A wind turbine based variable pitch processing method, apparatus, device, medium and system

Abstract

The application provides a wind turbine-based variable pitch processing method, device, equipment, medium and system. The method comprises the following steps: obtaining wind speed data within a first preset time, and inputting the wind speed data into a neural network to obtain a one-dimensional hidden vector corresponding to the wind speed data; transforming the one-dimensional hidden vector into a two-dimensional hidden vector matrix according to the acquisition time of the wind speed data, and performing spatial pyramid convolution on the two-dimensional hidden vector matrix to obtain a first characteristic value of the wind speed data; obtaining a discriminant complex vector container corresponding to the wind turbine, and comparing the first characteristic value with a second characteristic value in the discriminant complex vector container; when it is determined that the distance between the first characteristic value and the second characteristic value is less than a preset distance threshold, sending a variable pitch instruction to a variable pitch system of the wind turbine to make the variable pitch system perform variable pitch processing. The problem that the wind power generator cannot be timely pitched in the prior art, resulting in severe fluctuation of the power system and shortening of the service life of the wind power generator is solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a pitch control method, apparatus, equipment, medium and system based on wind turbines. Background Technology

[0002] In wind power generation systems, the rotational speed of wind turbines is mainly determined by wind speed. When the current wind speed is higher than the rated wind speed, pitch technology is used. The pitch angle is adjusted by a proportional-integral (PI) controller to control the rotational speed of the wind turbine, so that the rotational speed of the wind turbine is as close as possible to the target rotational speed corresponding to the rated wind speed, thus avoiding damage to the wind turbine caused by excessive speed.

[0003] However, wind speed changes are instantaneous, abrupt, and uncertain. By the time the PI controller determines a deviation between the actual and target rotational speeds of the wind turbine and issues a pitch control command, the wind speed may have already changed multiple times, and the actual rotational speed of the wind turbine may have already experienced several mismatches with the current wind speed, resulting in a failure to adjust the pitch to the appropriate angle in time. When the current wind speed exceeds the rated wind speed, if the wind turbine fails to adjust the pitch to the appropriate angle in time, it will cause severe fluctuations in the power system and shorten the normal lifespan of the wind turbine. Summary of the Invention

[0004] This application provides a pitch control method, apparatus, equipment, medium, and system based on wind turbines, which solves the problem in the prior art that wind turbines cannot adjust pitch in time, resulting in severe fluctuations in the power system and shortened lifespan of wind turbines.

[0005] In a first aspect, this application provides a pitch processing method based on a wind turbine, comprising: acquiring wind speed data within a first preset time period and inputting the wind speed data into a neural network to obtain a one-dimensional latent vector corresponding to the wind speed data; transforming the one-dimensional latent vector into a two-dimensional latent vector matrix according to the acquisition time of the wind speed data, and performing spatial pyramid convolution on the two-dimensional latent vector matrix to obtain a first feature value of the wind speed data; acquiring a discriminative complex vector container corresponding to the wind turbine, and comparing the first feature value with a second feature value in the discriminative complex vector container; when it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, sending a pitch command to the pitch system of the wind turbine to enable the pitch system to perform pitch processing.

[0006] In one specific implementation, the step of transforming the one-dimensional latent vector into a two-dimensional latent vector matrix based on the acquisition time of the wind speed data includes: classifying the one-dimensional latent vector according to the acquisition time of the wind speed data to obtain a one-dimensional latent vector corresponding to daytime and a one-dimensional latent vector corresponding to nighttime; and combining the one-dimensional latent vector corresponding to daytime and the one-dimensional latent vector corresponding to nighttime to obtain a two-dimensional latent vector matrix.

[0007] In one specific implementation, performing spatial pyramid convolution on the two-dimensional latent vector matrix to obtain the first feature value of the wind speed data includes: obtaining multiple convolution kernel templates, and sequentially convolving the two-dimensional latent vector matrix with the multiple convolution kernel templates to obtain the first feature value of the wind speed data; wherein the size of the multiple convolution kernel templates decreases sequentially.

[0008] In one specific implementation, the step of obtaining the discriminative complex vector container corresponding to the wind turbine and comparing the first feature value with the second feature value in the discriminative complex vector container includes: obtaining multiple discriminative complex vector containers corresponding to the wind turbine, each of the multiple discriminative complex vector containers corresponding one-to-one with multiple pitch angles; comparing the first feature value with the second feature value in each of the discriminative complex vector containers; then, when it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, sending a pitch command to the pitch system of the wind turbine to cause the pitch system to perform pitch processing includes: for each discriminative complex vector container, when it is determined that the distance between the first feature value and the second feature value in the discriminative complex vector container is less than a preset distance threshold, sending a pitch command corresponding to the discriminative complex vector container to the pitch system of the wind turbine to cause the pitch system to perform pitch processing according to the pitch angle corresponding to the discriminative complex vector container.

