Wind power plant regulation and control method and device based on tensor decomposition, terminal equipment and storage medium

The expansion of wind power data through the four-dimensional tensor model solves the problem that two-dimensional matrix is ​​difficult to capture high-dimensional correlation in the prior art, and improves the accuracy of wind power prediction and the accuracy of wind farm regulation.

CN119030035APending Publication Date: 2024-11-26GUANGDONG POWER GRID CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411112942.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing wind power prediction methods are based on a two-dimensional matrix, making it difficult to effectively capture the high-dimensional correlation of wind power data, resulting in low prediction accuracy and regulation accuracy.

Method used

The four-dimensional tensor model is adopted to expand the wind power data through time, day, month and year dimensions, build an initial wind power prediction tensor model, and generate a wind power prediction tensor model through iterative training, breaking through the two-dimensional matrix limitation.

Benefits of technology

The accuracy and accuracy of wind power prediction are improved, thereby improving the regulation accuracy of wind farms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119030035A_ABST
    Figure CN119030035A_ABST
Patent Text Reader

Abstract

The invention discloses a tensor decomposition-based wind power plant regulation and control method and device, terminal equipment and a storage medium, and the method comprises the steps: obtaining historical wind power data, and selecting to-be-predicted wind power data; constructing an initial wind power prediction tensor model according to the historical wind power data and to-be-predicted wind power data; wherein the initial wind power prediction tensor model is a four-dimensional tensor model and comprises a moment dimension, a day dimension, a month dimension and a year dimension; performing iterative training on the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, and generating a wind power prediction tensor model; wherein during each iteration training, the factor matrix of each dimension is updated; determining predicted wind power data corresponding to wind power data to be predicted according to the wind power prediction tensor model; and regulating and controlling the wind power plant based on the predicted wind power data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of power grid technology, and in particular to a wind farm control method, device, terminal equipment and storage medium based on tensor decomposition. Background Art

[0002] Renewable energy mainly includes wind energy, solar energy, hydropower and geothermal energy. Among these renewable energy sources, wind energy has attracted much attention due to its sustainability, cost-effectiveness and cleanliness. The generation of wind energy is closely related to the unevenness of the thermal and dynamic movement of air in nature. Therefore, the site selection of wind farms has a direct impact on the power generation efficiency, economic benefits and profitability of wind turbines. Several principles should be considered when selecting a site: first, choose a location with stable wind direction and convenient for the layout of wind turbines; second, the annual average wind speed should be higher than 5 meters per second; third, the wind power density should be greater than 150 watts per square meter; fourth, the annual and daily changes in wind speed should match the local power load curve; fifth, the seasonal and daily changes in wind speed should be as small as possible to reduce the impact of grid connection; finally, considering geographical factors such as natural disasters, select areas with fewer natural disasters. However, the operation of wind farms is still affected by extreme weather conditions such as meteorological factors, which increases the volatility and intermittency of wind power and brings challenges to the real-time balance of the power system.

[0003] Wind power forecasting is a key technology for wind farm management. It predicts future power generation by analyzing wind power data such as historical wind speed data, meteorological parameters, and wind farm operating status. The main forecasting methods include physical model method, statistical model method, machine learning method, hybrid model method, and time series analysis method. These methods mainly learn patterns from historical data to predict future wind power. Although these methods perform well in specific situations, they also have problems such as dependence on training data, complex models, and limited generalization ability. Specifically, traditional methods such as ARIMA and SVR are mostly based on one-dimensional or two-dimensional data structures, that is, based on daily dimension matrices, or based on daily dimension matrices and monthly dimension matrices to determine wind power data. It is difficult to effectively capture the inherent high-dimensional correlation of wind power data, resulting in limited prediction accuracy, and then the problem of low control accuracy when regulating wind farms based on the predicted wind power data. Therefore, there is an urgent need for a method that can break through the limitations of two-dimensional matrices to improve the precision and accuracy of wind power forecasting, thereby improving the accuracy of wind farm regulation. Summary of the invention

[0004] The embodiments of the present invention provide a wind farm control method, apparatus, terminal device and storage medium based on tensor decomposition, which can break through the limitations of the existing two-dimensional matrix of wind power prediction, expand the dimension of wind power data from the original two dimensions to four dimensions through a four-dimensional tensor model, thereby improving the precision and accuracy of wind power prediction, and further improving the accuracy of wind farm control when using the predicted wind power data to control the wind farm.

[0005] An embodiment of the present invention provides a wind farm control method based on tensor decomposition, comprising:

[0006] Obtain historical wind power data and select wind power data to be predicted;

[0007] An initial wind power prediction tensor model is constructed according to the historical wind power data and the wind power data to be predicted; wherein the initial wind power prediction tensor model is a four-dimensional tensor model, including a time dimension, a day dimension, a month dimension, and a year dimension;

[0008] Iteratively training the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, thereby generating a wind power prediction tensor model; wherein, in each iterative training, the factor matrix of each dimension is updated;

[0009] Determining predicted wind power data corresponding to the wind power data to be predicted according to the wind power prediction tensor model;

[0010] The wind farm is regulated based on the predicted wind power data.

