multiple renewable energy sites or power system nodes. The data is in the form of time series data.

[0083] The historical meteorological data corresponding to the historical data of renewable energy output are obtained. The scale, time interval and length of the meteorological data are consistent with the renewable energy output data. The meteorological factors include solar radiation, temperature, wind speed, relative humidity and rainfall.

[0084] Historical data are preprocessed through outlier processing, missing value processing and data normalization methods to ensure the quality and consistency of real historical data.

[0085] Specifically, we first use the 3σ criterion to identify and judge outliers in historical data. When the sample distance from the average value is greater than 3σ, the data sample is considered an outlier. For the identified outliers, we can regard them as missing values ​​and process them according to the missing value method, or delete them directly.

[0086] Secondly, the K nearest neighbor algorithm fills the missing values ​​in the sample to ensure the integrity of the historical data.

[0087] Finally, the maximum and minimum normalization method is used to map the original historical data into the range of [0,1]. The normalization formula is as follows:

[0088]

[0089] Among them, v and v * Corresponding to the original and normalized historical data set vectors, v max and v min They correspond to the maximum and minimum values ​​in the historical data set respectively.

[0090] (2) Establish a mathematical model for generating controllable scenarios of renewable energy

[0091] The mathematical model of a typical renewable energy scenario can be expressed as:

[0092] x i ={x j,t |j∈N,t∈T}

[0093] Among them, x i represents a renewable energy output scenario; x j,t It represents the active output value of renewable energy of station or power system node j in time period t; N and Τ represent the set of stations or power system nodes and output time periods in a scene respectively, the number of stations or power system nodes in N is 1 or more, and Τ usually includes the output time periods within a day.

[0094] Traditional scenario generation tasks aim to directly generate scenarios that are similar to the distribution of historical data.

[0095] However, it is of greater significance and utility to generate renewable energy scenarios under specific conditions, which are usually intrinsic features of the scenarios or external factors associated with the scenarios.

[0096] Thus establishing a scenario with renewable energy output x i The corresponding control condition vector c i , which is mainly composed of meteorological conditions or the degree of fluctuation of renewable energy output.

[0097] Therefore, the goal of renewable energy controllable scenario generation is to generate scenarios with similar characteristics and patterns to the historical scenarios corresponding to the control conditions under given control conditions, and the data distribution of the generated scenarios is close to that of the historical scenarios.

[0098] (3) Classification and coding rules for control conditions

[0099] 1) Meteorological factors screening

[0100] Since historical data include a variety of meteorological factors, such as solar radiation, temperature, wind speed, relative humidity and rainfall, not all meteorological factors have a significant impact on renewable energy output. Moreover, for different renewable energy sources such as wind power and photovoltaics, the meteorological factors that have the greatest impact on them are also different.

[0101] The correlation coefficient is an indicator used to measure the strength of the correlation between two variables. The most common correlation coefficient is the Pearson correlation coefficient, which is calculated as follows:

[0102]

[0103] Where r represents the Pearson correlation coefficient; vi and u i They represent the i-th value of two variables, specifically the active output value of renewable energy and the value of meteorological factors; and represent the means of the two variables.

[0104] Meteorological factor screening is achieved by calculating the correlation between renewable energy output and various meteorological factors in historical data, and then selecting the meteorological factor with the highest correlation as the control condition.

[0105] 2) Coding rules for control conditions

[0106] Control condition vector c i Designed to:

[0107]

[0108] in, and They represent the meteorological condition category and renewable energy fluctuation category encoded in the i-th scenario respectively; c i The two expressions represent renewable energy scenarios under specific meteorological conditions or with specific fluctuation levels. and The classification coding rules are as follows:

[0109]

[0110] Among them, m is the meteorological factor with the highest correlation with the output value of renewable energy. i represents the mean value of meteorological factor m in the entire period in the i-th scene; m H and m L They correspond to the upper and lower thresholds for judging extreme meteorological conditions respectively.

[0111] Codes 2 and 1 represent two different types of extreme meteorological conditions respectively; code 0 represents normal meteorological conditions.

[0112]

[0113] Among them, δ i represents the maximum volatility of renewable energy output in the i-th scenario. If there are multiple sites or nodes, the average is taken; δ H and δ L It is the threshold used to judge violent fluctuations and moderate fluctuations.

[0114] Codes 2, 1, and 0 represent violent fluctuations, medium fluctuations, and small fluctuations, respectively. i The calculation formula is as follows:

[0115]

[0116] (4) Establish a conditional deep generation model for controllable generation of renewable energy scenarios

[0117] According to the mathematical model of controllable generation of renewable energy scenarios, the generator, discriminator, objective value function and loss function of the conditional deep generative model with the conditional adversarial generative network as the main body are constructed, and the Wasserstein distance and gradient penalty are introduced.

