Method, system, device and medium for generating wind power ramp-up scenarios under extreme weather conditions

Abstract

The present invention belongs to the technical field of scene generation. In order to solve the problem that existing wind power climbing scenes are difficult to adapt to extreme events, a method, system, equipment and medium for generating wind power climbing scenes under extreme weather are proposed. The spatiotemporal distribution of wind speed is obtained based on an extreme typhoon weather model, and an output scene is constructed in combination with the wind speed-power of the unit. A generative adversarial network is used to expand and enhance the scene, and then a climbing scene set is screened and constructed based on correlation theory and climbing evaluation indicators. The screening of typical climbing scenes is achieved based on a clustering algorithm. The proposed method meets the spatiotemporal correlation of the output of wind power clusters, can effectively represent the climbing output scene of wind power clusters under extreme weather, and is more in line with the operating status of wind power clusters.

Description

Technical Field

[0001] The present invention belongs to the technical field related to wind power cluster scene generation, and in particular to a method, system, equipment and medium for generating wind power ramp-up scenes under extreme weather conditions. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Wind power has strong intermittent, volatile, seasonal and random characteristics, which leads to increased uncertainty in the operation mode of the power grid and a larger fluctuation range. In particular, with the frequent occurrence of extreme weather events around the world, wind power fluctuates over a large range in a short period of time, that is, the potential threat of ramp events to the safe and stable operation of the power grid is becoming more and more prominent. In addition, wind power generation also has the characteristics of temporal and spatial correlation, which is mainly manifested in the output correlation of a single wind farm on a time scale and the output correlation of multiple wind farms on a spatial scale. This is because wind farms that are close to each other are in the same wind belt and the wind speeds are related to each other. The output characteristics of wind power make it difficult to construct a typical output scenario. Relevant studies have shown that for normal weather, the scenario analysis method characterized by scenario sampling and scenario reduction is currently one of the most commonly used and effective methods.

[0004] The scenario analysis method has high requirements for sample capacity. When faced with low-probability extreme events, due to the lack of sufficient samples, it is difficult to adapt to the low probability of extreme events, resulting in extremely low efficiency in generating climbing event scenarios and great difficulty in generating climbing scenarios.

[0005] Therefore, how to efficiently build a reliable wind power ramping scenario to improve the power system's ability to cope with large-scale wind power cluster ramping events under extreme weather conditions is a problem that needs to be solved at present. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method, system, equipment and medium for generating wind power climbing scenarios under extreme weather conditions. The proposed method meets the spatiotemporal correlation of the output of wind power clusters, can effectively represent the climbing output scenarios of wind power clusters under extreme weather conditions, and is more in line with the operating status of wind power clusters.

[0007] To achieve the above objectives, the first aspect of the present invention provides: a method for generating a wind power ramp-up scenario under extreme weather conditions, comprising:

[0008] Obtain extreme weather information and obtain the spatiotemporal distribution of wind speed based on the typhoon model;

[0009] Based on the spatiotemporal distribution of wind speed and combined with the wind speed-power characteristics of wind turbines, a wind power cluster output scenario under extreme weather conditions is constructed;

[0010] The trained generative adversarial network based on the attention mechanism is used to expand the constructed wind power cluster output scenarios under extreme weather conditions, and the output characteristics of each expanded wind power cluster output scenario are evaluated using the spatiotemporal correlation theory to obtain the wind power cluster output scenario that satisfies the spatiotemporal correlation.

[0011] Based on the ramp evaluation index, a clustering algorithm is used to perform cluster analysis on the wind power cluster output scenarios with temporal and spatial correlation, and the wind power cluster ramping scenarios under extreme weather conditions are obtained.

[0012] A second aspect of the present invention provides a system for generating wind power ramp-up scenarios under extreme weather conditions, comprising:

[0013] The acquisition module is used to obtain extreme weather information and obtain the spatiotemporal distribution of wind speed based on the typhoon model;

[0014] A construction module, for constructing a wind power cluster output scenario under extreme weather conditions based on the spatiotemporal distribution of wind speed and in combination with wind speed-power characteristics of wind turbines;

[0015] The screening module is used to expand the constructed wind power cluster output scenarios under extreme weather conditions using the trained generative adversarial network based on the attention mechanism, and to evaluate the output characteristics of each expanded wind power cluster output scenario using the spatiotemporal correlation theory to obtain the wind power cluster output scenario that satisfies the spatiotemporal correlation;

[0016] The clustering analysis module is used to perform cluster analysis on wind power cluster output scenarios with temporal and spatial correlation based on the ramp evaluation index and utilize clustering algorithms to obtain the wind power cluster ramp scenarios under extreme weather conditions.