[0011] Furthermore, before iteratively training the initial wind power prediction tensor model, it also includes:

[0012] Perform CP decomposition on the initial wind power prediction tensor model to obtain the factor matrix of each dimension;

[0013] The CP decomposition of the initial wind power prediction tensor model is as follows:

[0014]

[0015] Among them, X represents the initial wind power prediction tensor model; R is the CP rank; A represents the factor matrix of the moment dimension, with a size of I*R; B represents the factor matrix of the day dimension, with a size of J*R; C represents the factor matrix of the month dimension, with a size of K*R; D represents the factor matrix of the year dimension, with a size of L*R; refers to the vector outer product; a r is the factor of the factor matrix A, a r ∈R I ; b r is the factor of the factor matrix B, b r ∈R J ;c r is the factor of the factor matrix C, c r ∈R K ;d r is the factor of the factor matrix D, d r ∈R L .

[0016] Further, the initial wind power prediction tensor model is iteratively trained until the initial wind power prediction tensor model reaches a preset convergence condition, and a wind power prediction tensor model is generated, including:

[0017] Obtain the initial values ​​of the factor matrices of each dimension of the initial wind power prediction tensor model;

[0018] In each iterative training, a first update value of the moment dimension factor matrix in the current iteration is determined according to the initial value of each dimension factor matrix, and the initial value of the moment dimension factor matrix is ​​updated with the first update value to obtain an updated moment dimension factor matrix;

[0019] Determine a second update value of the day dimension factor matrix using the first update value and the initial values ​​of the factor matrices of the dimensions other than the time dimension, and update the initial value of the day dimension factor matrix using the second update value to obtain an updated day dimension factor matrix;

[0020] Determine a third update value of the monthly dimension factor matrix using the first update value, the second update value, the initial value of the monthly dimension factor matrix, and the initial value of the annual dimension factor matrix, and update the initial value of the monthly dimension factor matrix using the third update value to obtain an updated monthly dimension factor matrix;

[0021] Determine a fourth update value of the year dimension factor matrix using the first update value, the second update value, the third update value, and the initial value of the year dimension factor matrix, and update the initial value of the year dimension factor matrix using the fourth update value to obtain an updated year dimension factor matrix;

[0022] Determine whether the preset convergence condition is met according to the updated factor matrices of each dimension, and if so, stop iterative training to generate a wind power prediction model;

[0023] If not, the first update value, the second update value, the third update value and the fourth update value are respectively used as the initial values ​​of the moment dimension factor matrix, the day dimension factor matrix, the month dimension factor matrix and the year dimension factor matrix at the next iteration.

[0024] Furthermore, the preset convergence condition includes when the error function of the initial wind power prediction tensor model is minimum or when the maximum number of iterations is reached.

[0025] Based on the above method embodiment, the present invention provides a corresponding device embodiment;

[0026] An embodiment of the present invention provides a wind farm control device based on tensor decomposition, including: a data acquisition module, a tensor model construction module, a tensor model training module, a wind power data prediction module and a control module;

[0027] The data acquisition module is used to acquire historical wind power data and select wind power data to be predicted;

[0028] The tensor model building module is used to build an initial wind power prediction tensor model according to the historical wind power data and the wind power data to be predicted; wherein the initial wind power prediction tensor model is a four-dimensional tensor model, including a time dimension, a day dimension, a month dimension and a year dimension;

[0029] The tensor model training module is used to iteratively train the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, thereby generating a wind power prediction tensor model; wherein, during each iterative training, the factor matrix of each dimension is updated;

[0030] The wind power data prediction module is used to determine the predicted wind power data corresponding to the wind power data to be predicted according to the wind power prediction tensor model;

[0031] The control module is used to control the wind farm based on the predicted wind power data.

[0032] Further, a tensor decomposition module is included;

[0033] The tensor decomposition module is used to perform CP decomposition on the initial wind power prediction tensor model before iterative training on the initial wind power prediction tensor model to obtain factor matrices of various dimensions;

[0034] The CP decomposition of the initial wind power prediction tensor model is as follows:

[0035]

[0036] Among them, X represents the initial wind power prediction tensor model; R is the CP rank; A represents the factor matrix of the moment dimension, with a size of I*R; B represents the factor matrix of the day dimension, with a size of J*R; C represents the factor matrix of the month dimension, with a size of K*R; D represents the factor matrix of the year dimension, with a size of L*R; refers to the vector outer product; a r is the factor of the factor matrix A, a r ∈R I ; b r is the factor of the factor matrix B, b r ∈R J ;c r is the factor of the factor matrix C, c r ∈R K ;d r is the factor of the factor matrix D, d r ∈R L .