[0118] 1) Conditional Deep Generative Model

[0119] Generative adversarial networks provide an unsupervised learning framework that uses a game-theoretic adversarial process to train generative models, characterized by a dual network architecture consisting of a generator G and a discriminator D.

[0120] The main goal of the generator G is to generate data that is very similar to the distribution of historical data, while the discriminator D strives to distinguish between real historical data and generated data.

[0121] The conditional generative adversarial network introduces data sample labels as additional information during the training process.

[0122] In the controllable generation of renewable energy scenarios, the input of the generator of the conditional generative adversarial network is a network with a predetermined distribution P z (z) is a combination of random noise z and control condition c, and outputs the generated scene It can be expressed as:

[0123]

[0124] The input of the discriminator is the real scene x and its corresponding condition c from the historical data set or the scene generated by the generator and its control condition c, whose output is a scalar p real Indicates the probability that the input scene is the real scene.

[0125]

[0126] Both the generator and the discriminator can be instantiated as neural networks with trainable parameters, such as multi-layer perceptrons. The scene generation corresponding to multiple scenes or nodes can be a convolutional neural network or a graph neural network.

[0127] The training goal of both networks is to achieve a balance where the generator generates very realistic scenes, making it difficult for the discriminator to distinguish between real scenes and generated scenes, thus forming a min-max game with a value function V(D,G) expressed as follows:

[0128]

[0129] 2) Conditional Deep Generative Model with Wasserstein Distance and Gradient Penalty

[0130] Wasserstein distance can well quantify the difference between the distribution of real scenes and generated scenes, thereby improving training stability and avoiding model collapse. Wasserstein distance is defined as:

[0131]

[0132] in, Represents the real scene distribution p data (x) and the generated scene distribution The set of all possible joint distributions γ of .

[0133] Since it is difficult to directly calculate the Wasserstein distance, an approximation is achieved by optimizing a function that satisfies 1-Lipschitz continuity fitted by a parameterized neural network D (also called a critic network).

[0134] In addition, a gradient penalty is adopted to ensure that D satisfies the Lipschitz continuity constraint.

[0135] Therefore, the loss function of the conditional adversarial generation network with the introduction of Wasserstein distance and gradient penalty can be expressed as:

[0136]

[0137] Among them, λ is the coefficient of gradient penalty; Represents the sampling scene The data distribution of is, and its sampling formula is:

[0138]

[0139] Here, ε is a random variable with uniform distribution in [0,1].

[0140] 3) Category weight allocation method

[0141] In the controllable generation of renewable scenarios, some control conditions correspond to relatively few historical scenarios, such as scenarios under extreme meteorological conditions or scenarios with drastic fluctuations, which easily lead to category imbalance.

[0142] The class weight allocation method aims to redistribute weights among historical scenes in the training set so that the training model pays more attention to scenes with rare extreme meteorological conditions and scenes with more drastic fluctuations.

[0143] The formula for assigning category weights is as follows:

[0144]

[0145] Among them, N c is the number of categories of the control condition vector; N k is the number of scenes belonging to the kth type of control condition; N is the total number of scenes in the historical data set; Σ k Describes the set of scenarios that belong to the kth category of control conditions. Category weight w i Will be used as the coefficient of the loss function.

[0146] (5) Training of conditional deep generative models and controllable generation of reproducible scenarios

[0147] The training process of the conditional deep generation model is as follows:

[0148] 1) Set the gradient penalty coefficient λ, the training batch size m, and the number of discriminator iterations n per generator iteration D The training parameters of the model are included; and the network parameters θ of the generator and discriminator of the model are initialized G and θ D ;

[0149] 2) Sampling a real data sample {x i ,c i}, sampling a random noise vector and the noise data sample of the encoded control condition vector from the noise distribution {z i ,c i}, and sample a random variable ε from a uniform distribution in [0,1];

[0150] 3) The generator generates renewable energy scenarios based on noise data samples The discriminator discriminates the scenes generated by the generator based on real data samples;

[0151] 4) Calculate the loss function based on the results of the generator and discriminator, and use the optimization algorithm to perform gradient updates on the network parameters of the generator and discriminator according to the loss function.

[0152] Repeat process 2)-4) until the generator network parameters converge and obtain a trained generator model.

[0153] The controllable generation of renewable scenarios is to input the category code of the control conditions into the trained generator, so as to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

[0154] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the instructions, these steps are not necessarily executed in sequence according to the order of the instructions. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above-mentioned embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0155] Corresponding to the aforementioned embodiment of the controllable generation method of renewable energy scenarios based on a deep generative model, adopting the same technical concept, the present application also provides an embodiment of a controllable generation device of renewable energy scenarios based on a deep generative model.

