Distributed wind power prediction method, device and equipment

Abstract

The application relates to a distributed wind power prediction method, device and equipment. The method comprises the following steps: obtaining wind power data of distributed wind turbines as sample data, wherein the wind power data comprises multiple dimensions; preprocessing the sample data, constructing a time series data set according to the preprocessed sample data, determining the relative position relationship of each distributed wind turbine according to the wind power data, generating an adjacency matrix and an edge weight matrix according to the relative position relationship of each distributed wind turbine, and constructing a tensor graph according to the number of distributed wind turbines, the adjacency matrix and the time series data set. A prediction model is built and trained according to the tensor graph and the edge weight matrix, and the prediction model is used to output a prediction result according to the wind power data of the distributed wind turbines. In the application, the tensor graph used for training the prediction model considers the relative position relationship of each distributed wind turbine, thereby improving the wind power prediction accuracy of the prediction model for wind turbines in different physical environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power prediction, in particular to a decentralized wind power prediction method, device and equipment. BACKGROUND

[0002] Wind power is a clean renewable energy source with good application prospects. China's renewable energy natural endowments are mainly concentrated in the northwest region, so the wind power in the west develops rapidly. However, due to the vast area and few people in the west, the phenomenon of abandoned wind is significant, and the utilization rate of wind energy resources is not high. The eastern and southern parts of China, as load centers, have not been able to ease the pressure of electricity consumption.

[0003] Due to the complex terrain and uneven distribution of wind energy resources in the east and south of China, it is difficult to develop centralized wind power, so currently multiple points of decentralized wind turbines are connected to the power grid, thereby relieving the pressure of electricity consumption, optimizing the power quality at the end of the power grid, and reducing the line loss of the power grid. Therefore, the construction of decentralized wind power in the load center area is one of the future development trends.

[0004] The existing technology mainly focuses on centralized wind power prediction. Compared with centralized wind power, decentralized wind power has the characteristics of multiple point access and random distribution. The physical environment of different wind turbines is different, and the power prediction is difficult. The randomness and intermittency of wind power output have an impact on the stability of the power grid, which limits the development of decentralized wind power. SUMMARY

[0005] In order to at least partially overcome the problem of difficulty in predicting the power of decentralized wind power in the related art, the present application provides a decentralized wind power prediction method, device and equipment.

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

[0007] According to a first aspect of an embodiment of the present application, a decentralized wind power prediction method is provided, comprising:

[0008] Obtaining wind power data of a decentralized wind turbine as sample data; wherein the wind power data comprises multiple dimensions;

[0009] Preprocessing the sample data;

[0010] Constructing a time series data set according to the preprocessed sample data;

[0011] Determining the relative position relationship of each decentralized wind turbine according to the wind power data, and generating an adjacency matrix and an edge weight matrix according to the relative position relationship of each decentralized wind turbine;

[0012] constructing a tensor graph according to the number of the distributed wind turbines, the adjacency matrix and the time series data set;

[0013] building and training a prediction model according to the tensor graph and the edge weight matrix; the prediction model is used to output a prediction result according to wind power data of a distributed wind turbine.

[0014] Preferably, in an implementable manner of the application, the dimension of the wind power data at least includes:

[0015] geographical position data of each of the distributed wind turbines, topography data of each of the distributed wind turbines, meteorological data and power data; wherein the meteorological data at least includes wind speed data, wind direction data, temperature data and humidity data.

[0016] Preferably, in an implementable manner of the application, the preprocessing of the sample data includes:

[0017] filling in missing values of the sample data;

[0018] unifying data sampling time of the meteorological data and the power data;

[0019] performing data cleaning on the wind speed data based on a preset full load wind speed threshold;

[0020] vectorizing the geographical position data and the topography data of the distributed wind turbines;

[0021] performing trigonometric functionization on the wind direction data;

[0022] performing normalization processing on different types of the sample data respectively.

[0023] Preferably, in an implementable manner of the application, the constructing of the time series data set according to the preprocessed sample data includes:

[0024] constructing a time series data set X = {X t |t∈[0,T]}, wherein X t = {x 1,t ,L x i,t ,L x m,t}, x t = {x i,t |i∈[1,m]}, X t represents a wind power data sequence in a state at time t, x i,t represents an i th wind power data element at time t and there are m wind power data elements in total.