[0037] Further, the initial wind power prediction tensor model is iteratively trained until the initial wind power prediction tensor model reaches a preset convergence condition, and a wind power prediction tensor model is generated, including:

[0038] Obtain the initial values ​​of the factor matrices of each dimension of the initial wind power prediction tensor model;

[0039] In each iterative training, a first update value of the moment dimension factor matrix in the current iteration is determined according to the initial value of each dimension factor matrix, and the initial value of the moment dimension factor matrix is ​​updated with the first update value to obtain an updated moment dimension factor matrix;

[0040] Determine a second update value of the day dimension factor matrix using the first update value and the initial values ​​of the factor matrices of the dimensions other than the time dimension, and update the initial value of the day dimension factor matrix using the second update value to obtain an updated day dimension factor matrix;

[0041] Determine a third update value of the monthly dimension factor matrix using the first update value, the second update value, the initial value of the monthly dimension factor matrix, and the initial value of the annual dimension factor matrix, and update the initial value of the monthly dimension factor matrix using the third update value to obtain an updated monthly dimension factor matrix;

[0042] Determine a fourth update value of the year dimension factor matrix using the first update value, the second update value, the third update value, and the initial value of the year dimension factor matrix, and update the initial value of the year dimension factor matrix using the fourth update value to obtain an updated year dimension factor matrix;

[0043] Determine whether the preset convergence condition is met according to the updated factor matrices of each dimension, and if so, stop iterative training to generate a wind power prediction model;

[0044] If not, the first update value, the second update value, the third update value and the fourth update value are respectively used as the initial values ​​of the moment dimension factor matrix, the day dimension factor matrix, the month dimension factor matrix and the year dimension factor matrix at the next iteration.

[0045] Furthermore, the preset convergence condition includes when the error function of the initial wind power prediction tensor model is minimum or when the maximum number of iterations is reached.

[0046] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, a wind farm control method based on tensor decomposition as described in the above-mentioned embodiment of the invention is implemented.

[0047] Another embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the wind farm control method based on tensor decomposition described in the above-mentioned embodiment of the invention.

[0048] The following beneficial effects are achieved by implementing the present invention:

[0049] The present invention provides a wind farm control method, device, terminal equipment and storage medium based on tensor decomposition. The method acquires historical wind power data and selects wind power data to be predicted. According to the acquired historical wind power data and the wind power data to be predicted, a four-dimensional tensor model including time dimension, day dimension, month dimension and year dimension is constructed as an initial wind power prediction tensor model; the initial wind power model is iteratively trained until the initial wind power prediction tensor model reaches a preset convergence condition, a wind power prediction tensor model is generated, and the predicted wind power data corresponding to the wind power data to be predicted is determined according to the wind power prediction tensor model, and then the wind farm is controlled based on the predicted wind power data. By constructing a four-dimensional tensor model including time dimension, day dimension, month dimension and year dimension as the initial wind power prediction tensor model, the limitation of the two-dimensional matrix in the existing wind power prediction is broken through, and the dimension of the wind power data is expanded from the original two-dimensional to four-dimensional through the four-dimensional tensor model, so as to improve the precision and accuracy of wind power prediction, and further improve the accuracy of wind farm control when the predicted wind power data is used to control the wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of a wind farm control method based on tensor decomposition provided by an embodiment of the present invention.

[0051] Figure 2 This is a wind power thermal diagram corresponding to the existing matrix filling method provided by the present invention.

[0052] Figure 3 It is a schematic diagram of a four-dimensional tensor model provided by an embodiment of the present invention.

[0053] Figure 4 Schematic diagram of CP decomposition of a third-order tensor provided by an embodiment of the present invention

[0054] Figure 5 It is a structural schematic diagram of a wind farm control device based on tensor decomposition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] In order to better understand the solutions of the present invention, relevant terms involved in the present invention are explained in advance.

[0057] Tensor: A tensor is a concept that describes a multidimensional array in data. It has applications in many fields such as physics, engineering, and computer science. A tensor can have multiple dimensions, each dimension can be regarded as a direction of the data, and the order of the tensor is the number of these dimensions. Data arranged in the same direction is called a one-way array. For example, a row vector is a one-way array of data along the horizontal direction. A two-way array of data arranged in the horizontal and vertical directions constitutes a matrix, and a tensor is a form of a multi-way array. According to the form of composition from low to high dimensions, it can be found that a tensor is an extension of a matrix to a higher dimension. For example, a zero-order tensor is a scalar, generally represented by non-bold lowercase letters, that is, a scalar a; a first-order tensor is a row or column vector, represented by bold lowercase letters, such as vector x; a second-order tensor is a matrix, represented by non-bold uppercase letters, such as matrix A; third-order and higher orders are called high-order tensors, represented by bold uppercase letters, such as X, T, etc. Vector a∈R m The i-th element in is represented by a i Indicates that the matrix A∈R m*n The (i,j)th element in is represented by a i,j Indicates that, by analogy, the n-order tensor The first (i1,i2,...,i n ) elements can be represented by a i1,i2,..in To express.

[0058] The scheme of the present invention is described in detail below.

