New energy extreme meteorological disturbance modeling-oriented scene generation method and system

Abstract

The invention discloses a scene generation method and system for new energy extreme meteorological disturbance modeling, and relates to the technical field of power systems, and the method comprises the steps: recognizing a strong disturbance event under meteorological drive according to a constructed new energy effective output sample set, and constructing a generative adversarial neural network model embedded in a wind power photovoltaic physical boundary condition, acquiring a virtual disturbance section; and based on the virtual disturbance section, constructing a new energy annual extreme disturbance scene set by integrating a plurality of typical days embedded with disturbance, and applying the new energy annual extreme disturbance scene set to new energy extreme meteorological disturbance modeling and sample enhancement. The method has strong representation capability, good training stability and scene expansion capability, and can provide more representative and diversified input scenes for dispatching optimization and supporting power supply planning of the power system, so that the robustness and risk perception capability of the system for coping with extreme events are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a scenario generation method and system for modeling extreme weather disturbances caused by new energy. Background Art

[0002] As renewable energy penetration continues to increase, the output uncertainty and volatility of renewable energy sources such as wind power and photovoltaics are becoming increasingly pronounced. This is particularly true during extreme weather events (such as severe temperature drops, strong winds, and sandstorms). Renewable energy output is prone to sudden and drastic drops or prolonged lows, posing a serious threat to the safety and stability of power grid operations. To improve the adaptability of dispatch models to extreme conditions, it is imperative to construct representative renewable energy output scenarios that accurately reflect the characteristics of extreme disturbances. Existing technologies often use probabilistic and statistical methods to generate renewable energy disturbance scenarios, such as Monte Carlo simulation, time series modeling, and Copula function modeling. Alternatively, they employ artificial perturbation rules to structurally deform and expand historical output series to simulate fluctuations. While these methods have some effectiveness in modeling conventional fluctuations, they often suffer from the following drawbacks when dealing with extreme disturbances: difficulty capturing non-Gaussianity, heavy tails, and complex time series structures; significant discrepancies between generated samples and actual anomaly samples; and unstable generation quality when samples are scarce. Furthermore, some methods ignore the physical feasibility constraints of the output series, potentially generating anomaly scenarios that do not conform to actual operating characteristics, impacting the accuracy of subsequent dispatch and analysis. Therefore, there is an urgent need to introduce a generation mechanism with strong nonlinear representation capabilities, good generalization performance for scarce samples, and the ability to learn the deep structural characteristics of extreme disturbances, to provide technical support for high-fidelity modeling of new energy output scenarios, thereby more effectively serving the scheduling optimization and risk assessment work under high-proportion new energy systems. Summary of the Invention

[0003] To address the shortcomings of existing renewable energy output scenario generation methods in responding to extreme weather disturbances, especially the inability to effectively capture the non-Gaussianity, heavy-tailedness, strong temporal dependence, and sample scarcity of extreme events, as well as the problems of low generation quality, poor generalization, and insufficient physical consistency of existing traditional methods based on probability statistics or regular perturbations when simulating sudden and persistent abnormal fluctuation scenarios, making it difficult to meet the higher requirements for robustness and security of power grid dispatch models under high-proportion renewable energy access, the present invention aims to provide a scenario generation technology for modeling extreme weather disturbances for renewable energy. Based on a generative adversarial neural network, it fully explores extreme disturbance patterns in historical output data and achieves high-fidelity synthetic modeling of abnormal output segments of wind power and photovoltaic power under extreme weather conditions through a data-driven approach. This method has strong representation capabilities, good training stability, and scenario expansion capabilities. It can provide more representative and diverse input scenarios for power system dispatch optimization and supporting power planning, thereby improving the system's robustness and risk perception capabilities in responding to extreme events.

[0004] To achieve the above technical objectives, the present application provides a scenario generation method for new energy extreme weather disturbance modeling, comprising the following steps:

[0005] Based on the constructed renewable energy effective output sample set, strong disturbance events driven by meteorological factors are identified, and virtual disturbance segments are obtained by constructing a generative adversarial neural network model embedded in the physical boundary conditions of wind power and photovoltaic power.

[0006] Based on virtual disturbance segments, a set of annual extreme disturbance scenarios for new energy is constructed by integrating multiple typical days with embedded disturbances, which is applied to the modeling and sample enhancement of extreme meteorological disturbances for new energy.

