Business operation method and equipment based on wind power and storage medium

Abstract

The invention discloses a business operation method and equipment based on wind power, and a storage medium. The method comprises the following steps: reconstructing a plurality of actual powers generated by a wind generating set in history into a phase space; if the last first phase point is selected as the central point, extending the central point into a plurality of second phase points according to the step length; searching a plurality of first phase points adjacent to the central point as central adjacent points; if the first second phase point is selected as an actual prediction point, searching a plurality of first phase points adjacent to the actual prediction point to obtain low-dimensional adjacent points; taking an intersection of the center neighbor point and the low-dimensional neighbor point to obtain a crossed neighbor point; fusing the plurality of cross neighbor points into a plurality of information points; inputting the multiple information points and the central point into a power prediction model to predict expected power generated by the wind generating set in a future duration range; and performing business operation on the wind generating set according to the expected power. According to the embodiment, the accuracy of the power prediction model in ultra-short-term wind power prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular, to a service operation method, device, and storage medium based on wind power. Background Art

[0002] In the process of energy structure transformation, wind power generation occupies an increasingly important position in the power system due to its clean and renewable characteristics. Because wind energy resources are highly random and uncertain, wind power prediction is carried out in the power system for business operations such as wind power consumption, stable operation of the power grid, and optimal allocation of energy.

[0003] Wind power prediction can generally be divided into ultra-short-term (from a few minutes to four hours), short-term (from several hours to three days), and medium- and long-term (from several days to several years).

[0004] Since the inherent chaotic dynamics characteristics of wind power are the main reasons for its external volatility, and this chaotic characteristic needs to be reflected in a high-dimensional space. Therefore, for ultra-short-term wind power prediction, the wind power can be reconstructed and mapped into a high-dimensional phase space. In the phase space, the points adjacent to the phase point at the next moment in the known set of phase points are used to predict the phase point at the next moment.

[0005] However, in some cases, the correlation between the adjacent points and the phase point at the next moment is weak, resulting in low accuracy of ultra-short-term wind power prediction and affecting the quality of business operations. Summary of the Invention

[0006] In view of this, the present invention provides a service operation method, device, and storage medium based on wind power to improve the accuracy of ultra-short-term wind power prediction.

[0007] The first aspect of the present invention provides a service operation method based on wind power, including:

[0008] Reconstructing a plurality of actual powers generated by a wind turbine generator set in history into a phase space according to a delay time and an embedding dimension; there are a plurality of first phase points in the phase space; the sampling time is spaced between two adjacent actual powers;

[0009] If the last first phase point is selected as the center point, extending the center point into a plurality of second phase points according to a step size; the product of the step size and the sampling time is a time range with a length less than the ultra-short-term time threshold;

[0010] Finding a plurality of the first phase points adjacent to the center point as center adjacent points;

[0011] If the first of the second phase points is selected as the actual prediction point, then under the condition of reducing the embedding dimension, find multiple first phase points that are neighbors of the actual prediction point to obtain low-dimensional neighboring points;

[0012] Take the intersection of the central neighboring points and the low-dimensional neighboring points to obtain cross neighboring points;

[0013] Fuse multiple cross neighboring points into multi-information points;

[0014] Input the multi-information points and the central point into a power prediction model to predict the expected power generated by the wind turbine within the future time range;

[0015] Perform an operation on the wind turbine according to the expected power.

[0016] The second aspect of the present invention provides an operation based on wind power, including:

[0017] A phase space reconstruction module for reconstructing multiple actual powers generated by the wind turbine in history into a phase space according to a delay time and an embedding dimension; there are multiple first phase points in the phase space; the sampling time is spaced between two adjacent actual powers;

[0018] A phase point extension module for extending the central point into multiple second phase points according to a step length if the last of the first phase points is selected as the central point; the product of the step length and the sampling time is a time range with a length less than the ultra-short-term time threshold;

[0019] A central neighboring point search module for finding multiple first phase points that are neighbors of the central point as central neighboring points;

[0020] A low-dimensional neighboring point search module for finding multiple first phase points that are neighbors of the actual prediction point under the condition of reducing the embedding dimension to obtain low-dimensional neighboring points if the first of the second phase points is selected as the actual prediction point;

[0021] A neighboring point intersection module for taking the intersection of the central neighboring points and the low-dimensional neighboring points to obtain cross neighboring points;

[0022] A cross neighboring point fusion module for fusing multiple cross neighboring points into multi-information points;

[0023] An expected power prediction module for inputting the multi-information points and the central point into a power prediction model to predict the expected power generated by the wind turbine within the future time range;

[0024] An operation execution module for performing an operation on the wind turbine according to the expected power.

[0025] A third aspect of the present invention provides an electronic device, the electronic device comprising:

[0026] at least one processor; and

[0027] a memory communicatively connected to the at least one processor; wherein,

[0028] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the business operation method based on wind power as described in the first aspect above.

[0029] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the business operation method based on wind power as described in the first aspect above.

[0030] A fifth aspect of the present invention provides a computer program product comprising a computer program, and when the computer program is executed by a processor, it implements the business operation method based on wind power as described in the first aspect above.

[0031] In this embodiment, a plurality of actual powers generated by a wind turbine in history are reconstructed into a phase space according to a delay time and an embedding dimension; there are a plurality of first phase points in the phase space; the sampling time is spaced between two adjacent actual powers; if the last first phase point is selected as the center point, the center point is extended into a plurality of second phase points according to a step size; the product of the step size and the sampling time is a time range with a length less than the ultra-short-term time threshold; a plurality of first phase points adjacent to the center point are found as the center neighboring points; if the first second phase point is selected as the actual prediction point, under the condition of reducing the embedding dimension, a plurality of first phase points adjacent to the actual prediction point are found to obtain low-dimensional neighboring points; the intersection of the center neighboring points and the low-dimensional neighboring points is taken to obtain the cross neighboring points; the plurality of cross neighboring points are fused into multi-information points; the multi-information points and the center point are input into a power prediction model to predict the expected power generated by the wind turbine in a future time range; the wind turbine is operated according to the power generated by the wind turbine in the future time range. In this embodiment, the low-dimensional neighboring points are used to further screen the center neighboring points, ensuring that the selected points can be more relevant to the prediction target, improving the representativeness of the input of the power prediction model, thereby improving the accuracy of the power prediction model in ultra-short-term prediction of wind power and ensuring the quality of the business operation.