[0017] The third aspect of the present invention provides a computer device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, a method for generating wind power ramping scenarios under extreme weather conditions is executed.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, a method for generating a wind power ramping scenario under extreme weather conditions is executed.

[0019] One or more of the above technical solutions have the following beneficial effects:

[0020] In the present invention, the spatiotemporal distribution of wind speed is obtained based on the extreme typhoon weather model, and the output scenario is constructed in combination with the wind speed-power of the unit. The generative adversarial network with self-attention mechanism is used to realize the expansion and enhancement of the scenario. Then, the climbing scenario set is screened and constructed based on the correlation theory and climbing evaluation index, and the screening of typical climbing scenarios is realized based on the clustering algorithm. The proposed method meets the spatiotemporal correlation of the output of wind power clusters, can effectively represent the climbing output scenario of wind power clusters under extreme weather, and is more in line with the operating status of wind power clusters.

[0021] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 This is a flow chart of a method for generating a wind power cluster ramp-up scenario under extreme weather conditions in Embodiment 1 of the present invention;

[0024] Figure 2 The Batts typhoon wind field model in the first embodiment of the present invention;

[0025] Figure 3 Sensor working principle diagram;

[0026] Figure 4 Sensor working principle block diagram;

[0027] Figure 5 The generative adversarial process in the first embodiment of the present invention, wherein (a) is the distribution result of the generator and the discrimination result of the discriminator before training; (b) is the distribution result of the generator and the discrimination result of the discriminator after the discriminator is trained; (c) is the distribution result of the generator and the discrimination result of the discriminator after the discriminator is trained and the generator is retrained; (d) is the distribution result of the generator and the discrimination result of the discriminator after the discriminator and the generator are trained multiple times;

[0028] Figure 6 This is a structural framework diagram of a generative adversarial network with a self-attention mechanism added in Embodiment 1 of the present invention;

[0029] Figure 7 Schematic diagram of power variation in a hill climbing event in Embodiment 1 of the present invention;

[0030] Figure 8 This is a clustering flow chart in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0032] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0033] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0034] Embodiment 1

[0035] This embodiment discloses a method for generating a wind power ramp-up scenario under extreme weather conditions, including:

[0036] Step 1: Taking a typhoon as an example, obtain typhoon meteorological information and obtain the spatiotemporal distribution of wind speed based on the typhoon model;

[0037] Step 2: Based on the spatiotemporal distribution of wind speed and combined with the wind speed-power characteristics of the wind turbines, a wind power cluster output scenario under extreme weather conditions is constructed;

[0038] Step 3: A generative adversarial network based on the self-attention mechanism is proposed to expand the generated wind power cluster output scenarios, and the output characteristics of each expanded wind power cluster output scenario are evaluated using the spatiotemporal correlation theory to obtain a wind power cluster output scenario that satisfies the spatiotemporal correlation;

[0039] Step 4: Based on the ramp evaluation index, a clustering algorithm is proposed to perform cluster analysis on the wind power cluster output scenarios with temporal and spatial correlation, and the wind power cluster ramping scenario under typhoon weather is obtained.

[0040] This embodiment obtains the spatiotemporal distribution of wind speed based on the typhoon model, and constructs the output scenario in combination with the wind speed-power of the unit. A generative adversarial network based on the self-attention mechanism is proposed to realize the expansion and enhancement of the scenario. Then, a climbing scenario set is screened and constructed based on the correlation theory and the climbing evaluation index. An adaptive DBSCAN clustering is proposed to realize the screening of typical climbing scenarios. The proposed method meets the spatiotemporal correlation of the output of the wind power cluster, can effectively represent the climbing output scenario of the wind power cluster under typhoon extreme weather, and is more in line with the operating status of the wind power cluster.

[0041] In step 1 of this embodiment, the wind farm cluster to which the scenario generation method is applied is determined, the moving path and moving speed information of the typhoon center are obtained according to the weather forecast, and the spatiotemporal distribution model of wind speed of the large-scale wind farm cluster is constructed based on the Batts typhoon model. The Batts typhoon model is constructed according to the basic theory of the relationship between the atmospheric pressure field and the wind speed, including the sea surface typhoon wind field model and the attenuation model considering the typhoon after landing, such as Figure 2 As shown, with high accuracy, the wind speed of the study point can be calculated based on the positional relationship between the typhoon center and the target point.

[0042] Specifically, after the typhoon landed, The central pressure difference at the moment is:

[0043] (1)

[0044] in, Indicates the change in central air pressure difference. represents the initial pressure difference, and They respectively represent the angle between the typhoon's moving direction and due north, and the angle between the landing coastline direction and due north.

[0045] Based on the change in central pressure difference, after the typhoon lands, The maximum wind speed radius at the moment is expressed as:

[0046] (2)

[0047] in, Indicates the change of central air pressure difference; Indicates the maximum wind speed radius in kilometers.