[0059] like Figure 1 As shown, a wind farm control method based on tensor decomposition is provided in one embodiment of the present invention, comprising:

[0060] Step S1: Obtain historical wind power data and select wind power data to be predicted;

[0061] Step S2: constructing an initial wind power prediction tensor model according to the historical wind power data and the wind power data to be predicted; wherein the initial wind power prediction tensor model is a four-dimensional tensor model, including a time dimension, a day dimension, a month dimension and a year dimension;

[0062] Step S3: iteratively training the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, and generating a wind power prediction tensor model; wherein, in each iterative training, the factor matrix of each dimension is updated;

[0063] Step S4: determining predicted wind power data corresponding to the wind power data to be predicted according to the wind power prediction tensor model;

[0064] Step S5: regulating the wind farm based on the predicted wind power data.

[0065] For step S1, it is first necessary to obtain historical wind power data and determine the wind power data to be predicted. Preferably, the historical wind power data obtained can be historical data that changes over time; further, in order to avoid the accumulation of data volume, which leads to increased training complexity, historical wind power data within a certain time period earlier than the earliest moment of the wind power data to be predicted can be selected, and after the prediction of the wind power data to be predicted is completed, the predicted wind power data is added as the historical wind power data, so as to facilitate the prediction of wind power data in the subsequent new time period, so that the historical wind power data can achieve a dynamic adjustment process.

[0066] Regarding steps S2-S3, before describing the solution of the present invention, it is necessary to introduce the concept of the existing traditional prediction method based on matrix completion.

[0067] like Figure 2 The figure shows a two-dimensional matrix heat map constructed from wind power data. In this figure, the filled area represents the historical wind power data interval, and the unfilled area represents the wind power data interval to be predicted. The prediction problem of the wind power data to be predicted can be transformed into the problem of filling the missing data in the matrix based on the known data in the matrix. Extending to the general case, it can be expressed in mathematical form as follows: Assume that there is a wind turbine power data matrix model X∈R (l*h+l*p)×n , where l represents the length of the time slice selected in the model, h and p represent the historical interval span and the forecast interval span respectively, and n represents the number of adjacent wind turbines. In this model, assuming that the current time is t i , the problem will be transformed into a known history interval [t i ,t i -a,t i -2a,...,t i -h*a], for the prediction interval [t i ,t i +a,t i +2a,...,t i+p*a] data. That is, the prediction problem of historical data for the prediction interval data is transformed into the problem of matrix completion of missing data. A large number of algorithms have been proposed for the study of filling in missing values ​​in matrices, such as the most commonly used filling algorithm based on SVD decomposition.

[0068] Based on the above-mentioned existing wind power prediction principle based on two-dimensional matrix, the present invention expands it into a wind power prediction tensor model filled with high-dimensional tensors. Figure 3 As shown, a four-dimensional tensor model including time dimension (Time), day dimension (Day), month dimension (Month) and year dimension (Year) provided by the present invention is used as the initial wind power prediction tensor model by the present invention.

[0069] Since the characteristics of real wind power data in real life are not only reflected in the time dimension correlation between future data and historical data, but also in the periodicity of wind power data in units of years, months and days, in the present invention, four modes of wind power data, namely, time, day, month and year features are extracted for modeling, and they are constructed as follows: Figure 3 The format shown is a four-dimensional tensor structure of Time*Day*Month*Year, and its corresponding mathematical form is as follows:

[0070] X∈R Time*Day*Month*Year ; (1)

[0071] The tensor-based prediction algorithm of the present invention is similar to the traditional matrix-based prediction algorithm in processing methods. Both use the historical wind power data in the model as known information, and regard the wind power data to be predicted as missing data. The prediction of the wind power data to be predicted, that is, the future wind power data, is achieved by filling the missing data in the model. Compared with the matrix prediction model, the present invention uses a four-dimensional tensor model as a wind power prediction model to break through the limitations of the two-dimensional matrix, showing higher flexibility, and can more effectively utilize the periodic characteristics of wind power data, thereby improving the precision and accuracy of wind power data prediction.

[0072] It should be noted that, for the general two-dimensional matrix completion algorithm, the fundamental problem is to solve the minimum value problem of the function in formula (2);

[0073]

[0074] In the above formula, function f is the error function, which is used to represent the original missing matrix and the decomposition model The error between ′ , B ′ represents the factor matrix in the two-dimensional matrix decomposition model; formulas (3) and (4) are respectively the matrix A′ , B ′ The decomposition model representation of; matrix W ′ Representation and missing matrix X ′ Weight matrices of the same size;

[0075]

[0076] Then the problem of filling missing values ​​will be transformed into the problem of solving the optimal error function. There are many optimization algorithms to solve formula (2), such as the commonly used EM (maximum expected value algorithm), etc. Its principle is the iterative process of multiple interpolation and optimization. The specific process is: first, the original missing matrix X ′ Initialize the missing values ​​in by inserting random numbers, the mode of elements in adjacent positions, or the average of the row and column, and initialize the factor matrix A ′ , B ′ , according to formula (3), we get the matrix, and then according to formula (4), we get the original missing matrix X ′ The known data in the matrix are recombined with the filled data in the matrix. The decomposition and filling process is repeated until the error is less than the set threshold or the maximum number of iterations is reached.