[0032] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0034] Figure 1 is a flowchart of a service operation method based on wind power generation power provided in Embodiment 1 of the present invention.

[0035] Figure 2 is a schematic structural diagram of a power prediction model provided in Embodiment 1 of the present invention.

[0036] Figure 3 is a schematic structural diagram of a service operation device based on wind power generation power provided in Embodiment 2 of the present invention.

[0037] Figure 4 is a schematic structural diagram of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0039] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can cover arrangements other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0040] Example 1

[0041] See Figure 1 , which shows a flowchart of a service operation method provided in Example 1 of the present invention. This method can be executed by a service operation device based on wind power. The service operation device based on wind power can be implemented in the form of hardware and / or software, and the service operation device based on wind power can be configured in an electronic device. As Figure 1 shown, the method includes:

[0042] Step 101: Reconstruct the multiple actual powers generated by the wind turbine in history into a phase space according to the delay time and the embedding dimension.

[0043] In a wind power plant such as an offshore wind farm or an onshore wind farm, multiple wind turbines can be configured. A wind turbine is a device that converts the kinetic energy of the wind into mechanical kinetic energy and then converts the mechanical energy into electrical kinetic energy. It is usually divided into parts such as a wind wheel, a generator, and a tower barrel.

[0044] Sample the actual power of the wind turbine at N moments at a fixed sampling time c within a historical continuous period to obtain a wind power time series x(t) = {x1, x2, …, x i , …, x N}, where x i is the actual power at moment i, i ∈ [1, N], N is the length of the wind power time series, and the interval between two adjacent actual powers is the sampling time.

[0045] In a specific implementation, the mutual information method can be used to select the delay time τ, and the saturated correlation dimension method can be used to select the embedding dimension m. Reconstruct the multiple actual powers generated by the wind turbine in history into a phase space according to the delay time τ and the embedding dimension m to obtain a phase space X m .

[0046] Among them, there are multiple first phase points in the phase space, and the dimension number of each first phase point is the embedding dimension m. The coordinate values between two adjacent dimensions in the same first phase point are separated by the delay time τ.

[0047] Then, the phase space X m can be expressed as:

[0048] ;

[0049] Among them, X(i) is the i-th first phase point, i ∈ [1, n], n is the number of first phase points, x i+(j-1)τ is the coordinate value under the j-th dimension in X(i), j ∈ [1, m], m is the embedding dimension, and τ is the delay time.

[0050] Generally, n satisfies the condition: n = N - (m - 1)τ. Then, the coordinate value x of the last dimension of X(n) n+(m-1)τ is the actual power of the Nth sample (i.e., the last sample point) in the wind power time series x(t) = {x1, x2, …, x i , …, x N}.

[0051] Since the coordinate values of all the first phase points in the phase space X m are known, the phase space X m can be called a known set of m-dimensional phase space points.

[0052] Step 102: If the last first phase point is selected as the center point, extend the center point into multiple second phase points according to the step size.

[0053] Among the n first phase points of the known set of m-dimensional phase space points X m , the last first phase point X(n) can be selected as the center point.

[0054] At this time, the center point X(n) can be extended into multiple second phase points according to the step size s of wind power prediction. The second phase points construct a new m-dimensional phase space, called the predicted target phase point set X m ´. The predicted target phase point set X m ´ can be regarded as an extension of the known set of m-dimensional phase space points X m .

[0055] Among them, the product (s × c) of the step size s and the sampling time c is a time range with a length less than the ultra-short-term time threshold (such as 4 hours), which belongs to ultra-short-term wind power prediction.

[0056] In specific implementation, variables can be determined. Among them, the value range of the variable is a positive integer s between 1 and the step size, that is, [1, s].

[0057] Extend the coordinate values of each dimension in the center point according to multiple variables respectively to obtain the coordinate values of each dimension in multiple second phase points. Among them, the coordinate value of the last dimension in the second phase point corresponds to the power (i.e., the sampling point) with a quantity of step size s after multiple actual powers (i.e., N actual powers) of the wind turbine generator.

[0058] Then, the predicted target phase point set X m ´ can be expressed as:

[0059] ;

[0060] where X(n + i) is the ith second phase point, i ∈ [1, s], s is the step size, and x n+i+(j-1)τis the coordinate value of the j-th dimension in X(n + i), where j ∈ [1, m], m is the embedding dimension, and τ is the delay time.

[0061] Step 103: Find multiple first phase points that are close to the center point as the central neighboring points.

[0062] In the known m-dimensional phase space point set X m except for the center point, other first phase points can be traversed to find multiple other first phase points that are close to the center point X(n), denoted as central neighboring points.

[0063] In a specific implementation, in the known m-dimensional phase space point set X m for other first phase points except the center point, calculate the first distance between the center point X(n) and other first phase points.

[0064] Taking the Euclidean distance as an example of the first distance, the first distance between the center point X(n) and other first phase points can be expressed as:

[0065] ;

[0066] where, x i+(j-1)τ is the coordinate value of the j-th dimension in the i-th first phase point X(i), x n+(j-1)τ is the coordinate value of the j-th dimension in the n-th first phase point X(n) (i.e., the center point), d m is the first distance between the center point X(n) and the i-th first phase point X(i), i ∈ [1, n - 1], n is the number of first phase points, j ∈ [1, m], m is the embedding dimension, and τ is the delay time.

[0067] Compare the first distances of each other first phase point, and select multiple (assume k) other first phase points with the smallest first distance as the central neighboring points P center .

[0068] Step 104: If the first second phase point is selected as the actual prediction point, under the condition of reducing the embedding dimension, find multiple first phase points that are close to the actual prediction point to obtain low-dimensional neighboring points.