[0007] Preferably, in the process of constructing the new energy effective output sample set, the new energy effective output sample set is constructed based on the wind power and photovoltaic output data throughout the year through physical validity screening, time period elimination and outlier cleaning.

[0008] Preferably, in the process of identifying strong disturbance events, based on the new energy effective output sample set, a dual threshold criterion of combined power drop and continuous trough is adopted to dynamically extract short-term extreme disturbance fragments and identify strong disturbance events driven by meteorological factors.

[0009] Preferably, in the process of extracting short-term extreme disturbance segments, the short-term extreme disturbance segments are extracted by decoupling amplitude fluctuation and timing dependency features based on a dual-threshold criterion.

[0010] Preferably, in the process of constructing a generative adversarial neural network model, a generative adversarial network framework based on a fully connected layer is used to embed the physical constraint mechanism of wind power and photovoltaic output in the generator output layer, and through the adversarial training process of discriminator feedback, high-fidelity generation and physical consistency modeling of extreme meteorological disturbance sequences are achieved to construct a generative adversarial neural network model.

[0011] Preferably, when constructing a generative adversarial neural network model, the tanh activation function is used as the activation function of the last layer of the generator of the generative adversarial neural network model.

[0012] Preferably, when constructing a generative adversarial neural network model, a random noise vector with a dimension of 10 is input to the generator, which is gradually mapped to an output dimension consistent with the training sample through a multi-layer fully connected network, and an abnormal perturbation sequence with a length of 5 is output; the discriminator also adopts a multi-layer fully connected structure, and uses LeakyReLU nonlinear units in each layer to enhance the gradient propagation capability. The last layer uses a tanh activation function to compress the output to (-1,1), indicating the probability estimate that the input sample is a true sample.

[0013] Preferably, when constructing the new energy annual extreme disturbance scenario set, the virtual disturbance segment is embedded in the non-night time period of the actual new energy output curve to generate the new energy extreme weather disturbance scenario and construct the new energy annual extreme disturbance scenario set.

[0014] Preferably, when the virtual disturbance segment is embedded in the non-nighttime period of the actual renewable energy output curve, a smooth transition function is used to process the interpolation boundary.

[0015] The present invention also discloses a scenario generation system for new energy extreme weather disturbance modeling, comprising:

[0016] The data processing module is used to identify strong disturbance events driven by meteorological factors based on the constructed renewable energy effective output sample set. By constructing a generative adversarial neural network model embedded with the physical boundary conditions of wind power and photovoltaic power, the virtual disturbance segments are obtained.

[0017] The scenario generation and application module is used to construct a new energy annual extreme disturbance scenario set based on virtual disturbance segments by integrating multiple typical days with embedded disturbances, which is applied to the modeling and sample enhancement of new energy extreme meteorological disturbances.

[0018] The present invention discloses the following technical effects:

[0019] The present invention has a number of significant advantages and technological advances in the modeling of extreme meteorological disturbances in new energy. First, physical validity screening, photovoltaic nighttime shutdown elimination and outlier cleaning mechanisms are introduced in the data preprocessing stage to ensure that the training sample set truly reflects the characteristics of natural disturbances and avoids model learning deviations caused by equipment shutdowns or measurement anomalies. Secondly, a sliding window is used to combine the dual threshold criteria of power sag and continuous trough to identify extreme disturbance fragments, which can more accurately capture the strong disturbance behavior driven by meteorology than the traditional empirical method and improve the representativeness of the samples. In terms of model construction, by embedding the physical boundary constraints of wind and solar power in the generator output and combining it with discriminator feedback training, the physical consistency modeling of extreme disturbance samples is achieved, effectively avoiding the problem of traditional GAN ​​generating non-physical disturbances. At the same time, this method supports the batch generation of diverse disturbance fragments, and combined with the smooth interpolation function, it can naturally embed disturbances in the typical daily power curve, maintaining the continuity and rationality of the original time series structure. Finally, by constructing a set of typical output days throughout the year that includes disturbance fragments, a representative, temporally consistent and diverse annual abnormal disturbance scenario set is formed, providing systematic and structured scenario support for the dispatch optimization and flexibility analysis of new energy high-penetration power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 It is a schematic flow chart of the method described in the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0023] like Figure 1 As shown, the present invention discloses a scenario generation technology for new energy extreme weather disturbance modeling, including the following process:

[0024] Step S1: Based on the historical 8760 hours of wind power and photovoltaic output data throughout the year, a standardized time series dataset for extreme weather disturbance modeling is constructed through physical validity screening, time period elimination (photovoltaic nighttime shutdown period) and outlier cleaning.