[0048] Based on the gradient wind equation ignoring friction, the maximum gradient wind speed is obtained, which is expressed as:

[0049] (3)

[0050] in, Indicates the maximum gradient wind speed; represents a constant coefficient, here it is 6.72; Indicates the change of central air pressure difference; represents the Coriolis coefficient; Represents the angular velocity of the earth's rotation, in rad / s; Indicates the latitude of the air particle.

[0051] Based on the maximum wind radius, the maximum average wind speed can be expressed as:

[0052] (4)

[0053] in, is the overall moving speed of the typhoon, Indicates the maximum gradient wind speed.

[0054] Based on the real-time meteorological information of typhoon, the spatiotemporal distribution of wind speed of large-scale wind power clusters shows dynamic characteristics. The wind speed at any point in the typhoon wind field is expressed as:

[0055] (5)

[0056] in, Indicates the wind speed at the midfield station in the wind farm cluster, Indicates the distance between the typhoon center and the wind farm station. Indicates the maximum wind speed radius of the typhoon, Indicates the maximum average wind speed of the typhoon, Set according to different typhoon types.

[0057] Considering the evolution process of typhoon weather, the wind speed time series curve of each unit is generated based on the spatiotemporal distribution model of wind speed in the wind power cluster.

[0058] In step 2 of this embodiment, based on the spatiotemporal distribution of wind speed and combined with the wind speed-power characteristics of the wind turbine generator set, a wind power cluster output scenario under typhoon weather is constructed, specifically:

[0059] Step 21: Considering the equipment differences of each wind turbine, based on the operating parameters and historical wind speed and output data, a piecewise linear function is used to fit the wind speed-power characteristics of each unit. There are certain differences in the characteristics between different units.

[0060] Taking three-segment linearization as an example, the wind speed-power characteristic of the unit can be expressed as:

[0061] (6)

[0062] in, Indicates the wind speed at the midfield station in the wind farm cluster, Indicates the output power of the unit. Indicates the rated wind speed of the unit. Indicates the rated power of the unit. Indicates the unit cut-in wind speed, Indicates the unit cut-out wind speed.

[0063] Step 22: Based on the wind speed time series curve and the wind speed-power characteristics of the unit, the wind power of each unit is obtained. Considering the difference in the angle between the blade surface of each unit and the wind direction, the wind power is corrected to obtain the output scenario of the entire wind power cluster.

[0064] The correction method of wind power is expressed as:

[0065] (7)

[0066] in, Indicates The wind power after correction of the unit, Indicates Theoretical wind power of the unit, Indicates wind speed and fan The angle of the plane where the blade lies is calculated as follows:

[0067] The plane direction of the wind turbine blades can be obtained from the wind turbine parameter data, and the wind direction at the location of the wind power cluster can be measured using a wind speed sensor based on piezoresistive detection.

[0068] The wind speed and direction sensor consists of four identical cantilever beam piezoresistive sensors, which are installed on a cube base platform in a centrally symmetrical manner. On the outside of the cantilever beam, piezoresistance is formed by doping or ion implantation. Figure 3 As shown, with the coordinates x and y axes as the standard, when the wind blows from any direction, assuming that the wind blows from the direction shown in the figure, it is in the positive direction of the x axis. horn( >0), blows over the surface of the structure and acts on the four cantilever beams. Then the four cantilever beams are deformed accordingly under the action of the wind. Figure 3 The four sensors are marked as 1, 2, 3, and 4. Obviously, the structural deformation of sensors 1 and 3 is that the upper end (i.e., the free end) is bent toward the x-axis. Similarly, sensors 2 and 4 are bent toward the y-axis. The horizontal wind speed can be considered as a two-dimensional vector, which can be decomposed into components in the x-axis and y-axis directions. The actual effect of the wind speed on the cantilever beam is the sum of the effects of the two components applied to the cantilever beam separately. Assume that the cantilever beam is as thin as possible and its thickness can be ignored. Then, the component of the wind speed acting on the front of the cantilever beam can be equivalent to the effect of the wind speed on the cantilever beam. The piezoresistors on the cantilever beams 1 and 3 placed opposite each other along the x-axis have the same bending direction, but one is subjected to compression and the other is subjected to tension. Theoretically, the absolute values ​​of the tensile stress and compressive stress generated by the piezoresistors on the two cantilever beams placed opposite each other are the same. Similarly, the same is true for the piezoresistors of the two sensors on the y-axis. By measuring the change in resistance, or connecting to a general measurement circuit, a voltage output is obtained by constant current drive. The output voltage V( , , , The relationship between the voltage output of the piezoresistors on cantilever beams 1, 2, 3, and 4 and the wind speed W and wind direction D can be calculated using equations (8) and (9):

[0069] (8)

[0070] (9)

[0071] Among them, k represents the correction factor.