[0069] Among the s second phase points of the predicted target phase point set X m ´, the first second phase point X(n + 1) can be selected as the actual prediction point.

[0070] Under the condition that both the actual prediction point X(n + 1) and the first phase point X(i) reduce the embedding dimension (i.e., dimension reduction), find multiple first phase points X(i) that are close to the actual prediction point X(n + 1) to obtain low-dimensional neighboring points P low .

[0071] In a specific implementation, when the delay time τ is less than the sum of the number N of actual power and 1 (i.e., τ < N + 1), the actual prediction point X(n + 1) is reduced by one dimension to obtain the low-dimensional prediction target point X(n + 1). m-1 , the low-dimensional prediction target point X(n + 1) m-1 can be expressed as: X(n + 1) m-1 = [x n+1 , x n+1+τ , …, x n+1+(j-1)τ , …, x n+1+(m-2)τ , where x n+1+(j-1)τ is the coordinate value in the j-th dimension of X(n + 1), j ∈ [1, m - 1], m is the embedding dimension, and τ is the delay time.

[0072] The coordinate value x m-1 of the last dimension of the low-dimensional prediction target point X(n + 1) n+1+(m-2)τ is the (n + 1 + (m - 2)τ)-th element of the wind power time series x(t). According to n = N - (m - 1)τ, that is, the coordinate value x m-1 of the last dimension of the low-dimensional prediction target point X(n + 1) n+1+(m-2)τ is the (N + 1 - τ)-th element of the wind power time series x(t). When τ < N + 1, N + 1 - τ > 0, making the low-dimensional prediction target point X(n + 1) m-1 known. The first m - 1 dimensions of the low-dimensional prediction target point X(n + 1) m-1 are equal to the first m - 1 dimensions of the actual prediction point X(n + 1), and the low-dimensional prediction target point X(n + 1) m-1 is more relevant to the prediction target.

[0073] In addition, for the known m-dimensional phase space point set X m , low-dimensional projection is performed to remove the coordinate value of the last dimension in each first phase point X(i) to obtain low-dimensional phase points. The low-dimensional phase points can form a new phase space, denoted as the known (m - 1)-dimensional phase space point set X m-1 .

[0074] Then, the known (m - 1)-dimensional phase space point set X m-1 can be expressed as:

[0075] ;

[0076] where X(i) m-1 is the i-th low-dimensional phase point, i ∈ [1, n], n is the number of low-dimensional phase points, and x i+(j-1)τ is the coordinate value in the j-th dimension of X(i) m-1 , j ∈ [1, m - 1], m is the embedding dimension, and τ is the delay time.

[0077] In the known m - 1 - dimensional phase - space point set X m-1 calculate the low - dimensional predicted target point X(n + 1) m-1 and the second distance between it and other low - dimensional phase points X(i) m-1

[0078] Taking the Euclidean distance as an example of the second distance, the second distance between the low - dimensional predicted target point X(n + 1) m-1 and other low - dimensional phase points X(i) m-1 can be expressed as:

[0079] ;

[0080] where x i+(j-1)τ is the coordinate value of the j - th dimension in the i - th low - dimensional phase point X(i) m-1 x n+1+(j-1)τ is the coordinate value of the j - th dimension in the low - dimensional predicted target point X(n + 1) m-1 d m-1 is the second distance between the low - dimensional predicted target point X(n + 1) m-1 and the i - th low - dimensional phase point X(i) m-1 i ∈ [1, n], n is the number of low - dimensional phase points, j ∈ [1, m - 1], m is the embedding dimension, and τ is the delay time.

[0081] Compare the second distances of each low - dimensional phase point, and select the first - phase points corresponding to multiple (assume k) low - dimensional phase points with the smallest second distance as the low - dimensional nearest - neighbor points P m-1 nearest to the low - dimensional predicted target point X(n + 1) low .

[0082] For example, if the i - th low - dimensional phase point X(i) m-1 is one of the k low - dimensional phase points with the smallest second distance, and the i - th first - phase point X(i) is the point corresponding to the i - th low - dimensional phase point X(i) m-1 then the i - th first - phase point X(i) is the low - dimensional nearest - neighbor point P m-1 nearest to the low - dimensional predicted target point X(n + 1) low .

[0083] Step 105: Take the intersection of the central nearest - neighbor points and the low - dimensional nearest - neighbor points to obtain the cross - nearest - neighbor points.

[0084] In this embodiment, the intersection of the central nearest - neighbor point P center and the low - dimensional nearest - neighbor point P low can be taken to obtain the cross - nearest - neighbor point P inter , that is, P inter = P center ∩P low . ​

[0085] Then, the cross-nearest neighbor point P inter can be expressed as:

[0086] ;

[0087] where p i is the i-th cross-nearest neighbor point, i ∈ [1, N inter , N inter is the number of cross-nearest neighbor points, is the coordinate value of p i under the j-th dimension, j ∈ [1, m], and m is the embedding dimension.

[0088] Step 106: Fuse multiple cross-nearest neighbor points into multi-information points.

[0089] In this embodiment, multiple cross-nearest neighbor points can be fused in a linear or non-linear manner to obtain multi-information points.

[0090] In a fusion method, in the known m-dimensional phase space point set X m , there is a central point X(n) that is relatively relevant to the prediction target. In the known (m - 1)-dimensional phase space point set X m-1 , there is a low-dimensional prediction target point X(n + 1) m-1 that is relatively relevant to the prediction target. Therefore, a multi-dimensional space distance metric can be constructed based on the central point X(n) and the low-dimensional prediction target point X(n + 1) m-1 .

[0091] In specific implementation, on the one hand, calculate the third distance between each cross-nearest neighbor point p i and the central point X(n) respectively. On the other hand, calculate the fourth distance between each cross-nearest neighbor point p i and the low-dimensional prediction target point X(n + 1) m-1 respectively;

[0092] For the same cross-nearest neighbor point p i , add the third distance and the fourth distance to obtain the original distance metric.