[0025] Preferably, in an implementable manner of the present application, the constructing the time series dataset according to the preprocessed sample data further comprises:

[0026] Based on the preset time interval, a plurality of time series segments with different lengths are intercepted along the time axis as the periodic components, wherein each periodic component is an integer multiple of the preset time interval;

[0027] Constructing a component time series dataset corresponding to each periodic component.

[0028] Preferably, in an implementable manner of the present application, the generating an adjacency matrix according to the relative position relationship of each of the distributed wind turbines comprises:

[0029] Calculating the element a ij in the adjacency matrix A:

[0030]

[0031] wherein a ij represents an element in the adjacency matrix A, D ij represents the distance between the distributed wind turbines i and j; σ and ε both represent a preset connectivity threshold of the graph;

[0032] Generating the adjacency matrix based on the elements in the adjacency matrix.

[0033] Preferably, in an implementable manner of the present application, the generating an edge weight matrix according to the relative position relationship of each of the distributed wind turbines comprises:

[0034] Calculating the weight between the distributed wind turbines i and j in the edge weight matrix:

[0035] w i,j =f(Δ||dis' i,j ||,Δh' i,j ,Δ||g' i,j ||,d ij )

[0036] =1 / (αgΔ||dis' i,j ||+βgΔh' i,j +μgΔ||g' i,j ||+νgd ij +k)

[0037]

[0038] wherein w i,jrepresents the weight between the distributed wind turbines i and j in the edge weight matrix W, f(g) represents an initial weight calculation function, ||dis|| represents the wind direction data norm after data preprocessing, Δ||dis i,j || is the difference between the wind direction data norm values between the distributed wind turbines i and j after data preprocessing, Δh' i,j represents the difference in altitude between the distributed wind turbines i and j after data preprocessing, Δ||g' i,j || represents the terrain difference condition between the distributed wind turbines i and j, d i,j represents the straight-line distance between the distributed wind turbines i and j, α, β, μ, ν, k represent preset function coefficients, w′ i,j represents the weight between the distributed wind turbines i and j in the normalized edge weight matrix.

[0039] Preferably, in an implementable manner of the present application, the prediction model is built and trained according to the tensor graph and the edge weight matrix, comprising:

[0040] Model training is performed according to the tensor graph and the edge weight matrix constructed according to each component time series data set;

[0041] The prediction results and weight parameters trained for each component time series data set are outputted;

[0042] According to the prediction results and weight parameters corresponding to each component time series data set, a final prediction result is obtained.

[0043] Preferably, in an implementable manner of the present application, the prediction model is a Transformer model, and the Transformer model framework is composed of an encoder and a decoder, and the encoder and the decoder are composed of an attention mechanism combination; the attention mechanism combination includes a multi-head spatial attention module and a multi-head temporal attention module, the multi-head spatial attention module is used to learn the node information in the spatial dimension, and the multi-head temporal attention module is used to learn the node information in the time dimension.

[0044] According to a second aspect of the embodiments of the present application, a distributed wind power prediction device is provided, comprising:

[0045] a processor and a memory;

[0046] The processor and the memory are connected through a communication bus;

[0047] The processor is configured to call and execute the program stored in the memory.

[0048] The memory is configured to store a program, and the program is configured to at least implement the distributed wind power prediction method according to any one of the preceding embodiments.

[0049] According to a third aspect of the embodiments of the present application, a distributed wind power prediction device is provided, comprising:

[0050] The acquisition module is configured to acquire wind power data of the distributed wind power generators as sample data, wherein the wind power data comprises multiple dimensions.

[0051] The preprocessing module is configured to preprocess the sample data.

[0052] The data set construction module is configured to construct a time series data set according to the preprocessed sample data.

[0053] The matrix generation module is configured to determine relative position relationships of the distributed wind power generators according to the wind power data, and generate an adjacency matrix and an edge weight matrix according to the relative position relationships of the distributed wind power generators.

[0054] The tensor graph construction module is configured to construct a tensor graph according to the number of the distributed wind power generators, the adjacency matrix and the time series data set.

[0055] The model training module is configured to build and train a prediction model according to the tensor graph and the edge weight matrix, and the prediction model is configured to output a prediction result according to the wind power data of the distributed wind power generators.