[0077] Based on the above matrix missing value filling algorithm, its principle is also applicable to the filling of high-order tensors, the main difference is that the decomposition model is more complicated. Applying this principle to the four-dimensional initial wind power prediction tensor model, the process of decomposing and filling it is actually the process of generating a wind power prediction tensor model based on iterative training of the initial wind power prediction tensor model. When the wind power prediction model is generated, that is, when the filling is completed, the corresponding wind power data to be predicted has been filled, and the predicted wind power data corresponding to the wind power data to be predicted can be determined according to the wind power prediction tensor model. The process of decomposing and filling the tensor model is described in detail below:

[0078] The present invention solves the initial wind power prediction tensor model based on the iterative least squares method of CP decomposition. For CP decomposition, a tensor is decomposed into the sum of multiple single-rank tensors. For example, given a 3rd-order tensor X∈R I*J*K , the CP decomposition diagram of the tensor is as follows Figure 4 shown.

[0079] Will Figure 4 The CP decomposition process is formulated as:

[0080]

[0081] Based on the expansion of the above third-order tensor, the fourth-order tensor representation of the present invention can be obtained.

[0082] In a preferred embodiment, before iteratively training the initial wind power prediction tensor model, it also includes: performing CP decomposition on the initial wind power prediction tensor model to obtain factor matrices of various dimensions;

[0083] The CP decomposition of the initial wind power prediction tensor model is as follows:

[0084]

[0085] Where X represents the initial wind power prediction tensor model; R is a positive integer, called the CP rank; It is an abbreviation for the CP decomposition model; A represents the factor matrix of the moment dimension, with a size of I*R; B represents the factor matrix of the day dimension, with a size of J*R; C represents the factor matrix of the month dimension, with a size of K*R; D represents the factor matrix of the year dimension, with a size of L*R; refers to the vector outer product; a r is the factor of the factor matrix A, a r ∈R I ; b r is the factor of the factor matrix B, b r ∈R J ;c r is the factor of the factor matrix C, c r ∈R K ;d r is the factor of the factor matrix D, d r ∈R L .

[0086] Expand the CP decomposition of formula (6), that is, the element value of tensor X and factor a r , b r , c r , d r The corresponding relationship between the element values ​​is expressed as follows:

[0087] x i,j,k,l =a i,r b j,r c k,r d l,r ; (7)

[0088] Therefore, the problem of how to fill the original tensor with the CP decomposition model is transformed into the error minimization problem of the following formula;

[0089]

[0090] That is, find the appropriate parameters A, B, C, D and R, and obtain the tensor to be completed according to the CP decomposition model formula Minimizing the 2-norm of the error between X and the original tensor X turns the problem into a typical numerical optimization problem.

[0091] Then the CP decomposition model of the fourth-order initial wind power prediction tensor model is: The corresponding error function is:

[0092]

[0093] In the CP decomposition model, the most important and difficult problem in determining the parameters A, B, C, D and R is the determination of the CP rank R of the tensor, because the solution of the rank of the tensor is an NP-hard problem, and currently no algorithm can obtain a uniquely determined rank of the tensor. In real life, tensors composed of real data are noisy, and only through a process of continuous trial and error can the rank R suitable for the original tensor be found. This means that there will be errors between the padded tensor obtained by the CP decomposition model and the original tensor. This error is inevitable and is called the error coefficient. In the present invention, the R value is set by human experience.

[0094] The parameters to be solved are A, B, C, and D. There are many algorithms that can be used to optimize these parameters. In this embodiment, the simplest and most effective method is selected to solve them, namely, the iterative least squares method based on CP decomposition, referred to as CP-ALS.

[0095] In a preferred embodiment, the initial wind power prediction tensor model is iteratively trained until the initial wind power prediction tensor model reaches a preset convergence condition to generate a wind power prediction tensor model, including: obtaining the initial value of the factor matrix of each dimension of the initial wind power prediction tensor model; in each iterative training, determining the first update value of the moment dimension factor matrix in the current iteration according to the initial value of the factor matrix of each dimension, and updating the initial value of the moment dimension factor matrix with the first update value to obtain an updated moment dimension factor matrix; determining the second update value of the day dimension factor matrix with the first update value and the initial values ​​of the factor matrices of the remaining dimensions except the time dimension, and updating the initial value of the day dimension factor matrix with the second update value to obtain an updated day dimension factor matrix; and The initial value of the factor matrix and the initial value of the annual dimension factor matrix determine the third update value of the monthly dimension factor matrix, and the initial value of the monthly dimension factor matrix is ​​updated with the third update value to obtain an updated monthly dimension factor matrix; the fourth update value of the annual dimension factor matrix is ​​determined with the first update value, the second update value, the third update value and the initial value of the annual dimension factor matrix, and the initial value of the annual dimension factor matrix is ​​updated with the fourth update value to obtain an updated annual dimension factor matrix; according to the updated dimension factor matrices, it is determined whether the preset convergence condition is met, and if so, the iterative training is stopped to generate a wind power prediction model; if not, the first update value, the second update value, the third update value and the fourth update value are used as the initial values ​​of the moment dimension factor matrix, the day dimension factor matrix, the month dimension factor matrix and the year dimension factor matrix at the next iteration respectively.