[0093] Taking the Euclidean distance as an example of the third distance and the fourth distance, the original distance metric can be expressed as:

[0094] ;

[0095] where is the original distance metric of the i-th cross-nearest neighbor point p i , is the coordinate value of p i under the j-th dimension, x n+(j-1)τis the coordinate value of the center point X(n) in the j-th dimension, x n+1+(j-1)τ is the low-dimensional prediction target point X(n + 1) m-1 in the coordinate value of the j-th dimension, i ∈ [1, N inter , N inter is the number of cross-nearest neighbor points, j ∈ [1, m], m is the embedding dimension, and τ is the delay time.

[0096] Then, the set F of the original distance metrics of each cross-nearest neighbor point can be expressed as: .

[0097] Using algorithms such as Min-Max (minimum-maximum) and Z-Score (Z-score) to normalize the distance metrics corresponding to multiple cross-nearest neighbor points to obtain the target distance metrics corresponding to multiple cross-nearest neighbor points. Then, the set of the target distance metrics of each cross-nearest neighbor point can be expressed as:

[0098] , where is the target distance metric of the i-th cross-nearest neighbor point p i i ∈ [1, N inter , N inter is the number of cross-nearest neighbor points.

[0099] Configure weights for the cross-nearest neighbor points according to the target distance metrics; among them, the weight of the current cross-nearest neighbor point is the ratio between the inverse metric of the current cross-nearest neighbor point and the inverse metrics of all cross-nearest neighbor points, and the inverse metric is the difference between 1 and the target distance metric.

[0100] Then, the weight can be expressed as:

[0101] ;

[0102] where, w i is the weight of the i-th cross-nearest neighbor point, is the target distance metric of the i-th cross-nearest neighbor point, is the target distance metric of the j-th cross-nearest neighbor point, i, j ∈ [1, N inter , N inter is the number of cross-nearest neighbor points.

[0103] For the same dimension, sum the products of the weights and the coordinate values of the cross-nearest neighbor points in this dimension to obtain the coordinate value of the multi-information point in this dimension.

[0104] Let the multi-information point A = {a1, a2, …, a j , …, a m}, where, a jis the coordinate value of the multi-information point in the j-th dimension.

[0105] ;

[0106] where w i is the weight of the i-th cross-neighbor point p i and is the coordinate value of the i-th cross-neighbor point p i in the j-th dimension, j ∈ [1, m], m is the number of dimensions, i ∈ [1, N inter , and N inter is the number of cross-neighbor points.

[0107] Step 107: Input the multi-information point and the center point into the power prediction model to predict the expected power generated by the wind turbine within the future time range.

[0108] In practical applications, for very short-term wind power prediction, a power prediction model can be pre-constructed in advance, and the power prediction model can be supervised-trained and verified using historical samples (multi-information points and center points) collected from the wind turbine. When the power prediction model completes supervised training and verification, the power prediction model is deployed for online operation.

[0109] At this time, the multi-information point and the center point can be input into the power prediction model separately or spliced for processing, so as to predict multiple powers generated by the wind turbine within the future time range, denoted as the expected power.

[0110] Among them, the power prediction model can include machine learning models, such as the Least Square Support Vector Machine (LSSVM), etc., or can also include deep learning models, such as the Deep Belief Network (DBN), etc. This embodiment does not limit this. Use historical data to train a high-precision power prediction model, and optimize the parameters of the power prediction model according to different data sets to improve the stability and robustness of the power prediction model for very short-term wind power prediction.

[0111] Furthermore, for deep learning models, the structure of the power prediction model is not limited to the artificially designed neural network, and can also be a neural network optimized by model quantization methods, a neural network searched by the NAS (Neural Architecture Search) method for the characteristics of very short-term wind power prediction, etc. This embodiment does not limit this.

[0112] In one embodiment of the present invention, step 107 may include the following steps:

[0113] Step 1071: Load the power prediction model.

[0114] In this embodiment, as Figure 2 shown, the power prediction model adopts the architecture of an encoder-decoder, which includes a first encoder Encoder_1, a second encoder Encoder_2, a third encoder Encoder_3, a fusion module Fusion_Module, and a decoder Decoder. During training, a cross-entropy loss function can be used for supervised training.

[0115] Among them, the first encoder Encoder_1, the second encoder Encoder_2, and the third encoder Encoder_3 are all parts of a complete encoder, which are respectively used for encoding multi-information points and / or center points, and extracting high-level feature maps from the multi-information points and / or center points.

[0116] The fusion module Fusion_Module is used to process and fuse the encoded multiple feature maps to enhance the expression ability of the feature maps.

[0117] The decoder Decoder is used to decode the feature maps with enhanced expression ability to achieve ultra-short-term wind power prediction.

[0118] Step 1072: Input the multi-information points into the first encoder to encode them into target multi-information features.

[0119] In this embodiment, as Figure 2 shown, the multi-information points are separately input into the first encoder Encoder_1 for encoding to obtain target multi-information features.

[0120] In one design, as Figure 2 shown, the first encoder includes two layers of long short-term memory networks (Long Short-Term Memory, LSTM), which are respectively denoted as the first long short-term memory network LSTM_1 and the second long short-term memory network LSTM_2.

[0121] In this design, the multi-information points are input into the first long short-term memory network LSTM_1 to extract candidate multi-information features, and the candidate multi-information features are input into the second long short-term memory network LSTM_2 to extract target multi-information features.

[0122] Two layers of unidirectional LSTM can capture unidirectional temporal dependencies from multi-information points, realize layer-by-layer abstraction of features, the bottom LSTM captures local patterns (i.e., short-term dependencies), and the top LSTM captures global patterns (i.e., long-term dependencies), enhancing the semantic level of features.

[0123] Step 1073: Concatenate the multi-information points and the center points into enhanced points.

[0124] In this embodiment, as Figure 2 shown, the Concat function can be used to splice multiple information points with the center point into enhanced points to form a richer feature representation.

[0125] Step 1074: Input the enhanced points into the second encoder to encode them into target enhanced features.