[0009] In one specific implementation, the discriminative complex vector container is obtained as follows: historical wind speed data within a second preset time period is acquired, and the historical wind speed data is divided into historical data and future data according to the acquisition time of the historical wind speed data; feature values ​​of the historical data and feature values ​​of the future data are acquired respectively, and feature values ​​of excess data are selected from the feature values ​​of the future data; based on the feature values ​​of the historical data, the feature values ​​of the future data, and the feature values ​​of the excess data, a loss function is used to obtain the discriminative complex vector container corresponding to the historical wind speed data.

[0010] In one specific implementation, obtaining the feature values ​​of the historical data includes: inputting the historical data into a neural network to obtain a one-dimensional latent vector corresponding to the historical data; transforming the one-dimensional latent vector corresponding to the historical data into a two-dimensional latent vector matrix corresponding to the historical data according to the acquisition time of the historical data, and performing spatial pyramid convolution on the two-dimensional latent vector matrix corresponding to the historical data to obtain the feature values ​​of the historical data.

[0011] Secondly, this application provides a pitch processing device based on a wind turbine, comprising: an acquisition module, configured to acquire wind speed data within a first preset time period and input the wind speed data into a neural network to obtain a one-dimensional latent vector corresponding to the wind speed data; a processing module, configured to transform the one-dimensional latent vector into a two-dimensional latent vector matrix according to the acquisition time of the wind speed data, and perform spatial pyramid convolution on the two-dimensional latent vector matrix to obtain a first feature value of the wind speed data; the processing module is further configured to acquire a discriminative complex vector container corresponding to the wind turbine, and compare the first feature value with a second feature value in the discriminative complex vector container, and when it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, send a pitch command to the pitch system of the wind turbine to enable the pitch system to perform pitch processing.

[0012] Thirdly, this application provides an electronic device, including: a processor, a memory, and a communication interface; the memory is used to store executable instructions of the processor; wherein the processor is configured to execute the wind turbine-based pitch processing method described in the first aspect by executing the executable instructions.

[0013] Fourthly, this application provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the wind turbine-based pitch control method described in the first aspect.

[0014] Fifthly, this application provides a wind turbine-based pitch control system, comprising: a wind turbine-based pitch control device as described in any one of the second to fourth aspects, and a wind turbine; wherein the wind turbine includes a pitch control system.

[0015] This application provides a pitch control method, apparatus, device, medium, and system based on a wind turbine. The method includes: acquiring wind speed data within a first preset time period and inputting the wind speed data into a neural network to obtain a one-dimensional latent vector corresponding to the wind speed data; transforming the one-dimensional latent vector into a two-dimensional latent vector matrix according to the acquisition time of the wind speed data, and performing spatial pyramid convolution on the two-dimensional latent vector matrix to obtain a first feature value of the wind speed data; acquiring a discriminative complex vector container corresponding to the wind turbine, and comparing the first feature value with a second feature value in the discriminative complex vector container; when the distance between the first feature value and the second feature value is less than a preset distance threshold, sending a pitch control command to the pitch control system of the wind turbine to enable the pitch control system to perform pitch control. Compared to existing technologies that only issue pitch control commands when the actual speed of a wind turbine deviates from the target speed, the pitch control method based on wind turbines in this application inputs the acquired wind speed data into a neural network to obtain a one-dimensional latent vector. Then, based on the acquisition time, the one-dimensional latent vector is transformed into a two-dimensional latent vector matrix. A spatial pyramid convolution is then performed on the two-dimensional latent vector matrix to obtain significant feature values ​​of the wind speed data. These feature values ​​are compared with the feature values ​​in the discriminative complex vector container. When the distance between the feature values ​​is less than a distance threshold, it is determined that the wind speed is about to exceed the rated wind speed. Consequently, a pitch control command is promptly sent to the pitch control system for pitch control processing. This solves the problem in existing technologies where wind turbines cannot adjust pitch in time, leading to severe fluctuations in the power system and shortened lifespan of the wind turbines. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic flowchart of an embodiment of a pitch control method based on a wind turbine provided in this application;

[0018] Figure 2 A schematic flowchart of a second embodiment of a pitch control method based on a wind turbine provided in this application;

[0019] Figure 3 A schematic flowchart of a third embodiment of a pitch control method based on a wind turbine provided in this application;

[0020] Figure 4 A schematic flowchart of Embodiment 4 of a pitch control method based on a wind turbine provided in this application;

[0021] Figure 5A schematic diagram of the structure of an embodiment of a wind turbine-based pitch control device provided in this application;

[0022] Figure 6 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments made by those skilled in the art under the guidance of these embodiments are within the scope of protection of this application.