[0025] Step S2: Using the sliding window combined with the dual criteria of power mutation threshold and continuous low power threshold, the extreme disturbance samples are dynamically extracted from the full-cycle data, and their amplitude fluctuation and timing dependency characteristics are decoupled.

[0026] Step S3: Based on the fully connected generative adversarial network (GAN) framework, wind and solar physical constraints are embedded in the generator output layer, and the physical consistency modeling of extreme meteorological disturbance sequences is achieved through the adversarial training mechanism of discriminator feedback.

[0027] Step S4: Use the trained generator to synthesize virtual disturbance segments with diversity and temporal coherence, and embed them into the non-nighttime period of the actual renewable energy output curve to generate scenarios for renewable energy extreme weather disturbance modeling and sample enhancement.

[0028] In step S1, based on the historical 8760 hours of wind power and photovoltaic output data throughout the year, a standardized time series dataset for extreme weather disturbance modeling is constructed through physical validity screening, time period elimination, and outlier cleaning, which includes the following steps:

[0029] Specifically, let the time series of new energy output be:

[0030] P={P1,P2,…,P T}, P t ∈[0,1];

[0031] Where, P t is the unit output per unit value (pu) at time t; T is the total time length.

[0032] Define a validity screening threshold δ, then the effective output meets the conditions:

[0033] P t >δ;

[0034] Where δ is an empirically set threshold to eliminate equipment outages or invalid data.

[0035] Construct the indicator function:

[0036]

[0037] By performing sliding statistics on the sequence, continuous valid segments can be identified and extracted The continuous time length L1 meets the minimum physical requirements and is used for the subsequent identification and extraction of extreme disturbance samples. In order to eliminate the extremely low output segments caused by non-weather reasons such as equipment failure and communication interruption, the continuity of the time points where the output is 0 or close to 0 is further verified. If it does not conform to the extreme weather change pattern, it is removed from the candidate data set to avoid misleading model learning. After the above processing, the cleaned new energy effective output sequence set P is obtained. eff , as the basic data set for subsequent training and sample construction.

[0038] In step S2, the sliding window is used to combine the dual criteria of power mutation threshold and continuous low power threshold to dynamically extract extreme disturbance samples from the full cycle data and decouple their amplitude fluctuation and timing dependency characteristics, including the following steps:

[0039] Define the sliding window length as L hours, and perform a sliding scan of length L on P. Let the starting position of the window be i, then the local output sequence corresponding to the window is:

[0040] P={P i ,P i+1 ,…,P i+L-1};

[0041] In each sliding window, determine whether the following two conditions are met. If both conditions are met, the window is determined to be an extreme disturbance sample:

[0042]

[0043] Here, starting from time i within the window, within one hour, the output suddenly drops by more than a fluctuation threshold Δ, and the average output remains below a mean threshold θ for at least L-1 hours. This criterion reflects the typical characteristics of sustained low wind and photovoltaic power generation during periods of extreme weather. The extracted samples constitute the training set, with a data dimension of N × L, where N is the number of extracted samples.

[0044] To improve the training stability of the neural network model, all input samples need to be normalized uniformly. Using linear interval normalization, the output value is scaled to [-1, 1] according to the minimum-maximum interval:

[0045]

[0046] Where, P max and P min are the minimum and maximum values ​​of all samples in the training set, P t norm is the output value after linear normalization.

[0047] This approach can keep the output of the tanh activation function consistent with the last layer of the generator, while avoiding the problem of gradient saturation of functions such as Sigmoid in low-value areas. Before using the GAN model for training, all extracted abnormal fragments must be reorganized into samples of uniform length to form a training set matrix:

[0048]

[0049] Where X∈R N×L is the training set matrix, R L is a matrix of dimension 1×L.