[0056] The technical scheme provided by the present application can have the following beneficial effects: the distributed wind power prediction method provided by the present application comprises: acquiring wind power data of distributed wind power generators as sample data, and the wind power data comprises multiple dimensions. The sample data is preprocessed, a time series data set is constructed according to the preprocessed sample data, relative position relationships of the distributed wind power generators are determined according to the wind power data, an adjacency matrix and an edge weight matrix are generated according to the relative position relationships of the distributed wind power generators, and a tensor graph is constructed according to the number of the distributed wind power generators, the adjacency matrix and the time series data set. A prediction model is built and trained according to the tensor graph and the edge weight matrix, and the prediction model is configured to output a prediction result according to the wind power data of the distributed wind power generators. Since the tensor graph used to train the prediction model in the present application considers the relative position relationships of the distributed wind power generators, the correlation between the physical environment dimension data and other dimension data in the wind power data is established, the utilization rate of data resources is improved, and the wind power prediction accuracy of the prediction model for wind power generators in different physical environments is improved.

[0057] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0058] The accompanying drawings, which are incorporated herein and constitute part of this specification, illustrate embodiments consistent with the application and, together with the description, further serve to explain the principles of the application.

[0059] Figure 1 is a flowchart of a distributed wind power prediction method provided by an embodiment of the application;

[0060] Figure 2 is a flowchart of preprocessing sample data in a distributed wind power prediction method provided by an embodiment of the application;

[0061] Figure 3 is a flowchart of building and training a prediction model according to the tensor graph and the edge weight matrix in a distributed wind power prediction method provided by an embodiment of the application;

[0062] Figure 4 is a Transformer model framework diagram in a distributed wind power prediction method provided by an embodiment of the application;

[0063] Figure 5 is a structural diagram of a distributed wind power prediction device provided by an embodiment of the application;

[0064] Figure 6 is a structural diagram of a distributed wind power prediction device provided by an embodiment of the application.

[0065] Reference signs: processor-21; memory-22; acquisition module-31; preprocessing module-32; data set construction module-33; matrix generation module-34; tensor graph construction module-35; model training module-36. DETAILED DESCRIPTION

[0066] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The following description is only exemplary and is not intended to represent all embodiments consistent with the present application. Rather, the following description is intended only to describe some embodiments consistent with the present application, as detailed in the appended claims.

[0067] A distributed wind power prediction method, with reference to Figure 1 , comprising:

[0068] S11: acquiring wind power data of a distributed wind turbine as sample data; wherein the wind power data comprises multiple dimensions;

[0069] The wind power data in the embodiment has the characteristics of multi-dimension, multi-source and multi-modal. The dimensions of the wind power data at least include geographical position data of each distributed wind turbine, terrain data where each distributed wind turbine is located, meteorological data (collected from a micro-meteorological station) and power data. The meteorological data at least includes wind speed data, wind direction data, temperature data and humidity data.

[0070] S12: pre-processing the sample data;

[0071] The wind power data in the embodiment has significant differences in numerical size and physical meaning, but there is a certain correlation, so data pre-processing is needed.

[0072] Specifically, the sample data is pre-processed, referring to Figure 2 , including:

[0073] S121: filling the missing values of the sample data;

[0074] S122: unifying the data sampling time of the meteorological data and the power data;

[0075] S123: performing flat data cleaning on the wind speed data based on a preset full load wind speed threshold;

[0076] S124: vectorizing the geographical position data and the terrain data of the distributed wind turbine;

[0077] S125: trigonometric functionizing the wind direction data;

[0078] S126: performing normalization processing on different types of sample data respectively.

[0079] In the embodiment, it is judged whether the sample data has missing values. If the sample data has missing values, the missing values are filled with the mean value of similar samples.

[0080] Since the meteorological data and the power data in the wind power data are time series data, the data sampling time of the meteorological data and the power data needs to be unified, and the preset sampling frequency is, for example, st times per day.

[0081] To enhance the data correlation, the minimum wind speed sw H of the full power output of the unit is selected, the minimum wind speed ensuring the full power output of all units is defined as the preset full load wind speed, the values exceeding the preset full load wind speed are flattened to the preset full load wind speed, and the values lower than the full load wind speed remain unchanged, as shown in the following formula, wherein sw is the wind speed data collected by the micro-meteorological station, and sw' is the pre-processed wind speed data.