[0156] Figure 5 It is a structural schematic diagram of a controllable generation device for renewable energy scenarios based on a deep generative model, shown as an exemplary embodiment of the present application.

[0157] In an exemplary embodiment, Figure 5 As shown, the renewable energy scene controllable generation device based on the deep generation model includes:

[0158] The data acquisition module 1 is used to acquire the historical output data of renewable energy and the corresponding historical meteorological data, and perform data preprocessing.

[0159] The first model building module 2 is used to generate a mathematical model based on a renewable energy controllable scenario, wherein the control condition is a meteorological condition or a fluctuation degree of renewable energy output.

[0160] The rule establishment module 3 is used to select the meteorological factor with the highest correlation as the control condition by analyzing the correlation between the output of renewable energy and various meteorological factors in historical data, and to establish classification and coding rules for the control conditions according to whether the meteorological conditions are extreme or whether the output of renewable energy fluctuates violently.

[0161] The second model building module 4 is used to establish a conditional deep generation model suitable for the controllable generation of renewable energy scenarios, and to design a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output.

[0162] The model training module 5 is used to train the conditional deep generation model using the pre-processed historical data, and then input the category code of the control condition into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

[0163] For the specific limitations of the controllable generation device for renewable energy scenarios based on deep generative models, please refer to the limitations of the controllable generation method for renewable energy scenarios based on deep generative models mentioned above, which will not be repeated here.

[0164] Each module in the above-mentioned renewable energy scenario controllable generation device based on deep generative model can be fully or partially implemented by software, hardware and their combination.

[0165] The above modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above modules.

[0166] Figure 3 A wind power output scenario with violent fluctuation characteristics generated by a controllable generation method of renewable energy scenarios based on a deep generative model shown in an exemplary embodiment of the present application, Figure 4 It is a photovoltaic output scenario under extreme weather conditions generated by a controllable generation method of renewable energy scenarios based on a deep generation model shown in an exemplary embodiment of the present application.

[0167] In order to verify the effectiveness of the controllable generation method of renewable energy scenarios based on deep generative models, its performance was tested using real renewable energy output data and corresponding meteorological condition data.

[0168] like Figure 3 and Figure 4 As shown, considering one wind power station and one photovoltaic station, the time interval in a scene is 15 minutes, which includes 96 output periods.

[0169] According to the results of correlation analysis, the factor most correlated with wind power output is wind speed, and the factor most correlated with photovoltaic output is solar radiation.

[0170] The results of case analysis show that the controllable generation method of renewable energy scenarios based on deep generative models can effectively generate renewable energy scenarios with similar characteristics and patterns to real historical scenarios corresponding to given control conditions, especially for some rare extreme scenarios.

[0171] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A controllable generation method for renewable energy scenarios based on a deep generative model, characterized in that: include: Obtain historical output data of renewable energy and corresponding historical meteorological data, and perform data preprocessing; Generate mathematical models based on renewable energy controllable scenarios, where the control conditions are meteorological conditions or the degree of fluctuation of renewable energy output; By analyzing the correlation between renewable energy output and various meteorological factors in historical data, the meteorological factors with the highest correlation are selected as control conditions, and classification and coding rules for control conditions are established based on whether the meteorological conditions are extreme or whether the renewable energy output fluctuates violently; Establish a conditional deep generation model suitable for the controllable generation of renewable energy scenarios, and design a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output; The conditional deep generation model is trained using the preprocessed historical data, and the category code of the control condition is input into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

2. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 1, characterized in that: The obtaining of historical output data of renewable energy and corresponding historical meteorological data and performing data preprocessing further includes: Obtain historical data on wind power, photovoltaic and other renewable energy output corresponding to multiple renewable energy sites or power system nodes; Obtain historical meteorological data corresponding to the historical data of renewable energy output. The scale, time interval and length of meteorological data are consistent with the renewable energy output data. Meteorological factors include solar radiation, temperature, wind speed, relative humidity and rainfall, etc.; Historical data are preprocessed through outlier processing, missing value processing and data normalization methods.

3. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 1, characterized in that: The controllable scenario-based mathematical model for renewable energy generation, wherein the control condition is the meteorological condition or the fluctuation degree of renewable energy output, further includes: Mathematical modeling of renewable energy scenarios: x i ={x j,t |j∈N,t∈T} Among them, x i represents a renewable energy output scenario; x j,t represents the active output value of renewable energy of the station or power system node j in time period t; N and Τ represent the set of stations or power system nodes and output time periods in a scenario respectively; Establish the control condition vector c corresponding to the renewable energy output scenario i , which is mainly composed of meteorological conditions or the degree of fluctuation of renewable energy output; The goal of renewable energy controllable scenario generation is to generate scenarios with similar characteristics and patterns to historical scenarios corresponding to the control conditions under given control conditions, and the data distribution of the generated scenarios is close to that of the historical scenarios.

4. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 3, characterized in that: The above-mentioned method analyzes the correlation between the output of renewable energy and various meteorological factors in historical data, selects the meteorological factor with the highest correlation as the control condition, and establishes classification and coding rules for the control condition according to whether the meteorological condition is extreme or whether the output of renewable energy fluctuates violently, further comprising: According to the distribution of historical data of various meteorological factors, the upper and lower thresholds for determining extreme meteorological conditions are set. Meteorological conditions with meteorological factor values ​​exceeding the upper threshold or below the lower threshold are considered extreme meteorological conditions, and the corresponding renewable energy scenarios are considered extreme scenarios. Meteorological conditions and renewable energy scenarios corresponding to other meteorological factor values ​​are normal meteorological conditions and normal scenarios. According to the distribution of historical data on renewable energy output fluctuations, threshold ranges for determining severe, moderate and small fluctuation conditions are set, corresponding to severe fluctuation scenarios, moderate fluctuation scenarios and small fluctuation scenarios respectively; Extreme meteorological conditions and normal meteorological conditions as well as severe, moderate and small fluctuation conditions are classified and coded separately.

5. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 1, characterized in that: The method of establishing a conditional deep generation model suitable for controllable generation of renewable energy scenarios and designing a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output further includes: According to the mathematical model of controllable generation of renewable energy scenarios, the generator, discriminator, target value function and loss function of the conditional deep generative model based on the conditional adversarial generative network are constructed, and the Wasserstein distance and gradient penalty are introduced; According to the classification encoding of the control conditions, a category weight allocation method is designed to alleviate the category imbalance phenomenon, and it is introduced into the loss function to improve the learning efficiency of small sample categories.

6. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 5, characterized in that: The method of using the preprocessed historical data to train the conditional deep generation model, and then inputting the category code of the control condition into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding features and patterns, further includes: Set training parameters and initialize the network parameters of the model's generator and discriminator; Sampling a real data sample containing a renewable energy scenario and an encoded control condition vector from a historical data distribution, sampling a random noise vector and a noise data sample of the encoded control condition vector from a noise distribution, and sampling a random variable from a uniform distribution of [0,1]; The generator generates renewable energy scenarios based on noise data samples, and the discriminator discriminates the scenarios generated by the generator based on real data samples; The loss function is calculated based on the results of the generator and the discriminator, and the network parameters of the generator and the discriminator are gradient updated according to the loss function through the optimization algorithm until the generator network parameters converge.

7. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 2, characterized in that: The preprocessing of historical data by outlier processing, missing value processing and data normalization methods further includes: According to the 3σ criterion, outliers in historical data are identified and judged. When the sample distance average is greater than 3σ, the data sample is considered to be an outlier. For the identified outliers, they can be regarded as missing values ​​and processed according to the missing value method, or directly deleted. The missing values ​​in the sample are filled by the K nearest neighbor algorithm to ensure the integrity of historical data; The maximum and minimum normalization method is used to map the original historical data into the range of [0,1].

8. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 7, characterized in that: The normalization formula is as follows: Among them, v and v* correspond to the original and normalized historical data set vectors respectively, and vmax and vmin correspond to the maximum and minimum values ​​in the historical data set respectively.

9. The controllable generation method of renewable energy scenarios based on deep generative models according to claim 1, characterized in that: Also includes: The performance of the conditional deep generation model is tested using real renewable energy output data and corresponding meteorological condition data.

10. A controllable generation device for renewable energy scenarios based on a deep generative model, characterized in that: include: A data acquisition module is used to obtain historical output data of renewable energy and corresponding historical meteorological data, and perform data preprocessing; The first model building module is used to generate a mathematical model based on a renewable energy controllable scenario, wherein the control condition is a meteorological condition or a fluctuation degree of renewable energy output; The rule establishment module is used to analyze the correlation between renewable energy output and various meteorological factors in historical data, select the meteorological factor with the highest correlation as the control condition, and establish classification and coding rules for the control condition according to whether the meteorological condition is extreme or whether the renewable energy output fluctuates violently; The second model building module is used to establish a conditional deep generation model suitable for the controllable generation of renewable energy scenarios, and to design a category weight allocation method to improve the learning efficiency of renewable energy scenarios under extreme meteorological conditions with fewer samples or under drastic fluctuations in renewable energy output; The model training module is used to train the conditional deep generation model using the preprocessed historical data, and then input the category code of the control condition into the trained conditional deep generation model to controllably generate renewable energy scenarios with corresponding characteristics and patterns.

Citation Information

Patent Citations

  • Long-term scene generation method and system based on attention condition generative adversarial network

    CN117370825A