[0050] In step S3, the fully connected layer-based generative adversarial network framework embeds the physical constraint mechanism of wind power and photovoltaic output in the generator output layer, and achieves high-fidelity generation and physical consistency modeling of extreme weather disturbance sequences through an adversarial training process with discriminator feedback. The framework includes the following steps:

[0051] The input of the generator is a random noise vector of dimension 10, which is gradually mapped to an output dimension consistent with the training sample through a multi-layer fully connected network, and finally outputs an abnormal perturbation sequence of length 5. The training goal of the generator is to maximize the probability of the discriminator misclassifying its generated samples. Its loss function is defined as follows:

[0052]

[0053] Where G(z) is the generated sample, D(G(z)) is the probability of judging it as a true sample, and z~N(0,1) is the random Gaussian noise input.

[0054] The discriminator receives the input sample x∈R 5 , whose source may be real samples x~p data (x) or the sample x = G(z) generated by the generator. The discriminator also uses a multi-layer fully connected structure and uses LeakyReLU nonlinear units in each layer to enhance gradient propagation capabilities. The final layer uses a tanh activation function to compress the output to (-1, 1), indicating the probability estimate of the input sample being a real sample. The discriminator training goal is to distinguish between real samples and generated samples as much as possible. Its loss function is defined as follows:

[0055]

[0056] Where x is the real sample and G(z) is the generated sample.

[0057] In the theoretical optimal state, if the data distribution generated by the generator is p g (x) and the real data distribution p data (x) completely overlap, the discriminator cannot distinguish the two, and its optimal output should be as follows:

[0058]

[0059] At this point, the generator and the discriminator reach a Nash equilibrium state, and the training process converges.

[0060] In step S4, the trained generator is used to synthesize virtual disturbance segments with diversity and temporal coherence, and the segments are embedded into the non-nighttime period of the actual renewable energy output curve to generate a scenario of extreme weather disturbance for renewable energy, which includes the following steps:

[0061] The above noise z~N(0,1) is input into the trained generator to obtain a set of perturbation sequences:

[0062]

[0063] Where M is the number of generated samples, is the perturbation sample matrix.

[0064] Divide the annual output sequence into typical day sequences:

[0065] P d ={P 1,d ,P 2,d ,…P 24,d},d=1,…,365;

[0066] Defines the set of hours where interpolation is allowed PV excludes night time. Randomly select the interpolation starting point Satisfy t ins +L-1≤24.

[0067] Insert the perturbation sequence into the original output:

[0068]

[0069] The interpolation boundary can be processed using a smooth transition function:

[0070] P trans (t) = (1-α t )·P orig (t)+α t ·P gen (t),α t ∈[0,1];

[0071] Where, P trans (t) is the new energy output value after being processed by the smooth transition function; P orig (t) is the value of the new energy processing curve at time t of the original typical day; P gen (t) is the value of the generated perturbation fragment at time t; α tis the transition weight factor, which controls the fusion ratio of the original curve and the perturbation segment. It can usually be set to:

[0072]

[0073] Several virtual abnormal fragments are embedded into multiple typical daily sequences throughout the year to construct a set of new energy annual output scenarios with extreme disturbance characteristics:

[0074]

[0075] Where, P abn The set of annual abnormal disturbance output scenarios constructed; represents the power generation sequence of the i-th abnormal day, which contains the embedded disturbance fragment; k is the number of typical days with embedded disturbance, which does not exceed 365 days; i∈{1,2,…,k} is the typical day index.

[0076] The present invention provides a scenario generation technology for modeling extreme meteorological disturbances for new energy. First, based on a physical constraint screening mechanism, a new energy effective output sample set is constructed from the wind power and photovoltaic output data throughout the year by eliminating nighttime shutdowns, invalid or abnormal data. Second, a dual-threshold criterion of combined power sag and sustained trough is adopted to dynamically extract short-term extreme disturbance fragments and accurately identify strong disturbance events driven by meteorological conditions. Subsequently, a generative adversarial neural network model is constructed that embeds the physical boundary conditions of wind power and photovoltaics. Through the generator output physical constraint design and the discriminator adversarial training mechanism, the generation of high-fidelity abnormal disturbance samples is achieved. After the disturbance fragments are generated, a structural smoothing interpolation method is used to embed them into the actual new energy typical day output curve to maintain the continuity and rationality of the original time series structure. Finally, multiple typical days with embedded disturbances are integrated to construct an annual new energy extreme disturbance scenario set covering multiple time periods and multiple features, supporting the dispatch optimization and risk assessment model call of the power system. The present invention has strong characterization capabilities, good training stability and scenario expansion capabilities, and can provide more representative and diverse input scenarios for power system scheduling optimization and supporting power supply planning, thereby improving the system's robustness and risk perception capabilities in dealing with extreme events.