[0082]

[0083] In this embodiment, the geographical position data (terrain elevation) of the distributed wind turbine is denoted as h, and the terrain data is encoded and denoted as g, and vectorization is performed.

[0084] The wind direction data is denoted as dis, and since the range of wind direction is 0°-360°, the wind direction data is trigonometrically processed to obtain the processed wind direction data dis'=(sindis, cosdis).

[0085] Finally, different types of wind power data are normalized respectively:

[0086]

[0087] wherein a represents the minimum existing unit of any type of data, a min represents the minimum value in the type of data, a max represents the maximum value in the type of data.

[0088] S13: Constructing a time series data set according to the preprocessed sample data;

[0089] including:

[0090] Constructing a time series data set X={X t |t∈[0,T]} wherein X t ={x 1,t ,L x i,t ,L x m,t}, x t ={x i,t |i∈[1,m]} X t represents the wind power data sequence at time t, x i,t represents the i-th wind power data element at time t, and there are m wind power data elements in total.

[0091] S14: Determining the relative position relationship of each distributed wind turbine according to the wind power data, and generating an adjacency matrix and an edge weight matrix according to the relative position relationship of each distributed wind turbine;

[0092] Generating an adjacency matrix according to the relative position relationship of each distributed wind turbine, including:

[0093] Calculating the element a ij in the adjacency matrix:

[0094]

[0095] wherein a ij represents the element in the adjacency matrix A, and D ijrepresents the distance between the distributed wind turbines i and j; σ and ε both represent the preset connectivity threshold of the graph;

[0096] The adjacency matrix is generated based on the elements in the adjacency matrix.

[0097] Preferably, σ and ε are set to 20 and 0.5 respectively.

[0098] According to the relative position relationship of each distributed wind turbine, an edge weight matrix is generated, including:

[0099] The weight between the distributed wind turbines i and j in the edge weight matrix is calculated as:

[0100] w i,j =f(Δ||dis' i,j ||,Δh' i,j ,Δ||g' i,j ||,d ij )

[0101] =1 / (αgΔ||dis' i,j ||+βgΔh' i,j +μgΔ||g' i,j ||+νgd ij +k)

[0102]

[0103] wherein w i,j represents the weight between the distributed wind turbines i and j in the edge weight matrix W, f(g) represents an initial weight calculation function, ||dis'|| represents the wind direction data norm after data preprocessing, Δ||dis' i,j || is the difference between the wind direction data norm values between the distributed wind turbines i and j after data preprocessing, Δh' i,j represents the difference in altitude between the distributed wind turbines i and j after data preprocessing, Δ||g' i,j || represents the terrain difference condition between the distributed wind turbines i and j, d i,j represents the straight-line distance between the distributed wind turbines i and j, α, β, μ, ν, k all represent preset function coefficients, and w′ i,j represents the weight between the distributed wind turbines i and j in the normalized edge weight matrix.

[0104] wherein α, β, μ, ν, k can be determined as optimal coefficients through actual examples.

[0105] S15: According to the number of distributed wind turbines, the adjacency matrix and the time series data set, a tensor graph is constructed;

[0106] The data is converted into a tensor graph form: G=(V, A, X), where V represents nodes, each distributed wind turbine is a node, and there are N nodes in total, A represents an adjacency matrix, and X represents a feature vector of the tensor graph, i.e., a time series data set.

[0107] S16: Building and training a prediction model according to the tensor graph and the edge weight matrix; the prediction model is used to output a prediction result according to the wind power data of the distributed wind turbine.

[0108] The distributed wind power prediction method in this embodiment includes: obtaining wind power data of distributed wind turbines as sample data, the wind power data including multiple dimensions. The sample data is preprocessed, a time series data set is constructed according to the preprocessed sample data, the relative position relationship of each distributed wind turbine is determined according to the wind power data, an adjacency matrix and an edge weight matrix are generated according to the relative position relationship of each distributed wind turbine, and a tensor graph is constructed according to the number of distributed wind turbines, the adjacency matrix and the time series data set. A prediction model is built and trained according to the tensor graph and the edge weight matrix, and the prediction model is used to output a prediction result according to the wind power data of the distributed wind turbine. Since the tensor graph used to train the prediction model in this embodiment considers the relative position relationship of each distributed wind turbine, the correlation between the physical environment dimension data and other dimension data in the wind power data is established, the utilization rate of data resources is improved, and the wind power prediction accuracy of the prediction model for wind turbines in different physical environments is improved.