[0126] In this embodiment, as Figure 2 shown, input the enhanced points into the second encoder Encoder_2 for encoding to obtain target enhanced features.

[0127] In one design, as Figure 2 shown, the second encoder Encoder_2 includes two layers of bidirectional long short-term memory networks (Bidirectional Long Short Term Memory, Bi-LSTM), denoted as the first bidirectional long short-term memory network Bi-LSTM_1 and the second bidirectional long short-term memory network Bi-LSTM_2 respectively. Bi-LSTM is composed of two independent LSTM networks. When Bi-LSTM processes sequence data, the sequence data will be input into the two LSTM networks in forward and reverse orders respectively to extract features, and the features of the features will be spliced to form an output vector as the final output of this time step.

[0128] In this design, input the enhanced points into the first bidirectional long short-term memory network Bi-LSTM_1 to extract candidate enhanced features, and input the candidate enhanced features into the second bidirectional long short-term memory network Bi-LSTM_2 to extract target enhanced features.

[0129] The two-layer bidirectional Bi-LSTM can perform double-time series modeling on the enhanced points, capture the potential correlation between historical and future information, and enhance the context understanding, so that the target enhanced features contain more comprehensive sequence dependencies.

[0130] Step 1075: Input the center point into the third encoder to encode it into target center features.

[0131] In this embodiment, as Figure 2 shown, input the center point into the third encoder Encoder_3 for encoding to obtain target center features.

[0132] In one design, as Figure 2 shown, the third encoder Encoder_3 includes two layers of LSTM, denoted as the third long short-term memory network LSTM_3 and the fourth long short-term memory network LSTM_4 respectively.

[0133] In this design, the center point is input into the third long short-term memory network LSTM_3 to extract the candidate center features, and the candidate center features are input into the fourth long short-term memory network LSTM_4 to extract the target center features.

[0134] Two layers of unidirectional LSTM can capture unidirectional temporal dependencies from the central point and realize layer-by-layer abstraction of features. The bottom-level LSTM captures local patterns (i.e., short-term dependencies), and the high-level LSTM captures global patterns (i.e., long-term dependencies), thus enhancing the semantic hierarchy of features.

[0135] Step 1076: Input the target multi-information features, target enhancement features and target center features into a fusion module to fuse them into multimodal features.

[0136] In this embodiment, if Figure 2 As shown, the target multi-information features, target enhancement features and target center features can be input into the fusion module Fusion_Module for interaction and fusion to obtain multimodal features.

[0137] In one design, Figure 2 As shown in the figure, the fusion module Fusion_Module includes a multilayer perceptron (MLP), two layers of self-attention modules (respectively denoted as the first self-attention module Self-Attention_1 and the second self-attention module Self-Attention_2) and a convolutional block attention module (CBAM).

[0138] In this design, the target enhanced features are input into the multi-layer perceptron MLP and mapped into the first encoded intermediate features as the common basic features.

[0139] Use the Concat function to concatenate the target multi-information feature and the first encoding intermediate feature into the second encoding intermediate feature.

[0140] The second encoded intermediate feature is input into the first self-attention module Self-Attention_1, and the first attention feature is generated according to the self-attention mechanism. The self-attention mechanism can mine the hierarchical dependency between multiple information points and enhancement points to achieve homomodal feature enhancement.

[0141] Use the Concat function to concatenate the target center feature and the first encoded intermediate feature into the third encoded intermediate feature.

[0142] The third encoded intermediate feature is input into the second self-attention module Self-Attention_2, and the second attention feature is generated according to the self-attention mechanism. The self-attention mechanism can mine the hierarchical dependency between the center point and the enhancement point to achieve homomodal feature enhancement.

[0143] Use the Concat function to concatenate the first attention feature and the second attention feature into the third attention feature.

[0144] The third attention feature is input into the convolutional block attention module CBAM to generate multimodal features, channel attention and spatial attention are provided in the convolutional block attention module CBAM, and high-value features are extracted from the mixed features (i.e., the third attention feature), so that the power prediction model pays more attention to the areas and key semantic channels that are more relevant to the ultra-short-term wind power prediction.

[0145] Step 1077: Input the multimodal features into a decoder for decoding to obtain the expected power generated by the wind turbine generator set within a future time range.

[0146] In this embodiment, if Figure 2 As shown, the multimodal features are input into the decoder for decoding to obtain the expected power generated by the wind turbine within the future time range.

[0147] In one design, Figure 2 As shown, the decoder Decoder includes a first decoding layer Transformer_1, a second decoding layer Transformer_2 and a third decoding layer Transformer_3, that is, the first decoding layer Transformer_1, the second decoding layer Transformer_2 and the third decoding layer Transformer_3 are all Transformers.

[0148] To reduce computational costs, the three-layer Transformer can be lightweight (such as reducing the number of heads, using a bottleneck structure) or optimized for low-rank approximation.

[0149] In this design, the multimodal features are input into the first decoding layer Transformer_1 for decoding to obtain the first decoding intermediate features.

[0150] The first decoded intermediate feature is input into the second decoding layer Transformer_2 for decoding to obtain the second decoded intermediate feature.

[0151] The second decoded intermediate feature is input into the third decoding layer Transformer_3 for decoding to obtain the expected power generated by the wind turbine generator set within the future time range.

[0152] The three-layer Transformer can extract features layer by layer. The bottom-layer Transformer is responsible for capturing local details and building short-distance dependency relationships. The middle-layer Transformer is responsible for learning middle-layer semantics and establishing cross-region associations. The top-layer Transformer is responsible for learning high-layer semantics and modeling long-distance dependencies, with strong anti-interference ability, which helps to improve the accuracy of sequence generation tasks (i.e., ultra-short-term wind power prediction).

[0153] In an experiment, the ultra-short-term wind power prediction method in this embodiment (using cross-nearest neighbor points P inter prediction) was compared with the traditional ultra-short-term wind power prediction method (using the prediction of neighboring points near the prediction center).