[0024] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] In wind power generation systems, the rotational speed of wind turbines is mainly determined by wind speed. When the current wind speed is higher than the rated wind speed, pitch technology is used. The pitch angle is adjusted by a proportional-integral (PI) controller to control the rotational speed of the wind turbine, so that the rotational speed of the wind turbine is as close as possible to the target rotational speed corresponding to the rated wind speed, thus avoiding damage to the wind turbine caused by excessive speed.

[0026] However, wind speed changes are instantaneous, abrupt, and uncertain. By the time the PI controller determines a deviation between the actual and target rotational speeds of the wind turbine and issues a pitch control command, the wind speed may have already changed multiple times, and the actual rotational speed of the wind turbine may have already experienced several mismatches with the current wind speed, resulting in a failure to adjust the pitch to the appropriate angle in time. When the current wind speed exceeds the rated wind speed, if the wind turbine fails to adjust the pitch to the appropriate angle in time, it will cause severe fluctuations in the power system and shorten the normal lifespan of the wind turbine.

[0027] Based on the above-mentioned technical problems, the technical conception process of this application is as follows: How to provide a pitch control method to solve the problem in the prior art that wind turbines cannot adjust pitch in time, resulting in severe fluctuations in the power system and shortened lifespan of wind turbines.

[0028] The technical solution of this application will now be described in detail through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0029] Figure 1 This is a schematic flowchart illustrating an embodiment of a wind turbine-based pitch control method provided in this application. See also... Figure 1 The pitch control method based on wind turbines specifically includes the following steps:

[0030] Step S101: Obtain wind speed data within a first preset time period, and input the wind speed data into a neural network to obtain the one-dimensional hidden vector corresponding to the wind speed data.

[0031] In this embodiment, wind speed data within a first preset time period can be obtained. This wind speed data is vector data composed of wind speed and wind direction.

[0032] The wind speed data is input into a neural network to obtain the corresponding one-dimensional hidden vector. For example, this neural network can be a Long Short-Term Memory (LSTM) network. Inputting the wind speed data into this neural network allows the acquisition of the corresponding one-dimensional hidden vector.

[0033] Step S102: Based on the acquisition time of the wind speed data, transform the one-dimensional latent vector into a two-dimensional latent vector matrix, and perform spatial pyramid convolution on the two-dimensional latent vector matrix to obtain the first feature value of the wind speed data.

[0034] In this embodiment, the one-dimensional latent vector can be transformed into a two-dimensional latent vector matrix based on the acquisition time of the wind speed data. For example, the one-dimensional latent vector can be classified into two groups of one-dimensional latent vectors according to the acquisition time of the wind speed data: one-dimensional latent vectors corresponding to daytime and one-dimensional latent vectors corresponding to nighttime, and the two groups of one-dimensional latent vectors can be combined to transform into a two-dimensional latent vector matrix.

[0035] Spatial pyramid convolution is performed on the two-dimensional latent vector matrix to obtain the first feature value of the wind speed data. Specifically, the two-dimensional latent vector matrix can be convolved with convolution kernel templates of successively smaller sizes to obtain the first feature value of the wind speed data. The successively smaller size of the convolution kernel templates reduces the resolution of the two-dimensional latent vector, thereby extracting more abstract and representative features.

[0036] Step S103: Obtain the discriminative complex vector container corresponding to the wind turbine, and compare the first feature value with the second feature value in the discriminative complex vector container. When it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, send a pitch command to the pitch system of the wind turbine so that the pitch system can perform pitch processing.

[0037] In this embodiment, the discriminative complex vector container corresponding to the wind turbine can be obtained. The discriminative complex vector container includes a second feature value used to determine that the wind speed is about to exceed the rated wind speed.

[0038] The first eigenvalue is compared with the second eigenvalue in the discriminative complex vector container. For example, this comparison can be performed using a dictionary lookup. If the distance between the first and second eigenvalues ​​is less than a preset distance threshold, it is determined that the wind speed is about to exceed the rated wind speed, and a pitch control command is sent to the wind turbine's pitch control system to initiate pitch control.