[0072] The working principle block diagram of the device is as follows Figure 4As shown in the figure, when the wind passes through the sensor surface, the wind acts on the cantilever beam surface to generate wind pressure. The magnitude of the wind pressure on the cantilever beam structure in one-dimensional direction depends on the wind speed and wind direction. The cantilever beam produces corresponding deformation, and stress is generated inside the structure. The magnitude of the stress is determined by the deformation, the properties of the material, and the size of the structure. This stress and deformation causes the resistance value of the piezoresistance to change. By measuring the piezoresistance changes in two perpendicular directions and after certain calculations, the wind speed and wind direction can be obtained.

[0073] In step 3 of this embodiment, it is proposed to introduce a generative adversarial network with a self-attention mechanism to expand the generated wind power cluster output scenarios, and use the spatiotemporal correlation theory to evaluate the output characteristics of each expanded wind power cluster output scenario to obtain a wind power cluster output scenario that satisfies the spatiotemporal correlation.

[0074] Generative adversarial networks are based on simple input data sequences and generate samples that are indistinguishable from real samples through multiple learning training. They include two neural networks: the generator and the discriminator. Figure 5 As shown in the figure, the generator samples from the noise space to generate pseudo samples, whose distribution is represented by the solid line. The training purpose is to make the discriminator unable to identify the authenticity of the generated samples. The discriminator is used to distinguish whether the pseudo samples come from the real sample distribution. The output probability is represented by the denser dotted line, and the sparser dotted line represents the real sample distribution. It can be seen from the figure that Figure 5 In (a), the samples generated by the generator are quite different from the real samples. The results of the generator and discriminator are not accurate enough. In order to optimize the generator and discriminator, the discriminator is trained first. The results are as follows Figure 5 As shown in (b) in the figure, the probability value output by the discriminator is low when the sample is fake, and high when the sample is real. Retrain the generator, and the result is as follows Figure 5 As shown in (c) in the figure, the generator refers to the output value of the discriminator and generates samples where the discriminator output value is high, making the generated sample distribution closer to the real output distribution. In this way, the generator and the discriminator are trained alternately many times, and finally the samples generated by the generator can completely fit the real output distribution. The discriminator cannot distinguish between true and false, and the output value is 1 / 2 at any point in the data space, as shown in Figure 5 As shown in (d) in the figure, the generator and the discriminator are trained alternately. The training process is a minimax zero-sum game. The two sides compete with each other and are continuously optimized until a Nash equilibrium is reached. The generator can completely cover the original data distribution, making it impossible for the discriminator to distinguish between true and false, and the output value is 1 / 2 at any point in the data space.

[0075] The self-attention mechanism is introduced into the generative adversarial network. The self-attention mechanism module is inserted into the generative adversarial network architecture based on the original generative adversarial network. The self-attention mechanism module can help the model better understand the contextual information in the sequence, so as to process the sequence data more accurately.

[0076] Based on the wind power cluster output scenario, this embodiment proposes to introduce a generative adversarial network with a self-attention mechanism to expand and enhance the original data set of the wind power cluster output, and generate a wind power cluster output scenario set.

[0077] Specifically including: Step 31: Figure 6 As shown in the figure, the generator, self-attention mechanism and discriminator are combined to form a generative adversarial network, which is trained through error back propagation and gradient descent. An asynchronous training strategy is introduced during the training to obtain an expanded and enhanced set of wind power cluster output scenarios.

[0078] Specifically, the noise input is introduced into the generator for the original wind power cluster output scene set, and the output generates pseudo samples. The output of the generator is connected to the self-attention mechanism and the discriminator in turn. The pseudo scene and the real scene are processed by the self-attention mechanism before entering the discriminator, which can better capture the long-term dependency of the wind farm output sequence data, thereby improving the details and consistency of the generated data. , that is, the probability value of the sample being a real sample. The discriminant value output by the discriminator reacts to the generator and the discriminator, optimizing the realism of the pseudo samples generated by the generator and optimizing the discriminator's ability to distinguish between real samples and generated samples. The generator and the discriminator continue to compete and optimize until a balance is reached, that is, the discriminator cannot determine the correct source of the input sample, and the samples generated by the generator completely fit the distribution of real samples, and learn to get the distribution from the random noise sequence. Ability to transform to real data.

[0079] During the training process, the generator does not need to make complex transformations. Instead, it learns the distribution law of the training data based on the neural network training data, and outputs the generated data distribution that satisfies the specific distribution. The main goal of the generator is to make the generated samples as close to the real data distribution as possible to make the discriminator misjudge. In order to ensure the optimization direction in the early stage of training, the objective function can be expressed as:

[0080] (10)

[0081] in, represents the mathematical expectation, Indicates noise The probability distribution is obeyed, represents the objective function of the generator, represents the objective function of the discriminator.