[0154] The wind power time series selected for the experiment was the data of the entire system of a local power transmission system operator from January 1, 2018 to January 31, 2018, and the time granularity of the data was 15 minutes. The delay time of the phase space was selected as 19 by the mutual information method, and the embedding dimension of the phase space was selected as 5 by the saturated correlation dimension method to construct the phase space. In the phase space, the central neighboring point P center and the number of points contained in the low-dimensional neighboring point P low was 16. The prediction time scales were 60 minutes, 90 minutes, and 120 minutes in advance, and all three time scales were within the range of ultra-short-term prediction.

[0155] In this experiment, the mean absolute error (MAE) and the root mean square error (RMSE) were used as evaluation indicators to judge the prediction effect.

[0156] Among them, the mean absolute error (MAE) was:[[]]

[0157] ;

[0158] The root mean square error (RMSE) was:[[]]

[0159] ;

[0160] Among them, WS(i) is the power actually generated by the wind turbine at time i, WS(i)' represents the predicted power of the wind turbine at time i, and N is the length of time.

[0161] The smaller the MAE, the higher the accuracy of the ultra-short-term wind power prediction method. The smaller the RMSE, the closer the predicted power of the wind turbine by the ultra-short-term wind power prediction method is to the power actually generated by the wind turbine.

[0162] The performance comparison between the ultra-short-term wind power prediction method and the traditional ultra-short-term wind power prediction method is as follows:[[]]

[0163] ;

[0164] Comparing the evaluation indicators, it can be seen that in this embodiment, the overall prediction effect of the ultra-short-term wind power prediction method is better than that of the traditional ultra-short-term wind power prediction method, and it can effectively improve the accuracy of ultra-short-term wind power prediction.

[0165] Step 108, perform business operations on the wind turbine according to the expected power.

[0166] In practical applications, corresponding business operations can be performed on the wind turbine according to the expected power generated by the wind turbine within the future time range. For example, wind power consumption, stable operation of the power grid, energy optimization allocation, and so on.

[0167] For example, for the stable operation of the power grid, if it is predicted that the wind power will decrease in the next 2 hours, the power grid can start the standby coal-fired unit in advance to fill the power gap.

[0168] In this embodiment, according to the delay time and the embedding dimension, multiple actual powers generated by the wind turbine in history are reconstructed into a phase space; there are multiple first phase points in the phase space; the sampling time is spaced between two adjacent actual powers; if the last first phase point is selected as the center point, the center point is extended into multiple second phase points according to the step size; the product of the step size and the sampling time is a time range less than the ultra-short-term time threshold; find multiple first phase points adjacent to the center point as the center adjacent points; if the first second phase point is selected as the actual prediction point, under the condition of reducing the embedding dimension, find multiple first phase points adjacent to the actual prediction point to obtain low-dimensional adjacent points; take the intersection of the center adjacent points and the low-dimensional adjacent points to obtain cross adjacent points; fuse multiple cross adjacent points into multi-information points; input the multi-information points and the center point into the power prediction model to predict the expected power generated by the wind turbine within the future time range; perform business operations on the wind turbine according to the power generated by the wind turbine within the future time range. In this embodiment, the low-dimensional adjacent points are used to further screen the center adjacent points to ensure that the selected points can be more relevant to the prediction target, improve the representativeness of the input of the power prediction model, and thus improve the accuracy of the power prediction model in predicting ultra-short-term wind power and ensure the quality of business operations.

[0169] Embodiment 2

[0170] See Figure 3 , which shows a schematic structural diagram of a business operation device based on wind power provided by Embodiment 2 of the present invention. As Figure 3 shown, the device includes:

[0171] The phase space reconstruction module 301 is configured to reconstruct multiple actual powers generated by the wind turbine generator in history into a phase space according to a delay time and an embedding dimension; there are multiple first phase points in the phase space; there is a sampling time interval between two adjacent actual powers;

[0172] The phase point extension module 302 is configured to, if the last one of the first phase points is selected as the center point, extend the center point into multiple second phase points according to a step size; the product of the step size and the sampling time is a time duration range with a length less than the ultra-short-term time threshold;

[0173] The central near-neighbor point searching module 303 is configured to search for multiple first phase points that are near neighbors of the center point as central near-neighbor points;

[0174] The low-dimensional near-neighbor point searching module 304 is configured to, if the first one of the second phase points is selected as the actual prediction point, search for multiple first phase points that are near neighbors of the actual prediction point under the condition of reducing the embedding dimension to obtain low-dimensional near-neighbor points;

[0175] The near-neighbor point intersection module 305 is configured to take the intersection of the central near-neighbor points and the low-dimensional near-neighbor points to obtain intersection near-neighbor points;

[0176] The intersection near-neighbor point fusion module 306 is configured to fuse multiple intersection near-neighbor points into multi-information points;

[0177] The expected power prediction module 307 is configured to input the multi-information points and the center point into a power prediction model to predict the expected power generated by the wind turbine generator within the future time duration range;

[0178] The service operation execution module 308 is configured to execute service operations on the wind turbine generator according to the expected power.

[0179] In an embodiment of the present invention, the phase point extension module 302 includes:

[0180] The variable determination module is configured to determine multiple variables; the value range of the variables is a positive integer between 1 and the step size;

[0181] The coordinate value extension module is configured to extend the coordinate values in each dimension of the center point according to multiple variables respectively to obtain the coordinate values in each dimension of multiple second phase points; wherein, the coordinate value of the last dimension of the second phase point corresponds to the power with the quantity of the step size after multiple actual powers of the wind turbine generator.

[0182] In an embodiment of the present invention, the central near-neighbor point searching module 303 includes:

[0183] The first distance calculation module is configured to calculate, for other first phase points except the central point, a first distance between the central point and the other first phase points;

[0184] The first distance screening module is configured to screen out multiple other first phase points with the smallest first distance as central neighboring points neighboring the central point;

[0185] The low-dimensional neighboring point finding module 304 includes:

[0186] The point dimensionality reduction module is configured to, when the delay time is less than the sum value between the number of the actual power and 1, reduce the dimension of the actual prediction point by one to obtain a low-dimensional prediction target point, and remove the coordinate value of the last dimension in each of the first phase points to obtain low-dimensional phase points;

[0187] The second distance calculation module is configured to calculate a second distance between the low-dimensional prediction target point and the other low-dimensional phase points;

[0188] The second distance screening module is configured to screen out the first phase points corresponding to multiple low-dimensional phase points with the smallest second distance as low-dimensional neighboring points neighboring the low-dimensional prediction target point.