[0109] The distributed wind power prediction method in some embodiments selects a Transformer model for the prediction model, the Transformer model framework is composed of an encoder and a decoder, and the components of the encoder and the decoder are attention mechanism combinations; the attention mechanism combination includes: a multi-head spatial attention module and a multi-head temporal attention module, the multi-head spatial attention module is used to learn node information in the spatial dimension, and the multi-head temporal attention module is used to learn node information in the time dimension.

[0110] In this embodiment, the entire distributed wind power prediction training process is mainly completed by the Transformer model. In this embodiment, in order to better mine the dynamic correlation of space and time, the multi-head attention mechanism is divided into a multi-head spatial attention module Space_multihead and a multi-head temporal attention module Temp_multihead, which are used to expand the model's ability to focus on different positions.

[0111] The specific attention module is described as follows:

[0112] Temp_multihead(Q, K, V) = Concat(t_head1, …, t_head h )WO

[0113]

[0114] Space_multihead(Q,K,V) = Concat(s_head1,...,s_head h )W O

[0115]

[0116] Q i = XW i Q ,K i = XW i K ,V i = XW i V

[0117] where t_head i and s_head i denote the temporal and spatial attention mechanisms, respectively, Q, K, V denote the query, key and value vectors, respectively, Concat() denotes the concatenation function of tensors, denotes the dimension of the key vector, softmax() denotes the normalization function, W O , W i Q , W i K , W i V denote the learnable weight parameters for the corresponding vectors.

[0118] Referring to Figure 4The encoder of the Transformer model consists of an input layer, a positional encoding layer, a self-attention layer, and two encoder layers. The input layer maps the input data to a high-order tensor through a fully connected network. At each time point, each kind of wind power data corresponds to a one-dimensional tensor. The order information in the time series data can be encoded using the positional encoding with sine and cosine functions by element-wise addition of the input vector and the positional encoding vector. The resulting vector is fed into the encoder layer. The two encoder layers each contain two sub-layers: an attention module (spatial attention module / temporal attention module) and a fully connected feedforward neural network sub-layer. The two encoder layers correspond to the multi-head spatial attention module Space_multihead and the multi-head temporal attention module Temp_multihead, respectively. The former is used to learn the node information in the spatial dimension, and the latter is used to learn the node information in the temporal dimension. As shown in the following formula, the output of the two-layer encoder layer is learned through a residual network, and the output result is normalized. Finally, the normalized result is fed back to the decoder by the last encoder.

[0119]

[0120] F=W n2 σ(W n1 X)

[0121] wherein Wg, σ are learnable parameters, F(g) is a linear mapping function for matching dimensions, L represents different module units (i.e. the temporal attention module or the spatial attention module in the encoding layer), X L represents the features of any unit L. X i is the input time series, W i is the weight learned by the network.

[0122] Referring to Figure 4 , the decoder of the Transformer model consists of an input layer, a decoder layer, and an output layer. The input of the decoder starts from the last data point input by the encoder. The input layer maps the decoder input to a high-order tensor. Two decoder layers each contain an attention sub-layer (spatial attention layer / temporal attention layer), an encoding-decoding attention layer, and a fully connected feedforward neural network sub-layer. The encoding-decoding attention layer is used to receive the vector feedback from the encoder. Similarly, the two decoder layers correspond to the multi-head spatial attention module Space_multihead and the multi-head temporal attention module Temp_multihead, respectively. After learning the output of the two-layer encoder layer through a residual network, the output result is normalized. Finally, the output of the last decoder layer is mapped to the target time series through the output layer.

[0123] In this embodiment, the information of the spatial dimension and the information of the time dimension in the corresponding encoder and decoder are associated and processed.

[0124] In this embodiment, the dynamic association of wind power data in the time-space dimension is explored to support decentralized wind power prediction.

[0125] Based on this, the time series data set is constructed according to the preprocessed sample data, which also includes:

[0126] Based on the preset time interval, a plurality of time series segments of different lengths are intercepted along the time axis as each periodic component; each periodic component is an integer multiple of the preset time interval;

[0127] The component time series data set corresponding to each periodic component is constructed.