[0082] In this embodiment, the discriminator can learn the ability to correctly distinguish true and false data through training. One part of its input is real sample data, and the other part is sample data generated by the generator; the output of the discriminator is a scalar from 0 to 1.

[0083] When the input sequence comes from real data, the output scalar is close to 1, and when the input sequence comes from generated samples, the output scalar is close to 0. Therefore, the objective function of training the discriminator can be expressed as maximizing the cross entropy, expressed as:

[0084] (11)

[0085] in, represents the mathematical expectation, Represents real data The probability distribution is obeyed, Indicates noise The probability distribution is obeyed, represents the objective function of the generator, represents the objective function of the discriminator.

[0086] The model training objective function of the generative adversarial network based on the self-attention mechanism is:

[0087] (12)

[0088] in, represents the objective function of model training, represents the mathematical expectation, Represents real data The probability distribution is obeyed, Indicates noise The probability distribution is obeyed, represents the objective function of the generator, represents the objective function of the discriminator.

[0089] The generative adversarial network based on the self-attention mechanism is trained through back propagation and gradient descent, without the need for Markov chain state transfer. During the training process, an asynchronous training strategy is also required. This is because the training of the discriminator is difficult and it is easy to cause overfitting on limited data. In order to prevent the learning ability of the generator and the discriminator from being too different, it is necessary to train The discriminator is trained once, and the generator is trained once again to optimize the learning abilities of the two simultaneously, thereby avoiding the situation where the discriminator cannot distinguish between the true and the false or can distinguish the true and the false completely and accurately due to the mismatch in their learning abilities, making the training lose the adversarial relationship and causing the model to collapse.

[0090] Among them, the self-attention mechanism is a mechanism used in deep learning models, which is particularly effective when processing sequence data. It allows each element of the input sequence to be compared with other elements in the sequence to calculate the representation of the sequence. This mechanism enables the model to focus on the relationship between different positions in the input sequence, thereby capturing complex dependencies within the sequence. The core idea of ​​the self-attention mechanism is that for each element in the sequence, the correlation between it and all other elements is calculated, and a weighted vector is generated accordingly. This weighted vector can be regarded as a representation of the element in the current context. In this way, the model can automatically learn the dependencies between different positions, thereby capturing richer contextual information.

[0091] In this embodiment, for the wind farm output sequence, wind speed and wind direction are the main factors that determine the wind farm output. The wind speed may change within a certain period of time, and this change is usually related to the weather system, seasonal climate change and day and night cycle. Therefore, the wind speed in a specific time period is often related to the wind speed in its adjacent time period. Therefore, when applying the self-attention mechanism to process the wind farm output daily scene sequence, unlike the traditional self-attention mechanism that captures the correlation between the elements of the entire sequence, only the output at each moment of the daily scene and the output data of the two adjacent time steps are selected as the input of the self-attention mechanism. The processing process of the self-attention mechanism includes:

[0092] (1) Calculate attention scores: Generate a query, key, and value for the input data, and use these vectors to calculate the importance of each output scenario to other scenarios.

[0093] (2) Calculate the attention weight: Use the calculated attention score to perform a weighted average on the value, so that the attention score is converted into a number between 0 and 1 and the sum is 1, and then the attention weight is obtained.

[0094] (3) Calculate output: Finally, multiply the value vector of each output scenario by its corresponding attention weight, and then sum them to get the final output. These weights will be used to combine the input output scenario vectors to generate a new context-related output scenario vector. This output scenario vector contains not only the information of the current output scenario, but also the information of its upper and lower relationships. In this way, the model can better understand the relationship between each output scenario in a specific time series.

[0095] Step 32: Input the set of wind power cluster output scenarios under typhoon weather obtained in step 2 into the trained generative adversarial network based on the self-attention mechanism to obtain the expanded wind power cluster output scenarios under extreme weather conditions.

[0096] Step 33: Use the spatiotemporal correlation theory to evaluate the wind power cluster output characteristics of each scenario, and select the wind power cluster output scenarios that meet the correlation requirements. The spatiotemporal correlation includes the temporal autocorrelation of the output of a single unit in the wind power cluster and the spatial correlation of the output of units between stations.

[0097] Specifically, the spatiotemporal correlation of wind power output in each scenario is evaluated based on the Spearman coefficient, and the correlation threshold is set to screen the scenarios by considering the temporal characteristics of the unit output and the spatial characteristics of the station cluster. The Spearman coefficient is a non-parametric indicator to measure the dependency relationship between two variables. When one variable is monotonically increasing and the other variable is also monotonically increasing, that is, the two variables are completely positively correlated, the Spearman coefficient is 1; when one variable is monotonically increasing and the other variable is monotonically decreasing, that is, the two variables are completely negatively correlated, the Spearman coefficient is -1; when there is no correlation between the two variables, the Spearman coefficient is close to 0.