[0039] In this embodiment, wind speed data within a first preset time period is acquired and input into a neural network to obtain a one-dimensional latent vector corresponding to the wind speed data. Based on the acquisition time of the wind speed data, the one-dimensional latent vector is transformed into a two-dimensional latent vector matrix, and a spatial pyramid convolution is performed on the two-dimensional latent vector matrix to obtain a first feature value of the wind speed data. A discriminative complex vector container corresponding to the wind turbine is acquired, and the first feature value is compared with a second feature value in the discriminative complex vector container. When it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, a pitch command is sent to the pitch system of the wind turbine to enable the pitch system to perform pitch processing. Compared to existing technologies that only issue pitch control commands when the actual speed of a wind turbine deviates from the target speed, the pitch control method based on wind turbines in this application inputs the acquired wind speed data into a neural network to obtain a one-dimensional latent vector. Then, based on the acquisition time, the one-dimensional latent vector is transformed into a two-dimensional latent vector matrix. A spatial pyramid convolution is then performed on the two-dimensional latent vector matrix to obtain significant feature values ​​of the wind speed data. These feature values ​​are compared with the feature values ​​in the discriminative complex vector container. When the distance between the feature values ​​is less than a distance threshold, it is determined that the wind speed is about to exceed the rated wind speed. Consequently, a pitch control command is promptly sent to the pitch control system for pitch control processing. This solves the problem in existing technologies where wind turbines cannot adjust pitch in time, leading to severe fluctuations in the power system and shortened lifespan of the wind turbines.

[0040] Figure 2 This is a flowchart illustrating a second embodiment of a pitch control method based on a wind turbine provided in this application. Figure 1 Based on the illustrated embodiment, step S102 specifically includes the following steps:

[0041] Step S201: Based on the acquisition time of the wind speed data, classify the one-dimensional latent vector to obtain the one-dimensional latent vector corresponding to daytime and the one-dimensional latent vector corresponding to nighttime.

[0042] Step S202: Combine the one-dimensional latent vectors corresponding to daytime and nighttime to obtain a two-dimensional latent vector matrix.

[0043] Step S203: Obtain multiple convolution kernel templates, and sequentially convolve the two-dimensional latent vector matrix with the multiple convolution kernel templates to obtain the first feature value of the wind speed data.

[0044] Among them, the size of multiple convolution kernel templates decreases sequentially.

[0045] In this embodiment, the one-dimensional latent vector can be classified according to the acquisition time of the wind speed data. This reflects the interaction between weather and meteorological conditions in the same time period of each cycle, which is beneficial for exploring the periodic changes in wind speed. Specifically, based on whether the wind speed data is acquired during the day or night, the one-dimensional latent vector is divided into a one-dimensional latent vector corresponding to the daytime and a one-dimensional latent vector corresponding to the nighttime. For example, the one-dimensional latent vector can be "daytime 1, nighttime 1, daytime 2, nighttime 2, ...", and based on the acquisition time of the wind speed data, this one-dimensional latent vector is divided into a one-dimensional latent vector corresponding to the daytime "daytime 1, daytime 2, ..." and a one-dimensional latent vector corresponding to the nighttime "nighttime 1, nighttime 2, ...".

[0046] For example, the one-dimensional latent vectors corresponding to daytime and nighttime can be combined to form a two-dimensional latent vector matrix as shown below, which facilitates the aggregation of the same features from different periods:

[0047]

[0048] After forming a two-dimensional latent vector matrix, multiple convolution kernel templates can be obtained. The two-dimensional latent vector matrix is ​​then convolved with the multiple convolution kernel templates in sequence to obtain the first feature value of the wind speed data.

[0049] Specifically, the sizes of these multiple convolutional kernel templates decrease sequentially. For example, there can be three convolutional kernel templates, with sizes of 7*7, 5*5, and 3*3 respectively. By convolving the two-dimensional latent vector matrix with the successively smaller convolutional kernel templates, the resolution of the two-dimensional latent vectors decreases. On the one hand, this allows for the extraction of more abstract and representative features, and on the other hand, it effectively blurs and removes redundant information that is irrelevant to the period.

[0050] In this embodiment, the one-dimensional latent vectors are classified according to the acquisition time of the wind speed data. This effectively reflects the interaction between weather and meteorological conditions in the same time period across different cycles, which is beneficial for exploring the periodic changes in wind speed. Combining the classified one-dimensional latent vectors into a two-dimensional latent vector matrix aggregates the same features from different cycles. Convolving the two-dimensional latent vector matrix with multiple convolution kernel templates of successively decreasing size yields the first feature value of the wind speed data. This extracts more abstract and representative features and effectively blurs and removes redundant information irrelevant to the cycle. This provides the prerequisite for accurately predicting that the wind speed is about to exceed the rated wind speed, and thus promptly sending pitch control commands to the pitch control system for pitch control. Furthermore, it solves the problem in existing technologies where wind turbines cannot adjust pitch in time, leading to severe power system fluctuations and shortened wind turbine lifespan.