[0189] In an embodiment of the present invention, the cross neighboring point fusion module 306 includes:

[0190] The third distance calculation module is configured to calculate a third distance between each of the cross neighboring points and the central point respectively;

[0191] The fourth distance calculation module is configured to calculate a fourth distance between each of the cross neighboring points and the low-dimensional prediction target point respectively;

[0192] The distance addition module is configured to add the third distance and the fourth distance for the same cross neighboring point to obtain an original distance index;

[0193] The normalization module is configured to normalize multiple distance indexes to obtain multiple target distance indexes;

[0194] The weight configuration module is configured to configure weights for the cross neighboring points according to the target distance indexes; the weight of the current cross neighboring point is the ratio between the anti-phase index of the current cross neighboring point and the anti-phase indexes of all the cross neighboring points, and the anti-phase index is the difference between 1 and the target distance index;

[0195] The weighted summation module is configured to sum the products of the weights and the coordinate values of the cross neighboring points in the same dimension to obtain the coordinate value of the multi-information point in the same dimension.

[0196] In one embodiment of the present invention, the expected power prediction module 307 includes:

[0197] A power prediction model loading module for loading a power prediction model; the power prediction model includes a first encoder, a second encoder, a third encoder, a fusion module, and a decoder;

[0198] A first encoding module for encoding the multi-information points into target multi-information features by inputting them into the first encoder;

[0199] An enhanced point splicing module for splicing the multi-information points and the center point into enhanced points;

[0200] A second encoding module for encoding the enhanced points into target enhanced features by inputting them into the second encoder;

[0201] A third encoding module for encoding the center point into target center features by inputting them into the third encoder;

[0202] A multi-modal feature fusion module for fusing the target multi-information features, the target enhanced features, and the target center features into multi-modal features by inputting them into the fusion module;

[0203] A decoding module for decoding the multi-modal features by inputting them into the decoder to obtain the expected power generated by the wind turbine within the future time range.

[0204] In one embodiment of the present invention, the first encoder includes a first long short-term memory network and a second long short-term memory network, and the first encoding module is further configured to:

[0205] Input the multi-information points into the first long short-term memory network to extract candidate multi-information features;

[0206] Input the candidate multi-information features into the second long short-term memory network to extract target multi-information features;

[0207] The second encoder includes a first bidirectional long short-term memory network and a second bidirectional long short-term memory network, and the second encoding module is further configured to:

[0208] Input the enhanced points into the first bidirectional long short-term memory network to extract candidate enhanced features;

[0209] Input the candidate enhanced features into the second bidirectional long short-term memory network to extract target enhanced features;

[0210] The third encoder includes a third long short-term memory network and a fourth long short-term memory network, and the third encoding module is further configured to:

[0211] Input the center point into the third long short-term memory network to extract candidate center features;

[0212] Input the candidate center features into the fourth long short-term memory network to extract target center features.

[0213] In an embodiment of the present invention, the fusion module includes a multi-layer perceptron, a first self-attention module, a second self-attention module, and a convolutional block attention module; the multi-modal feature fusion module is further configured to:

[0214] Input the target enhanced features into the multi-layer perceptron to map them into first encoded intermediate features;

[0215] Concatenate the target multi-information features and the first encoded intermediate features into second encoded intermediate features;

[0216] Input the second encoded intermediate features into the first self-attention module to generate first attention features;

[0217] Concatenate the target center features and the first encoded intermediate features into third encoded intermediate features;

[0218] Input the third encoded intermediate features into the second self-attention module to generate second attention features;

[0219] Concatenate the first attention features and the second attention features into third attention features;

[0220] Input the third attention features into the convolutional block attention module to generate multi-modal features.

[0221] In an embodiment of the present invention, the decoder includes a first decoding layer, a second decoding layer, and a third decoding layer, and the first decoding layer, the second decoding layer, and the third decoding layer are all Transformers; the decoding module is further configured to:

[0222] Input the multi-modal features into the first decoding layer for decoding to obtain first decoded intermediate features;

[0223] Input the first decoded intermediate features into the second decoding layer for decoding to obtain second decoded intermediate features;

[0224] Input the second decoded intermediate features into the third decoding layer for decoding to obtain the expected power generated by the wind turbine within the future time period.

[0225] The service operation device based on wind power provided by the embodiments of the present invention can execute the method for service operation based on wind power provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method for service operation based on wind power.

[0226] Embodiment III

[0227] Refer to Figure 4 , which shows a schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, blade servers, mainframe computers, and other suitable computers. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0228] As Figure 4 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0229] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0230] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the method for service operation based on wind power.

[0231] In some embodiments, a business operation method based on wind power can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the business operation method based on wind power described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the business operation method based on wind power by any other suitable means (e.g., by means of firmware).

[0232] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), systems on a chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0233] The computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer programs are executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0234] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0235] In order to provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0236] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0237] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0238] Embodiment 4

[0239] The embodiment of the present invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the business operation method based on wind power as provided in any embodiment of the present invention.

[0240] In the process of implementing the computer program product, the computer program code for performing the operations of the present invention can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN) - or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0241] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0242] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A business operation method based on wind power generation, characterized in that, Including: Reconstructing multiple actual powers generated by a wind turbine in history into a phase space according to a delay time and an embedding dimension; There are multiple first phase points in the phase space; the sampling time is spaced between two adjacent actual powers; If the last one of the first phase points is selected as the center point, extending the center point into multiple second phase points according to a step size; the product of the step size and the sampling time is a time range with a length less than the ultra-short-term time threshold; Finding multiple first phase points adjacent to the center point as center adjacent points; If the first one of the second phase points is selected as the actual prediction point, finding multiple first phase points adjacent to the actual prediction point under the condition of reducing the embedding dimension to obtain low-dimensional adjacent points; Taking the intersection of the center adjacent points and the low-dimensional adjacent points to obtain intersection adjacent points; Fusing multiple intersection adjacent points into multi-information points; Inputting the multi-information points and the center point into a power prediction model to predict the expected power generated by the wind turbine within the future time range; Performing business operations on the wind turbine according to the expected power.