[0098] The Spearman coefficient correlation is expressed as:

[0099] (13)

[0100] in, and Respectively represent the output power of different wind power clusters in each scenario, Represents the Spearman coefficient, the larger its value, the higher the correlation; express and The covariance between and Respectively and The standard deviation of and Respectively and The mean of represents mathematical expectation.

[0101] In step 4 of this embodiment, based on the ramp evaluation index, a clustering algorithm is used to perform cluster analysis on wind power cluster output scenarios with temporal and spatial correlation to obtain the wind power cluster ramp scenarios under extreme weather conditions.

[0102] Step 41: Construct wind power ramping indicators, identify large-scale wind power cluster power ramping events and screen the ramping scenario set.

[0103] For the ramping of wind power clusters, the up-ramp and down-ramp conditions of wind power are considered, an evaluation index is constructed based on the power difference on a certain time scale, the climbing index threshold is set considering the climbing regulation capability of the power system, and the scenarios where the climbing index exceeds the threshold are screened to construct a wind power cluster climbing scenario set.

[0104] like Figure 7 As shown, ramping refers to a large change in wind power cluster power in a short period of time, and the power change amplitude and rate exceed the regulation capacity of the power system control strategy. The identification and definition of ramping usually takes "event" as the main object of research and discussion, and analyzes the change in power during the duration of the ramping time. Upward climbing refers to a ramping event in which the power at the initial moment is less than the power at the end moment, and downward climbing refers to a ramping event in which the power at the initial moment is greater than the power at the end moment. During the ramping duration Inside, power from Change to The climbing amplitude is The climbing rate is , an upward slope occurs. During the slope duration Inside, power from Change to The climbing amplitude is The climbing rate is , a downward climb occurred.

[0105] Based on the power difference at a certain time scale, a climbing evaluation index is constructed, which is expressed as:

[0106] (14)

[0107] in, represents the power value at the end of the observation period; Indicates the power value at the beginning of the observation period; It represents a preset ramp threshold. As an optional solution, the ramp threshold can be set to 10% of the wind power cluster capacity.

[0108] Formula (14) is used to filter out the wind power cluster output scenarios where ramp events occur. The filtered ramp scenarios are then clustered to complete the generation of typical ramp scenarios for the wind power cluster.

[0109] Step 42: Design an adaptive DBSCAN clustering reduction ramp scenario set to obtain typical ramp scenarios of large-scale wind power clusters and their probability distribution information.

[0110] Step 421: First, a unified indicator is proposed to measure the distance of different output scenarios, expressed as a value:

[0111] (15)

[0112] in, For the Scene Hourly effort.

[0113] Adaptive DBSCAN is a density-based spatial clustering algorithm. , ) is used to describe the similarity of wind power cluster output scenario data in the neighborhood; among them, is the distance threshold describing the neighborhood of the hill climbing scene, It describes the climbing scene in which the distance to the point in the neighborhood is less than In addition, if the domain of a certain climbing scene contains at least the minimum number of If the neighborhood of a climbing scene is smaller than , but in the neighborhood of a core point, the scene is called a boundary point; if the neighborhood of a certain output scene is less than , and it is not a boundary point, then the scene is called a noise point.

[0114] The basic idea of ​​the adaptive DBSCAN clustering algorithm is: first set the threshold range of the number of clusters of the climbing scene, then determine the cluster category of each scene point by the density of the distribution of each climbing scene point in the climbing scene set, compare the number of cluster categories with the threshold, and adjust downward if the number of cluster categories is higher than the threshold range. If the number of cluster categories is lower than the threshold range, adjust upward , and then re-cluster until the number of cluster categories meets the threshold range.

[0115] The adaptive DBSCAN algorithm adjusts the parameters The size of the climbing scene is divided into different clusters with a reasonable number of clustering categories, such as Figure 8 As shown, the specific algorithm flow is as follows:

[0116] Step (1): Divide all climbing scenes into three categories: core points, boundary points, and noise points; all core points are marked as unvisited by default;

[0117] Step (2): Select an unvisited climbing scene P from all core objects and mark the climbing scene P as visited;

[0118] Step (3): Create a new cluster C for P, and add all the climbing scenes in the neighborhood of the climbing scene P to the generated candidate set N;

[0119] Step (4): Visit the climbing scene P' in the candidate set N. If the climbing scene P' is in the visited state, the climbing scene P' is classified as a boundary point into cluster C; if the climbing scene P' is in the unvisited state: first mark the climbing scene P' as visited; then determine whether the climbing scene P' is a core point. If the climbing scene P' is a core point, classify the climbing scene P' into cluster C and The climbing scene in the neighborhood is added to the candidate set N. If the climbing scene P' is a non-core point, it is classified into cluster C as a boundary point.