[0051] Figure 3 This is a flowchart illustrating a third embodiment of a pitch control method based on a wind turbine provided in this application. Figures 1 to 2 Based on the illustrated embodiment, see also Figure 3 The above step S103 specifically includes the following steps:

[0052] Step S301: Obtain multiple discriminative complex vector containers corresponding to the wind turbine, and each of the multiple discriminative complex vector containers corresponds one-to-one with multiple variable pitch angles.

[0053] In this embodiment, the discrimination complex vector container includes a second feature value used to determine that the wind speed is about to exceed the rated wind speed. There can be multiple discrimination complex vector containers corresponding to the wind turbine, each corresponding to a different excess target. Different excess targets correspond to different pitch angles.

[0054] For example, a wind speed exceeding the rated wind speed by 10% to 20% can be set as the first excess target, a wind speed exceeding the rated wind speed by 20% to 30% can be set as the second excess target, a wind speed exceeding the rated wind speed by 30% to 40% can be set as the third excess target, a wind speed exceeding the rated wind speed by 40% to 50% can be set as the fourth excess target, and a wind speed exceeding the rated wind speed by 50% can be set as the fifth excess target.

[0055] It can obtain multiple discriminative complex vector containers corresponding to the wind turbine, and each discriminative complex vector container corresponds to a different pitch angle.

[0056] Step S302: Compare the first eigenvalue with the second eigenvalue in each discriminative complex vector container.

[0057] Step S303: For each discriminative complex vector container, when it is determined that the distance between the first eigenvalue and the second eigenvalue in the discriminative complex vector container is less than a preset distance threshold, a pitch command corresponding to the discriminative complex vector container is sent to the pitch system of the wind turbine, so that the pitch system performs pitch processing according to the pitch angle corresponding to the discriminative complex vector container.

[0058] In this embodiment, the first feature value of the wind speed data is compared with the second feature value in each discriminative complex vector container. For example, the comparison between the first and second feature values ​​can be performed using a dictionary lookup.

[0059] For each discriminant complex vector container, when the distance between the first eigenvalue and the second eigenvalue in the discriminant complex vector container is determined to be less than a preset distance threshold, a pitch command corresponding to the discriminant complex vector container is sent to the wind turbine's pitch system. This pitch command may include the pitch angle corresponding to the discriminant complex vector container. The pitch system can then perform pitch adjustments based on the pitch angle corresponding to the discriminant complex vector container.

[0060] In this embodiment, there can be multiple discrimination complex vector containers, each corresponding to a different pitch angle. This allows the first feature value of the wind speed data to be compared with the second feature value in each discrimination complex vector container to determine the excess target met by the current wind speed. A pitch command corresponding to this excess target is then sent, causing the pitch system to adjust the pitch according to the corresponding pitch angle. This enables more targeted pitch adjustments, allowing for timely pitch adjustments when the wind speed is about to exceed the rated wind speed. This further solves the problem in existing technologies where wind turbines cannot adjust pitch in a timely manner, leading to severe power system fluctuations and shortened turbine lifespan.

[0061] Figure 4 This application provides a schematic flowchart of a fourth embodiment of a pitch control method based on a wind turbine. Figures 1 to 3 Based on the illustrated embodiment, see also Figure 4 The steps to determine how to obtain a complex vector container are as follows:

[0062] Step S401: Obtain historical wind speed data within a second preset time period, and divide the historical wind speed data into historical data and future data according to the acquisition time of the historical wind speed data. In this embodiment, historical wind speed data within a second preset time period can be obtained, and the historical wind speed data is vector data composed of wind speed and wind direction.

[0063] Based on the acquisition time of the historical wind speed data, the data can be divided into historical data and future data. For example, if the second preset time is 10 days, the historical wind speed data of the first 8 days can be used as historical data, and the historical wind speed data of the last 2 days can be used as future data.

[0064] Step S402: Obtain the feature values ​​of the historical data and the feature values ​​of the future data respectively, and select the feature values ​​of the excess data from the feature values ​​of the future data.

[0065] Step S403: Based on the feature values ​​of the historical data, the feature values ​​of the future data, and the feature values ​​of the excess data, use the loss function to obtain the discriminative complex vector container corresponding to the historical wind speed data.

[0066] In this embodiment, the feature values ​​of the historical data and the feature values ​​of the future data can be obtained respectively.