[0077] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0078] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0079] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A scenario generation method for modeling extreme weather disturbances caused by new energy, characterized in that: The following steps are involved: Based on the constructed renewable energy effective output sample set, strong disturbance events driven by meteorological factors are identified, and virtual disturbance segments are obtained by constructing a generative adversarial neural network model embedded in the physical boundary conditions of wind power and photovoltaic power. Based on the virtual disturbance segment, a set of new energy annual extreme disturbance scenarios is constructed by integrating multiple typical days with embedded disturbances, which is applied to the modeling and sample enhancement of new energy extreme meteorological disturbances.

2. The scenario generation method for modeling extreme weather disturbances caused by new energy according to claim 1, characterized in that: In the process of constructing the new energy effective output sample set, the new energy effective output sample set is constructed based on the annual wind power and photovoltaic output data through physical validity screening, time period elimination and outlier cleaning.

3. The scenario generation method for modeling extreme weather disturbances caused by new energy according to claim 2, characterized in that: In the process of identifying strong disturbance events, based on the new energy effective output sample set, a dual threshold criterion of combined power drop and continuous trough is adopted to dynamically extract short-term extreme disturbance fragments and identify strong disturbance events driven by meteorological factors.

4. The scenario generation method for modeling extreme weather disturbances caused by new energy according to claim 3, characterized in that: In the process of extracting short-term extreme disturbance segments, the short-term extreme disturbance segments are extracted by decoupling amplitude fluctuation and timing dependency features according to the dual-threshold criterion.

5. The scenario generation method for extreme weather disturbance modeling for new energy according to claim 4 is characterized in that: In the process of constructing a generative adversarial neural network model, a generative adversarial network framework based on a fully connected layer is used to embed the physical constraint mechanism of wind power and photovoltaic output in the generator output layer. Through the adversarial training process of discriminator feedback, high-fidelity generation and physical consistency modeling of extreme meteorological disturbance sequences are achieved to construct the generative adversarial neural network model.

6. The scenario generation method for modeling extreme weather disturbances caused by new energy according to claim 5, characterized in that: When constructing a generative adversarial neural network model, the tanh activation function is used as the activation function of the last layer of the generator of the generative adversarial neural network model.

7. The scenario generation method for extreme weather disturbance modeling for new energy according to claim 6, characterized in that: When constructing a generative adversarial neural network model, the input generator is a random noise vector with a dimension of 10, which is gradually mapped to an output dimension consistent with the training sample through a multi-layer fully connected network, and outputs an abnormal perturbation sequence with a length of 5; the discriminator also adopts a multi-layer fully connected structure, and uses LeakyReLU nonlinear units in each layer to enhance the gradient propagation capability. The last layer uses the tanh activation function to compress the output to (-1,1), indicating the probability estimate that the input sample is a true sample.

8. The scenario generation method for extreme weather disturbance modeling for new energy according to claim 7, characterized in that: When constructing the new energy annual extreme disturbance scenario set, the virtual disturbance segment is embedded in the non-night time period of the actual new energy output curve to generate a new energy extreme weather disturbance scenario, and construct the new energy annual extreme disturbance scenario set.

9. The scenario generation method for extreme weather disturbance modeling for new energy according to claim 8, characterized in that: When embedding the virtual disturbance segment into the non-nighttime period of the actual renewable energy output curve, a smooth transition function is used to process the interpolation boundary.

10. A scenario generation system for modeling extreme weather disturbances caused by new energy, characterized in that: include: The data processing module is used to identify strong disturbance events driven by meteorological factors based on the constructed renewable energy effective output sample set. By constructing a generative adversarial neural network model embedded with the physical boundary conditions of wind power and photovoltaic power, the virtual disturbance segments are obtained. The scenario generation and application module is used to construct a new energy annual extreme disturbance scenario set based on the virtual disturbance segment by integrating multiple typical days with embedded disturbances, which is applied to the new energy extreme meteorological disturbance modeling and sample enhancement.

Images