[0120] Step (5): Determine whether the candidate set N contains unvisited points. If the candidate set N still contains unvisited points, execute step (4); if all points in the candidate set N have been visited, execute step (6).

[0121] Step (6): Determine whether the scene set D contains unvisited core points. If D still contains unvisited core points, repeat steps (2) to (6); if all points in D have been visited, output the clustering results of the wind power cluster output scene set and the noise point set.

[0122] Step 422: Use the cluster center obtained in step 421 as the typical climbing scene, calculate the proportion of each scene cluster in the sample space as the probability distribution information of each typical climbing scene, and count the probability distribution of single machine climbing in each cluster.

[0123] A large number of screened ramp scenarios are clustered to complete the generation of typical ramp scenarios for wind power clusters.

[0124] The probability distribution information of a typical climbing scenario is visualized in the form of a bar graph. The probability distribution information of a typical climbing scenario can be expressed as:

[0125] (16)

[0126] in, Represents a typical climbing scenario The probability distribution of Represents a typical scenario The number of samples in the cluster, Indicates the number of cluster centers.

[0127] Step 423: Analyze the spatiotemporal correlation between the generated typical climbing scene and the test set, and calculate the Spearman coefficient to evaluate the model clustering effect.

[0128] In this embodiment, after DBSCAN clustering of the training sample set, a typical climbing scene is generated, and its spatiotemporal correlation with the test sample set is evaluated based on the Spearman coefficient, thereby verifying the generalization ability of the model and the representativeness of the typical climbing scene.

[0129] It should be noted that, in other embodiments, the values ​​of the above-mentioned specific parameters can be adjusted or changed according to specific prediction requirements.

[0130] Embodiment 2

[0131] The purpose of this embodiment is to provide a system for generating wind power ramping scenarios under extreme weather conditions, including:

[0132] The acquisition module is used to obtain typhoon meteorological information and obtain the spatiotemporal distribution of wind speed based on the typhoon model;

[0133] A construction module, for constructing a wind power cluster output scenario under extreme weather conditions based on the spatiotemporal distribution of wind speed and in combination with wind speed-power characteristics of wind turbines;

[0134] A screening module is used to expand the generated wind power cluster output scenarios using an improved generative adversarial network, and to evaluate the output characteristics of each expanded wind power cluster output scenario using the spatiotemporal correlation theory to obtain a wind power cluster output scenario that satisfies the spatiotemporal correlation;

[0135] The clustering analysis module is used to perform cluster analysis on wind power cluster output scenarios with temporal and spatial correlation based on the ramp evaluation index and utilize clustering algorithms to obtain the wind power cluster ramp scenarios under extreme weather conditions.

[0136] Embodiment 3

[0137] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0138] Embodiment 4

[0139] The purpose of this embodiment is to provide a computer-readable storage medium.

[0140] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0141] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0142] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0143] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for generating wind power ramp-up scenarios under extreme weather conditions, characterized in that: include: Obtain typhoon meteorological information and obtain the spatiotemporal distribution of wind speed based on the typhoon model; Based on the spatiotemporal distribution of wind speed and combined with the wind speed-power characteristics of wind turbines, a wind power cluster output scenario under extreme weather conditions is constructed; The trained generative adversarial network based on the attention mechanism is used to expand the constructed wind power cluster output scenarios under extreme weather conditions, and the output characteristics of each expanded wind power cluster output scenario are evaluated using the spatiotemporal correlation theory to obtain the wind power cluster output scenario that satisfies the spatiotemporal correlation. Based on the ramp evaluation index, clustering analysis is performed on the wind power cluster output scenarios with temporal and spatial correlation using clustering algorithms to obtain the wind power cluster ramp scenarios under extreme weather conditions. Based on the wind power cluster output scenario, a generative adversarial network with self-attention mechanism is proposed to expand and enhance the original data set of wind power cluster output and generate a wind power cluster output scenario set. Specifically, the generator, self-attention mechanism and discriminator are combined to form a generative adversarial network, which is trained through error back propagation and gradient descent, and an asynchronous training strategy is introduced during training to obtain an expanded and enhanced set of wind power cluster output scenarios; The obtained wind power cluster output scenario set under typhoon weather is input into the trained generative adversarial network based on the self-attention mechanism to obtain the expanded wind power cluster output scenario under extreme weather conditions; The spatiotemporal correlation theory is used to evaluate the wind power cluster output characteristics of each scenario, and the wind power cluster output scenarios that meet the correlation requirements are selected. The spatiotemporal correlation includes the temporal autocorrelation of the output of a single unit in the wind power cluster and the spatial correlation of the output of units between stations. Specifically, the spatiotemporal correlation of wind power output in each scenario is evaluated based on the Spearman coefficient, and the correlation threshold is set to screen the scenarios by considering the temporal characteristics of the unit output and the spatial characteristics of the station cluster.