[0067] Specifically, the historical data can be input into a neural network to obtain the corresponding one-dimensional hidden vector. For example, the neural network can be an LSTM. Inputting historical data into this neural network allows the acquisition of the corresponding one-dimensional hidden vector.

[0068] Based on the acquisition time of the historical data, the one-dimensional latent vector corresponding to the historical data is transformed into a two-dimensional latent vector matrix corresponding to the historical data. For example, based on the acquisition time of the historical data, the one-dimensional latent vectors corresponding to the historical data can be classified into two groups of one-dimensional latent vectors: one-dimensional latent vectors corresponding to daytime and one-dimensional latent vectors corresponding to nighttime, and the two groups of one-dimensional latent vectors can be combined to transform into a two-dimensional latent vector matrix corresponding to the historical data.

[0069] Spatial pyramid convolution is performed on the two-dimensional latent vector matrix corresponding to the historical data to obtain its feature values. Specifically, the two-dimensional latent vector matrix can be convolved with convolution kernel templates of successively smaller sizes to obtain the feature values ​​of the historical data. The successively smaller size of the convolution kernel templates reduces the resolution of the two-dimensional latent vectors corresponding to the historical data, thereby extracting more abstract and representative features.

[0070] Specifically, the future data can be input into a neural network to obtain a one-dimensional latent vector corresponding to the future data; based on the acquisition time of the future data, the one-dimensional latent vector corresponding to the future data is transformed into a two-dimensional latent vector matrix corresponding to the future data, and a spatial pyramid convolution is performed on the two-dimensional latent vector matrix corresponding to the future data to obtain the feature values ​​of the future data.

[0071] After obtaining the feature values ​​of future data, feature values ​​for excess data can be selected from these feature values. For example, feature values ​​for excess data can be selected from the feature values ​​of future data based on an excess target. An excess target could be, for example, a wind speed exceeding the rated wind speed by 10% to 20%.

[0072] In this embodiment, based on the feature values ​​of historical data, the feature values ​​of future data, and the feature values ​​of excess data, the discriminative complex vector container corresponding to the historical wind speed data can be obtained using a loss function.

[0073] Specifically, the eigenvalues ​​of historical data form the discriminative complex vector container d. g The feature values ​​of future data form a discriminative complex vector container f. a The eigenvalues ​​of the excess data form a discriminative complex vector container f. g The loss function is as follows:

[0074]

[0075] Where the molecule is d g with f g The similarity measurement results, with d as the denominator. g with f a The similarity measure is given by 'all', where 'all' represents the feature values ​​of all future data and 'goal' represents the feature values ​​of all excess data. The closer the loss function is to 0, the closer the numerator and denominator are, resulting in the discriminative complex vector container 'd'. g The better they are at distinguishing between excess and non-excess targets.

[0076] In this embodiment, multiple discriminative complex vector containers can be formed according to different excess targets. For example, a wind speed exceeding the rated wind speed by 10% to 20% can be set as the first excess target, a wind speed exceeding the rated wind speed by 20% to 30% can be set as the second excess target, a wind speed exceeding the rated wind speed by 30% to 40% can be set as the third excess target, a wind speed exceeding the rated wind speed by 40% to 50% can be set as the fourth excess target, and a wind speed exceeding the rated wind speed by 50% can be set as the fifth excess target.

[0077] For example, based on different excess targets, the feature values ​​of historical data can be divided into shared features and unique features. Shared features are features that all excess targets possess, i.e., the saliency features that determine all excess targets, forming a shared feature container; unique features are the distinguishing features that differentiate between different excess targets, forming multiple unique feature sub-containers. Each unique feature sub-container and the shared feature container are then combined to form multiple discriminative complex vector containers.

[0078] In this embodiment, based on the feature values ​​of historical data, future data, and excess data, a loss function is used to obtain a discriminative complex vector container corresponding to the historical wind speed data. This effectively distinguishes excess targets from non-excess targets, providing a prerequisite for accurately predicting that the wind speed will soon exceed the rated wind speed and promptly sending pitch control commands to the pitch control system for pitch control processing. This further solves the problem in existing technologies where wind turbines cannot adjust pitch in time, leading to severe power system fluctuations and shortened wind turbine lifespan.