2. The method according to claim 1, wherein The extending the center point into multiple second phase points according to a step size includes: Determining multiple variables; the value range of the variables is a positive integer between 1 and the step size; Respectively extending the coordinate values in each dimension of the center point according to the multiple variables to obtain the coordinate values in each dimension of multiple second phase points; wherein, the coordinate value of the last dimension of the second phase point corresponds to the power with the number of the step size after multiple actual powers of the wind turbine.

3. The method according to claim 1, wherein The finding multiple first phase points adjacent to the center point as center adjacent points includes: For other first phase points except the center point, calculating a first distance between the center point and other first phase points; Selecting multiple other first phase points with the smallest first distance as center adjacent points adjacent to the center point; The finding multiple first phase points adjacent to the actual prediction point under the condition of reducing the embedding dimension to obtain low-dimensional adjacent points includes: When the delay time is less than the sum value of the number of actual powers and 1, reducing one dimension of the actual prediction point to obtain a low-dimensional prediction target point, and removing the coordinate value of the last dimension of each first phase point to obtain a low-dimensional phase point; Calculating a second distance between the low-dimensional prediction target point and other low-dimensional phase points; Selecting the first phase points corresponding to multiple low-dimensional phase points with the smallest second distance as low-dimensional adjacent points adjacent to the low-dimensional prediction target point.

4. The method according to claim 3, wherein The fusing multiple intersection adjacent points into multi-information points includes: Respectively calculating a third distance between each intersection adjacent point and the center point; Respectively calculating a fourth distance between each intersection adjacent point and the low-dimensional prediction target point; For the same intersection adjacent point, adding the third distance and the fourth distance to obtain an original distance index; Normalizing multiple distance indexes to obtain multiple target distance indexes; Configure weights for the cross-nearest neighbors according to the target distance metric; currently, the weight of the cross-nearest neighbor is the ratio between the inverse metric of the current cross-nearest neighbor and the inverse metrics of all the cross-nearest neighbors, and the inverse metric is the difference between 1 and the target distance metric. For the same dimension, sum the products of the weights and the coordinate values of the cross-nearest neighbors in this dimension to obtain the coordinate value of the multi-information point in this dimension.

5. The method according to any one of claims 1-4, characterized in that, The step of inputting the multi-information point and the center point into the power prediction model to predict the expected power generated by the wind turbine within the future time range includes: Load the power prediction model; the power prediction model includes a first encoder, a second encoder, a third encoder, a fusion module, and a decoder. Input the multi-information point into the first encoder to be encoded into a target multi-information feature. Concatenate the multi-information point and the center point into an enhanced point. Input the enhanced point into the second encoder to be encoded into a target enhanced feature. Input the center point into the third encoder to be encoded into a target center feature. Input the target multi-information feature, the target enhanced feature, and the target center feature into the fusion module to be fused into a multi-modal feature. Input the multi-modal feature into the decoder to be decoded to obtain the expected power generated by the wind turbine within the future time range.

6. The method according to claim 5, wherein The first encoder includes a first long short-term memory network and a second long short-term memory network. The step of inputting the multi-information point into the first encoder to be encoded into a target multi-information feature includes: Input the multi-information point into the first long short-term memory network to extract a candidate multi-information feature. Input the candidate multi-information feature into the second long short-term memory network to extract a target multi-information feature. The second encoder includes a first bidirectional long short-term memory network and a second bidirectional long short-term memory network. The step of inputting the enhanced point into the second encoder to be encoded into a target enhanced feature includes: Input the enhanced point into the first bidirectional long short-term memory network to extract a candidate enhanced feature. Input the candidate enhanced feature into the second bidirectional long short-term memory network to extract a target enhanced feature. The third encoder includes a third long short-term memory network and a fourth long short-term memory network. The step of inputting the center point into the third encoder to be encoded into a target center feature includes: Input the center point into the third long short-term memory network to extract a candidate center feature. Input the candidate center feature into the fourth long short-term memory network to extract a target center feature.

7. The method according to claim 6, characterized in that, The fusion module includes a multi-layer perceptron, a first self-attention module, a second self-attention module, and a convolutional block attention module. The step of inputting the target multi-information feature, the target enhanced feature, and the target center feature into the fusion module to be fused into a multi-modal feature includes: Input the target enhanced feature into the multi-layer perceptron to be mapped into a first encoded intermediate feature. Concatenate the target multi-information feature and the first encoded intermediate feature into a second encoded intermediate feature. Input the second encoded intermediate feature into the first self-attention module to generate a first attention feature; Concatenate the target center feature and the first encoded intermediate feature into a third encoded intermediate feature; Input the third encoded intermediate feature into the second self-attention module to generate a second attention feature; Concatenate the first attention feature and the second attention feature into a third attention feature; Input the third attention feature into the convolutional block attention module to generate a multi-modal feature.

8. The method according to claim 7, characterized in that The decoder includes a first decoding layer, a second decoding layer, and a third decoding layer, and the first decoding layer, the second decoding layer, and the third decoding layer are all Transformers; The step of inputting the multi-modal feature into the decoder for decoding to obtain the expected power generated by the wind turbine within the future time range includes: Input the multi-modal feature into the first decoding layer for decoding to obtain a first decoded intermediate feature; Input the first decoded intermediate feature into the second decoding layer for decoding to obtain a second decoded intermediate feature; Input the second decoded intermediate feature into the third decoding layer for decoding to obtain the expected power generated by the wind turbine within the future time range.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the wind power-based service operation method according to any one of claims 1-8.

10. 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, it implements the wind power-based service operation method according to any one of claims 1-7.

Images