[0096] Specifically, firstly, a random initial value is given to each dimension factor matrix, that is, an initial value is given to A, B, C, and D respectively. After the initial value is determined, iterative training begins.

[0097] In each iteration, first, the new moment dimension factor matrix is ​​calculated using the initial values ​​of A, B, C, and D and formula (10): The moment dimension factor matrix A is updated with the value of the new moment dimension factor matrix as the first update value, and the updated moment dimension factor matrix A is obtained.

[0098]

[0099] Using the updated moment dimension factor matrix The initial values ​​of the factor matrices of the dimensions other than the time dimension and formula (11) are used to calculate the new daily dimension factor matrix The new daily dimension factor matrix value is used as the second update value to update the daily dimension factor matrix B to obtain the updated

[0100]

[0101] Using the updated moment dimension factor matrix Day dimension factor matrix The initial value of the monthly dimension factor matrix C, the initial value of the annual dimension factor matrix D and formula (12) are used to calculate the new monthly dimension factor matrix The monthly dimension factor matrix C is updated with the value of the new monthly dimension factor matrix as the third update value mentioned above, and the updated monthly dimension factor matrix C is obtained.

[0102]

[0103] Using the updated moment dimension factor matrix Day dimension factor matrix Month dimension factor matrix The initial value of the annual dimension factor matrix D and formula (13) are used to calculate the new annual dimension factor matrix The annual dimension factor matrix D is updated with the value of the new annual dimension factor matrix as the fourth update value, and the updated annual dimension factor matrix is ​​obtained.

[0104]

[0105] In the above formulas (10)-(13), represents the pseudo-inverse of the matrix, T represents the transpose of the matrix, X (1) , X (2) , X (3) , X (4) They represent the mode-1 matrixization, mode-2 matrixization, mode-3 matrixization, and mode-4 matrixization of the tensor X. For an N-order tensor X, the process of expanding it according to a certain dimension is called the n-dimensional matrixization of the tensor. The matrixization result is represented by X (n) express.

[0106] Determine whether the preset convergence condition is met based on the updated factor matrices of each dimension.

[0107] In a preferred embodiment, the preset convergence condition includes when the error function of the initial wind power prediction tensor model is minimum or when the maximum number of iterations is reached.

[0108] The calculation of the error function is shown in formula (14). When the result of the formula reaches convergence, that is, when the error is less than a preset threshold, the preset convergence condition is met.

[0109]

[0110] If the preset convergence conditions are met, the iterative training is stopped and the wind power prediction model is generated. If the preset convergence conditions are not met, the updated time dimension factor matrix is ​​used. As the moment dimension factor matrix A of the next iteration, the updated moment dimension factor matrix As the moment dimension factor matrix B of the next iteration, the updated moment dimension factor matrix As the moment dimension factor matrix C of the next iteration, the updated moment dimension factor matrix The above iterative process is repeated as the moment dimension factor matrix D of the next iteration.

[0111] The specific expansion method for the above formulas (10)-(13) is: for the n-order tensor X, the element x i_1,i_2,...,i_n will be converted to a mode-n matrix X through the mapping relationship (n) The element x i_n,j , where j is defined as follows:

[0112]

[0113] In the above formulas (10)-(13), ⊙ represents the Khatri-Rao product of the matrix. Assume that the matrix X∈R I*K , Y∈R J*K , the size of X⊙Y is IJ*K, and the calculation form is:

[0114]

[0115] In the above formula, represents the Kronecker product of matrices, assuming that matrix A∈R I*J , B∈R K*L , The size is IK*JL, and the calculation form is:

[0116]

[0117] In the above formula (14), Represents the square of the Frobenius norm of the tensor. Assuming there is an n-order tensor X, whose Frobenius norm is defined as the arithmetic square root of the sum of the squares of all elements in the tensor, there is the following calculation formula:

[0118]

[0119] For step S4, the process of iteratively training the initial wind power prediction tensor model through the above method is actually the process of completing the missing data, that is, the prediction process of the wind power data to be predicted; therefore, when the wind power prediction tensor model is generated, that is, when the training is completed, the data corresponding to the wind power data to be predicted in the model has been filled in, and the corresponding predicted wind power data can be obtained directly from the wind power prediction tensor model.

[0120] In step S5, the predicted wind power data obtained from the wind power prediction tensor model is provided to the wind farm operation and maintenance personnel, so that the operation and maintenance personnel can more accurately control the wind farm based on the predicted wind power data.