2. The method for generating wind power ramp-up scenarios under extreme weather conditions as claimed in claim 1, characterized in that: Based on the spatiotemporal distribution of wind speed and the wind speed-power characteristics of wind turbines, a wind power cluster output scenario under extreme weather conditions is constructed, specifically: Based on the operating parameters, historical wind speed and output data of each wind turbine, the wind speed-power characteristics of each wind turbine are obtained by piecewise linear function fitting; Based on the spatiotemporal distribution of wind speed, a wind speed time series curve is generated for each wind turbine location; Based on the wind speed time series curve, the wind speed-power characteristics of each wind turbine are corrected in consideration of the different angles between each wind turbine and the wind direction, so as to obtain the output scenario of the wind power cluster.

3. The method for generating wind power ramp-up scenarios under extreme weather conditions as claimed in claim 1, characterized in that: The generated wind power cluster output scenario is expanded using the trained generative adversarial network based on the self-attention mechanism, specifically: Noise is introduced into the original wind power cluster output scenario set, and a pseudo scenario set is generated using a generator; The generated pseudo scene set is input into the self-attention mechanism module to capture the long-term dependency of wind farm output sequence data. Then the discriminator distinguishes the real original wind power cluster processing scene set and the pseudo scene set processed by the self-attention mechanism. The adversarial optimization training is performed through back propagation and gradient descent to obtain a trained generative adversarial network based on the attention mechanism. The generated wind power cluster output scenario is expanded using the trained generative adversarial network based on the self-attention mechanism.

4. The method for generating wind power ramp-up scenarios under extreme weather conditions as claimed in claim 1, characterized in that: Based on the ramp evaluation index, clustering analysis is performed on the wind power cluster output scenarios with temporal and spatial correlation using clustering algorithms, specifically: Step (1): Divide all wind power cluster output scenarios with temporal and spatial correlation into core points, boundary points and noise points; Step (2): Select any unvisited scene from all core points and mark it as a visited scene; Step (3): Create a new cluster for the marked access scenario and add all scenarios in the neighborhood of the marked access scenario to the candidate set; Step (4): Determine the access status of the scene in the candidate set, whether it is a core point, and whether the corresponding scene is classified into the new cluster; Step (5): Determine whether there are unvisited points in the candidate set. If there are unvisited points in the candidate set, execute step (4); if all points in the candidate set have been visited, execute step (6); Step (6): Determine whether there are unvisited core points in the scene set. If there are unvisited core points in the scene set, repeat steps (2) to (6) until all scenes are visited. If all points in the scene set have been visited, the clustering results of the wind power cluster output scene set and the noise point set are obtained.

5. The method for generating wind power ramp-up scenarios under extreme weather conditions as claimed in claim 1, characterized in that: Based on the moving path and speed of the typhoon center, the spatiotemporal distribution of wind speed is obtained based on the Batts typhoon model.

6. The method for generating wind power ramp-up scenarios under extreme weather conditions as claimed in claim 1 or 2, characterized in that: The wind speed-power characteristics of the wind turbine are corrected as follows: ， in, Indicates The wind power after correction of the unit, Indicates Theoretical wind power of the unit, Indicates wind speed and fan The angle of the plane in which the blade lies.

7. A system for generating wind power ramp-up scenarios under extreme weather conditions, based on a method for generating wind power ramp-up scenarios under extreme weather conditions as claimed in any one of claims 1 to 6, characterized in that: include: The acquisition module is used to obtain typhoon meteorological information and obtain the spatiotemporal distribution of wind speed based on the typhoon model; A construction module, for constructing a wind power cluster output scenario under extreme weather conditions based on the spatiotemporal distribution of wind speed and in combination with wind speed-power characteristics of wind turbines; The screening module is used to expand the constructed wind power cluster output scenarios under extreme weather conditions using the trained generative adversarial network based on the attention mechanism, and to evaluate the output characteristics of each expanded wind power cluster output scenario using the spatiotemporal correlation theory to obtain the wind power cluster output scenario that satisfies the spatiotemporal correlation; The clustering analysis module is used to perform cluster analysis on wind power cluster output scenarios with temporal and spatial correlation based on the ramp evaluation index and utilize clustering algorithms to obtain the wind power cluster ramp scenarios under extreme weather conditions.

8. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the method for generating a wind power ramping scenario under extreme weather conditions as described in any one of claims 1 to 6 is performed.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for generating a wind power ramping scenario under extreme weather conditions as described in any one of claims 1 to 6 is executed.

Images