[0079] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0080] Figure 5 This application provides a schematic diagram of the structure of an embodiment of a wind turbine-based pitch control device; as shown below. Figure 5 As shown, the wind turbine-based pitch control device 50 includes an acquisition module 51 and a processing module 52. The acquisition module 51 acquires wind speed data within a first preset time period and inputs the wind speed data into a neural network to obtain a one-dimensional latent vector corresponding to the wind speed data. The processing module 52 transforms the one-dimensional latent vector into a two-dimensional latent vector matrix based on the acquisition time of the wind speed data, and performs spatial pyramid convolution on the two-dimensional latent vector matrix to obtain a first feature value of the wind speed data. The processing module 52 also acquires a discriminative complex vector container corresponding to the wind turbine, compares the first feature value with a second feature value in the discriminative complex vector container, and when the distance between the first feature value and the second feature value is less than a preset distance threshold, sends a pitch control command to the wind turbine's pitch control system to enable the pitch control system to perform pitch control processing.

[0081] The wind turbine-based pitch control device provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0082] In one possible implementation, the processing module 52 is specifically used to classify the one-dimensional latent vector according to the acquisition time of the wind speed data to obtain the one-dimensional latent vector corresponding to daytime and the one-dimensional latent vector corresponding to nighttime; and to combine the one-dimensional latent vector corresponding to daytime and the one-dimensional latent vector corresponding to nighttime to obtain a two-dimensional latent vector matrix.

[0083] In one possible implementation, the processing module 52 is specifically used to obtain multiple convolution kernel templates, and sequentially convolve the two-dimensional latent vector matrix with the multiple convolution kernel templates to obtain the first feature value of the wind speed data; wherein the size of the multiple convolution kernel templates decreases sequentially.

[0084] The wind turbine-based pitch control device provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0085] In one possible implementation, the processing module 52 is specifically used to obtain multiple discriminative complex vector containers corresponding to the wind turbine, each of which corresponds one-to-one with multiple pitch angles; compare the first feature value with the second feature value in each discriminative complex vector container; and for each discriminative complex vector container, when it is determined that the distance between the first feature value and the second feature value in the discriminative complex vector container is less than a preset distance threshold, send a pitch command corresponding to the discriminative complex vector container to the pitch system of the wind turbine, so that the pitch system performs pitch processing according to the pitch angle corresponding to the discriminative complex vector container.

[0086] The wind turbine-based pitch control device provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0087] In one possible implementation, the processing module 52 is further configured to acquire historical wind speed data within a second preset time period, and divide the historical wind speed data into historical data and future data according to the acquisition time of the historical wind speed data; acquire the feature values ​​of the historical data and the feature values ​​of the future data respectively, and select the feature values ​​of excess data from the feature values ​​of the future data; and, based on the feature values ​​of the historical data, the feature values ​​of the future data, and the feature values ​​of the excess data, use a loss function to obtain the discriminative complex vector container corresponding to the historical wind speed data.

[0088] In one possible implementation, the processing module 52 is specifically used to input the historical data into the neural network to obtain the one-dimensional latent vector corresponding to the historical data; according to the acquisition time of the historical data, the one-dimensional latent vector corresponding to the historical data is transformed into a two-dimensional latent vector matrix corresponding to the historical data, and spatial pyramid convolution is performed on the two-dimensional latent vector matrix corresponding to the historical data to obtain the feature values ​​of the historical data.

[0089] The wind turbine-based pitch control device provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0090] Figure 6 This is a schematic diagram of the structure of an electronic device provided in this application. Figure 6As shown, the electronic device 60 includes: a processor 61, a memory 62, and a communication interface 63; wherein, the memory 62 is used to store executable instructions of the processor 61; the processor 61 is configured to execute the technical solutions in any of the foregoing method embodiments by executing the executable instructions.

[0091] Optionally, the memory 62 can be either standalone or integrated with the processor 61.

[0092] Optionally, when the memory 62 is a device independent of the processor 61, the electronic device 60 may further include a bus 64 for connecting the aforementioned devices.

[0093] The electronic device is used to execute the technical solutions in any of the foregoing method embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0094] This application also provides a readable storage medium storing a computer program thereon, which, when executed by a processor, implements the technical solutions provided in any of the foregoing embodiments.

[0095] This application also provides a wind turbine-based pitch control system, including a wind turbine-based pitch control device and a wind turbine as provided in any of the foregoing embodiments; wherein the wind turbine includes a pitch control system.

[0096] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A pitch control method based on a wind turbine, characterized in that, include: Obtain wind speed data within a first preset time period, and input the wind speed data into a neural network to obtain a one-dimensional hidden vector corresponding to the wind speed data; Based on the acquisition time of the wind speed data, the one-dimensional latent vector is transformed into a two-dimensional latent vector matrix, and a spatial pyramid convolution is performed on the two-dimensional latent vector matrix to obtain the first feature value of the wind speed data. Obtain the discriminative complex vector container corresponding to the wind turbine, and compare the first feature value with the second feature value in the discriminative complex vector container. When it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, send a pitch command to the pitch system of the wind turbine so that the pitch system can perform pitch processing.