[0128] For example, suppose the current time is t0, and the sampling time interval is T st , three time series segments of lengths T h , T d and T w are intercepted along the time axis as the prediction period component, the daily period component and the weekly period component; wherein T h , T d and T w are integer multiples of st.

[0129] The specific construction of the component time series data set corresponding to the three periodic components is as follows:

[0130] The prediction period component time series data set:

[0131] The daily period component time series data set is composed of fragments of the past few days, and has the same day attribute and time interval as the prediction period:

[0132]

[0133] The weekly period component time series data set is composed of fragments of the past few weeks, and has the same week attribute and time interval as the prediction period:

[0134]

[0135] Further, the prediction model is built and trained according to the tensor graph and the edge weight matrix, and the prediction model is built and trained according to the tensor graph and the edge weight matrix. Referring to Figure 3 , including:

[0136] S161: Model training according to the tensor graph and the edge weight matrix constructed according to each component time series data set;

[0137] S162: output the prediction results and weight parameters trained by each component time series data set;

[0138] S163: obtain the final prediction result according to the prediction results and weight parameters of each component time series data set.

[0139] The transformer has great potential in improving the prediction ability, but it is slightly inferior in predicting long time scales. Therefore, in the embodiment, χ h , χ d , χ w These three different component time series data sets are superimposed to improve the training efficiency and more effectively capture data information between different time scales. Since the influence weights of the three components of each node are different, a link weight matrix is introduced for model training in the embodiment, and the weight data is learned from historical data, and then the outputs of different components are further fused. The final prediction result after fusion is:

[0140]

[0141] is Hadamard product, W h , W d and W w are weight parameters learned from historical data, reflecting the influence degree of three component time series data sets on the prediction target. h is the prediction result of the prediction period component time series data set, °Y d is the prediction result of the day period component time series data set, °Y w is the prediction result of the week period component time series data set, and °Y is the final prediction result.

[0142] A distributed wind power prediction device, referring to Figure 5 , comprising:

[0143] a processor 21 and a memory 22;

[0144] The processor 21 and the memory 22 are connected through a communication bus;

[0145] The processor 21 is configured to call and execute the program stored in the memory 22;

[0146] The memory 22 is configured to store a program, and the program is at least used to execute the distributed wind power prediction method in any one of the above embodiments.

[0147] A distributed wind power prediction device, referring to Figure 6 , comprising:

[0148] ​The acquisition module 31 is configured to acquire wind power data of the distributed wind turbines as sample data; wherein the wind power data comprises multiple dimensions;

[0149] The preprocessing module 32 is configured to preprocess the sample data;

[0150] The data set construction module 33 is configured to construct a time series data set according to the preprocessed sample data;

[0151] The matrix generation module 34 is configured to determine relative position relationships of the distributed wind turbines according to the wind power data, and generate an adjacency matrix and an edge weight matrix according to the relative position relationships of the distributed wind turbines;

[0152] The tensor graph construction module 35 is configured to construct a tensor graph according to the number of the distributed wind turbines, the adjacency matrix and the time series data set;

[0153] The model training module 36 is configured to build and train a prediction model according to the tensor graph and the edge weight matrix; the prediction model is configured to output a prediction result according to the wind power data of the distributed wind turbines.

[0154] The dimensions of the wind power data at least include geographical position data of each of the distributed wind turbines, terrain data where each of the distributed wind turbines is located, meteorological data and power data; wherein the meteorological data at least includes wind speed data, wind direction data, temperature data and humidity data.

[0155] The prediction model is a Transformer model, and a framework of the Transformer model comprises an encoder and a decoder; components of the encoder and the decoder are attention mechanism combinations; the attention mechanism combinations comprise a multi-head spatial attention module and a multi-head temporal attention module; the multi-head spatial attention module is configured to learn node information in a spatial dimension, and the multi-head temporal attention module is configured to learn node information in a temporal dimension.

[0156] In some embodiments, the preprocessing module 32 of the distributed wind power prediction device is specifically configured to perform missing value filling on the sample data; unify data sampling times of the meteorological data and the power data; perform data cleaning on the wind speed data based on a preset full-load wind speed threshold; vectorize the geographical position data and the terrain data of the distributed wind turbines; perform trigonometric functionization on the wind direction data; and perform normalization processing on different types of the sample data respectively.