[0121] Based on the above method embodiment, the present invention provides a corresponding device embodiment.

[0122] like Figure 5 As shown, an embodiment of the present invention provides a wind farm control device based on tensor decomposition, including: a data acquisition module, a tensor model construction module, a tensor model training module, a wind power data prediction module and a control module;

[0123] The data acquisition module is used to acquire historical wind power data and select wind power data to be predicted;

[0124] The tensor model building module is used to build an initial wind power prediction tensor model according to the historical wind power data and the wind power data to be predicted; wherein the initial wind power prediction tensor model is a four-dimensional tensor model, including a time dimension, a day dimension, a month dimension and a year dimension;

[0125] The tensor model training module is used to iteratively train the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, thereby generating a wind power prediction tensor model; wherein, during each iterative training, the factor matrix of each dimension is updated;

[0126] The wind power data prediction module is used to determine the predicted wind power data corresponding to the wind power data to be predicted according to the wind power prediction tensor model;

[0127] The control module is used to control the wind farm based on the predicted wind power data.

[0128] In a preferred embodiment, it also includes a tensor decomposition module;

[0129] The tensor decomposition module is used to perform CP decomposition on the initial wind power prediction tensor model before iterative training on the initial wind power prediction tensor model to obtain factor matrices of various dimensions;

[0130] The CP decomposition of the initial wind power prediction tensor model is as follows:

[0131]

[0132] Among them, X represents the initial wind power prediction tensor model; R is the CP rank; A represents the factor matrix of the moment dimension, with a size of I*R; B represents the factor matrix of the day dimension, with a size of J*R; C represents the factor matrix of the month dimension, with a size of K*R; D represents the factor matrix of the year dimension, with a size of L*R; refers to the vector outer product; a r is the factor of the factor matrix A, a r ∈R I ; b r is the factor of the factor matrix B, b r ∈R J ;c r is the factor of the factor matrix C, c r ∈R K ;d r is the factor of the factor matrix D, d r ∈R L .

[0133] In a preferred embodiment, the initial wind power prediction tensor model is iteratively trained until the initial wind power prediction tensor model reaches a preset convergence condition, and a wind power prediction tensor model is generated, including:

[0134] Obtain the initial values ​​of the factor matrices of each dimension of the initial wind power prediction tensor model;

[0135] In each iterative training, a first update value of the moment dimension factor matrix in the current iteration is determined according to the initial value of each dimension factor matrix, and the initial value of the moment dimension factor matrix is ​​updated with the first update value to obtain an updated moment dimension factor matrix;

[0136] Determine a second update value of the day dimension factor matrix using the first update value and the initial values ​​of the factor matrices of the dimensions other than the time dimension, and update the initial value of the day dimension factor matrix using the second update value to obtain an updated day dimension factor matrix;

[0137] Determine a third update value of the monthly dimension factor matrix using the first update value, the second update value, the initial value of the monthly dimension factor matrix, and the initial value of the annual dimension factor matrix, and update the initial value of the monthly dimension factor matrix using the third update value to obtain an updated monthly dimension factor matrix;

[0138] Determine a fourth update value of the year dimension factor matrix using the first update value, the second update value, the third update value, and the initial value of the year dimension factor matrix, and update the initial value of the year dimension factor matrix using the fourth update value to obtain an updated year dimension factor matrix;

[0139] Determine whether the preset convergence condition is met according to the updated factor matrices of each dimension, and if so, stop iterative training to generate a wind power prediction model;

[0140] If not, the first update value, the second update value, the third update value and the fourth update value are respectively used as the initial values ​​of the moment dimension factor matrix, the day dimension factor matrix, the month dimension factor matrix and the year dimension factor matrix at the next iteration.

[0141] In a preferred embodiment, the preset convergence condition includes when the error function of the initial wind power prediction tensor model is minimum or when the maximum number of iterations is reached.

[0142] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0143] Those skilled in the art can clearly understand that for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0144] Based on the above method item embodiments, the present invention provides corresponding terminal device item embodiments.

[0145] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a wind farm control method based on tensor decomposition as described in any one of the present invention is implemented.

[0146] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0147] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0148] The memory can be used to store the computer program, and the processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0149] Based on the above method item embodiments, the present invention provides a corresponding storage medium item embodiment.

[0150] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a wind farm control method based on tensor decomposition as described in any one of the present inventions.

[0151] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium.

[0152] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A wind farm control method based on tensor decomposition, characterized in that: include: Obtain historical wind power data and select wind power data to be predicted; An initial wind power prediction tensor model is constructed according to the historical wind power data and the wind power data to be predicted; wherein the initial wind power prediction tensor model is a four-dimensional tensor model, including a time dimension, a day dimension, a month dimension, and a year dimension; Iteratively training the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, thereby generating a wind power prediction tensor model; wherein, in each iterative training, the factor matrix of each dimension is updated; Determining predicted wind power data corresponding to the wind power data to be predicted according to the wind power prediction tensor model; The wind farm is regulated based on the predicted wind power data.