2. The pitch control method based on a wind turbine according to claim 1, characterized in that, The step of transforming the one-dimensional latent vector into a two-dimensional latent vector matrix based on the acquisition time of the wind speed data includes: Based on the acquisition time of the wind speed data, the one-dimensional latent vector is classified to obtain the one-dimensional latent vector corresponding to daytime and the one-dimensional latent vector corresponding to nighttime. The one-dimensional latent vector corresponding to daytime and the one-dimensional latent vector corresponding to nighttime are combined to obtain a two-dimensional latent vector matrix.

3. The pitch control method based on a wind turbine according to claim 1 or 2, characterized in that, The step of performing spatial pyramid convolution on the two-dimensional latent vector matrix to obtain the first feature value of the wind speed data includes: Multiple convolution kernel templates are obtained, and the two-dimensional hidden vector matrix is ​​convolved with the multiple convolution kernel templates in sequence to obtain the first feature value of the wind speed data; The sizes of the plurality of convolution kernel templates decrease sequentially.

4. The pitch control method based on a wind turbine according to claim 1 or 2, characterized in that, The step of obtaining the discriminative complex vector container corresponding to the wind turbine and comparing the first eigenvalue with the second eigenvalue in the discriminative complex vector container includes: Obtain multiple discriminative complex vector containers corresponding to the wind turbine, and each of the multiple discriminative complex vector containers corresponds one-to-one with multiple variable pitch angles; The first feature value is compared with the second feature value in each of the discriminative complex vector containers; The step of sending a pitch control command to the wind turbine's pitch control system when the distance between the first feature value and the second feature value is less than a preset distance threshold, so that the pitch control system performs pitch control processing, includes: For each of the discriminative complex vector containers, when it is determined that the distance between the first feature value and the second feature value in the discriminative complex vector container is less than a preset distance threshold, a pitch command corresponding to the discriminative complex vector container is sent to the pitch system of the wind turbine, so that the pitch system performs pitch processing according to the pitch angle corresponding to the discriminative complex vector container.

5. The pitch control method based on a wind turbine according to claim 1 or 2, characterized in that, The method for obtaining the discriminant complex vector container is as follows: Acquire historical wind speed data within a second preset time period, and divide the historical wind speed data into historical data and future data according to the acquisition time of the historical wind speed data; The feature values ​​of the historical data and the feature values ​​of the future data are obtained respectively, and the feature values ​​of the excess data are selected from the feature values ​​of the future data; Based on the feature values ​​of the historical data, the feature values ​​of the future data, and the feature values ​​of the excess data, a discriminative complex vector container corresponding to the historical wind speed data is obtained using a loss function.

6. The pitch control method based on a wind turbine according to claim 5, characterized in that, The process of obtaining the feature values ​​of the historical data includes: The historical data is input into a neural network to obtain a one-dimensional hidden vector corresponding to the historical data. Based on the acquisition time of the historical data, the one-dimensional latent vector corresponding to the historical data is transformed into a two-dimensional latent vector matrix corresponding to the historical data, and a spatial pyramid convolution is performed on the two-dimensional latent vector matrix corresponding to the historical data to obtain the feature values ​​of the historical data.

7. A pitch control device based on a wind turbine, characterized in that, include: The acquisition module is used to acquire wind speed data within a first preset time period and input the wind speed data into a neural network to obtain a one-dimensional hidden vector corresponding to the wind speed data. The processing module is used to transform the one-dimensional latent vector into a two-dimensional latent vector matrix according to the acquisition time of the wind speed data, and to perform spatial pyramid convolution on the two-dimensional latent vector matrix to obtain the first feature value of the wind speed data. The processing module is further configured to obtain the discriminative complex vector container corresponding to the wind turbine, compare the first feature value with the second feature value in the discriminative complex vector container, and when it is determined that the distance between the first feature value and the second feature value is less than a preset distance threshold, send a pitch command to the pitch system of the wind turbine so that the pitch system performs pitch processing.

8. An electronic device, characterized in that, include: Processor, memory, communication interface; The memory is used to store the executable instructions of the processor; The processor is configured to execute the wind turbine-based pitch processing method according to any one of claims 1 to 6 by executing the executable instructions.

9. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the wind turbine-based pitch processing method according to any one of claims 1 to 6.

10. A pitch control system based on a wind turbine, characterized in that, include: The pitch control device based on a wind turbine as described in any one of claims 7 to 9, and the wind turbine; wherein the wind turbine includes a pitch control system.

Images