[0157] In some embodiments, the data set construction module 33 of the distributed wind power prediction device is specifically configured to construct a time series data set X = {X t |t∈[0,T]}; wherein X t={x 1,t ,L x i,t ,L x m,t}, x t ={x i,t |i∈[1,m]},X t Let x represent the wind power data sequence at time t. i,t Let represent the i-th wind power data element at time t, and there are a total of m wind power data elements.

[0158] The dataset construction module 33 is further used to extract multiple time series segments of different lengths along the time axis based on a preset time interval, as each periodic component; wherein each periodic component is an integer multiple of the preset time interval; and to construct the component time series dataset corresponding to each periodic component.

[0159] In some embodiments of the distributed wind power prediction device, the matrix generation module 34 is specifically used to calculate element a in the adjacency matrix. ij :

[0160]

[0161] Among them, a ij D represents the elements in the adjacency matrix A. ij σ represents the distance between distributed wind turbine units i and j; σ and ε both represent the preset connectivity threshold of the graph.

[0162] The adjacency matrix is ​​generated based on the elements in the adjacency matrix.

[0163] Matrix generation module 34 is further used to calculate the weights between distributed wind turbine units i and j in the edge weight matrix:

[0164] w i,j =f(Δ||dis') i,j ||,Δh' i,j ,Δ||g' i,j ||,d ij )

[0165] =1 / (αgΔ||dis') i,j ||+βgΔh' i,j +μgΔ||g' i,j ||+νgd ij +k)

[0166]

[0167] Among them, w i,jrepresents the weight between the distributed wind turbines i and j in the edge weight matrix W, f(g) represents an initial weight calculation function, ||dis|| represents the wind direction data norm after data preprocessing, Ah i,j represents the wind direction data norm value difference between the distributed wind turbines i and j after data preprocessing, Ah i,j represents the altitude difference between the distributed wind turbines i and j after data preprocessing, Ah i,j represents the terrain difference condition between the distributed wind turbines i and j, d i,j represents the straight-line distance between the distributed wind turbines i and j, a, b, m, n, k represent preset function coefficients, w i,j represents the weight between the distributed wind turbines i and j in the normalized edge weight matrix.

[0168] The distributed wind power prediction device in some embodiments, the model training module 36 is specifically configured to train the model according to the tensor graph constructed according to each component time series data set and the edge weight matrix; output the prediction result and the weight parameter trained by each component time series data set; obtain the final prediction result according to the prediction result and the weight parameter corresponding to each component time series data set.

[0169] It can be understood that the same or similar parts in each of the above embodiments can be mutually referred to, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0170] It should be noted that in the description of the present application, the terms "first", "second" and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance. In addition, in the description of the present application, unless otherwise specified, the meaning of "a plurality of" is at least two.

[0171] Any process or method descriptions in flow charts or described elsewhere herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for performing specific logic functions or steps in the process, and that the scope of preferred embodiments of the present application encompasses alterations, modifications, and variations of these code modules, segments, or portions of code that can be performed in either the order shown or in a different order, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those skilled in the art.

[0172] It should be understood that each of the elements of the present application can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies known in the art or their combinations can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0173] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment methods can be completed by programs instructing related hardware, and the programs can be stored in a computer readable storage medium, and when executed, include one or a combination of steps of the method embodiments.

[0174] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can be physically present separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be realized in the form of hardware or in the form of a software function module. The integrated module, if realized in the form of a software function module and sold or used as an independent product, can also be stored in a computer readable storage medium.

[0175] The above-mentioned storage medium can be a read-only memory, a magnetic disk or an optical disk, etc.

[0176] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0177] Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A distributed wind power prediction method, characterized by, The method comprises the following steps: acquiring wind power data of distributed wind turbines as sample data; wherein the wind power data comprises multiple dimensions; preprocessing the sample data; constructing a time series data set according to the preprocessed sample data; determining the relative position relationship of each distributed wind turbine according to the wind power data, and generating an adjacency matrix and an edge weight matrix according to the relative position relationship of each distributed wind turbine; constructing a tensor graph according to the number of distributed wind turbines, the adjacency matrix and the time series data set; building and training a prediction model according to the tensor graph and the edge weight matrix; the prediction model is used to output a prediction result according to the wind power data of the distributed wind turbines; the dimensions of the wind power data at least include: geographical position data of each distributed wind turbine, topographic data of each distributed wind turbine, meteorological data and power data; wherein the meteorological data at least includes: wind speed data, wind direction data, temperature data and humidity data.

2. The method of claim 1, wherein, The preprocessing of the sample data comprises: filling in missing values of the sample data; unifying the data sampling time of the meteorological data and the power data; flattening and cleaning the wind speed data based on a preset full load wind speed threshold; vectorizing the geographical position data and topographic data of the distributed wind turbines; triangular functionizing the wind direction data; respectively normalizing different types of sample data.

3. The method of claim 1, wherein, The construction of the time series data set according to the preprocessed sample data comprises: Constructing a time series dataset ,in , , express t Wind power data sequence at any given time. express t The first moment i There are a total of wind power data elements, and the total number of wind power data elements is... m indivual.

4. The method of claim 1, wherein, The construction of the time series data set according to the preprocessed sample data further comprises: based on a preset time interval, multiple time series segments of different lengths are intercepted along the time axis as each periodic component; wherein each periodic component is an integer multiple of the preset time interval; constructing a component time series data set corresponding to each periodic component.

5. The method of claim 1, wherein, The generation of the adjacency matrix according to the relative position relationship of each distributed wind turbine comprises: computing elements in the adjacency matrix : wherein, denotes an element in the adjacency matrix A, denotes a distributed wind turbine i and j a distance between; and both denote a predetermined connectivity threshold of a graph; generating the adjacency matrix based on the elements in the adjacency matrix.

6. The method of claim 2, wherein, The generation of the edge weight matrix according to the relative position relationship of each distributed wind turbine comprises: computing weights between distributed wind turbines in the edge weight matrix i and j ​ wherein, represents the weight between the distributed wind turbines i and j in the edge weight matrix W, represents an initial weight calculation function, represents the wind direction data norm after data preprocessing, is the wind direction data norm value difference between the distributed wind turbines i and j after data preprocessing, represents the altitude difference between the distributed wind turbines i and j after data preprocessing, represents the terrain difference condition between the distributed wind turbines i and j , represents the straight-line distance between the distributed wind turbines i and j , , , , , all represent preset function coefficients, represents the weight between the distributed wind turbines i and j in the normalized edge weight matrix.

7. The method of claim 4, wherein, The building and training of the prediction model according to the tensor graph and the edge weight matrix comprises: model training is performed according to the tensor graph constructed based on each component time series data set and the edge weight matrix; outputting the prediction results and weight parameters obtained by training each component time series data set; obtaining the final prediction result according to the prediction results and weight parameters corresponding to each component time series data set.

8. The method of claim 1, wherein, The prediction model is a Transformer model, and the Transformer model framework is composed of an encoder and a decoder; the components of the encoder and the decoder are attention mechanism combinations; the attention mechanism combination comprises a multi-head spatial attention module and a multi-head temporal attention module; the multi-head spatial attention module is used to learn the node information of the spatial dimension, and the multi-head temporal attention module is used to learn the node information of the time dimension.

9. A decentralized wind power prediction device, characterized in that The method comprises the following steps: a processor and a memory; The processor is connected with the memory through a communication bus; The processor is configured to call and execute a program stored in the memory; The memory is configured to store a program, and the program is used to execute the distributed wind power prediction method according to any one of claims 1-8.

10. A distributed wind power prediction device, characterized by Comprise: An acquisition module is configured to acquire wind power data of the distributed wind turbine as sample data; wherein the wind power data comprises multiple dimensions; A preprocessing module is configured to preprocess the sample data; A data set construction module is configured to construct a time series data set according to the preprocessed sample data; A matrix generation module is configured to determine the relative position relationship of each distributed wind turbine according to the wind power data, and generate an adjacency matrix and an edge weight matrix according to the relative position relationship of each distributed wind turbine; A tensor graph construction module is configured to construct a tensor graph according to the number of distributed wind turbines, the adjacency matrix and the time series data set; A model training module is configured to build and train a prediction model according to the tensor graph and the edge weight matrix; the prediction model is used to output a prediction result according to the wind power data of the distributed wind turbine.

Images