2. A wind farm control method based on tensor decomposition according to claim 1, characterized in that: Before iteratively training the initial wind power prediction tensor model, the method further includes: Perform CP decomposition on the initial wind power prediction tensor model to obtain the factor matrix of each dimension; The CP decomposition of the initial wind power prediction tensor model is as follows: Among them, X represents the initial wind power prediction tensor model; R is the CP rank; A represents the factor matrix of the moment dimension, with a size of I*R; B represents the factor matrix of the day dimension, with a size of J*R; C represents the factor matrix of the month dimension, with a size of K*R; D represents the factor matrix of the year dimension, with a size of L*R; ° refers to the vector outer product; a r is the factor of the factor matrix A, a r ∈R I ; b r is the factor of the factor matrix B, b r ∈R J ;c r is the factor of the factor matrix C, c r ∈R K ;d r is the factor of the factor matrix D, d r ∈R L .

3. A wind farm control method based on tensor decomposition as claimed in claim 2, characterized in that: Iteratively training the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, and generating a wind power prediction tensor model, including: Obtain the initial values ​​of the factor matrices of each dimension of the initial wind power prediction tensor model; In each iterative training, a first update value of the moment dimension factor matrix in the current iteration is determined according to the initial value of each dimension factor matrix, and the initial value of the moment dimension factor matrix is ​​updated with the first update value to obtain an updated moment dimension factor matrix; Determine a second update value of the day dimension factor matrix using the first update value and the initial values ​​of the factor matrices of the dimensions other than the time dimension, and update the initial value of the day dimension factor matrix using the second update value to obtain an updated day dimension factor matrix; Determine a third update value of the monthly dimension factor matrix using the first update value, the second update value, the initial value of the monthly dimension factor matrix, and the initial value of the annual dimension factor matrix, and update the initial value of the monthly dimension factor matrix using the third update value to obtain an updated monthly dimension factor matrix; Determine a fourth update value of the year dimension factor matrix using the first update value, the second update value, the third update value, and the initial value of the year dimension factor matrix, and update the initial value of the year dimension factor matrix using the fourth update value to obtain an updated year dimension factor matrix; Determine whether the preset convergence condition is met according to the updated factor matrices of each dimension, and if so, stop iterative training to generate a wind power prediction model; If not, the first update value, the second update value, the third update value and the fourth update value are respectively used as the initial values ​​of the moment dimension factor matrix, the day dimension factor matrix, the month dimension factor matrix and the year dimension factor matrix at the next iteration.

4. A wind farm control method based on tensor decomposition as claimed in claim 3, characterized in that: The preset convergence condition includes when the error function of the initial wind power prediction tensor model is minimum or when the maximum number of iterations is reached.

5. A wind farm control device based on tensor decomposition, characterized in that: include: Data acquisition module, tensor model building module, tensor model training module, wind power data prediction module and control module; The data acquisition module is used to acquire historical wind power data and select wind power data to be predicted; The tensor model building module is used to build an initial wind power prediction tensor model according to the historical wind power data and the wind power data to be predicted; wherein the initial wind power prediction tensor model is a four-dimensional tensor model, including a time dimension, a day dimension, a month dimension and a year dimension; The tensor model training module is used to iteratively train the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, thereby generating a wind power prediction tensor model; wherein, during each iterative training, the factor matrix of each dimension is updated; The wind power data prediction module is used to determine the predicted wind power data corresponding to the wind power data to be predicted according to the wind power prediction tensor model; The control module is used to control the wind farm based on the predicted wind power data.

6. A wind farm control device based on tensor decomposition as claimed in claim 5, characterized in that: Also includes tensor decomposition module; The tensor decomposition module is used to perform CP decomposition on the initial wind power prediction tensor model before iterative training on the initial wind power prediction tensor model to obtain factor matrices of various dimensions; The CP decomposition of the initial wind power prediction tensor model is as follows: Among them, X represents the initial wind power prediction tensor model; R is the CP rank; A represents the factor matrix of the moment dimension, with a size of I*R; B represents the factor matrix of the day dimension, with a size of J*R; C represents the factor matrix of the month dimension, with a size of K*R; D represents the factor matrix of the year dimension, with a size of L*R; refers to the vector outer product; a r is the factor of the factor matrix A, a r ∈R I ; b r is the factor of the factor matrix B, b r ∈R J ;c r is the factor of the factor matrix C, c r ∈R K ;d r is the factor of the factor matrix D, d r ∈R L .

7. A wind farm control device based on tensor decomposition according to claim 6, characterized in that: Iteratively training the initial wind power prediction tensor model until the initial wind power prediction tensor model reaches a preset convergence condition, and generating a wind power prediction tensor model, including: Obtain the initial values ​​of the factor matrices of each dimension of the initial wind power prediction tensor model; In each iterative training, a first update value of the moment dimension factor matrix in the current iteration is determined according to the initial value of each dimension factor matrix, and the initial value of the moment dimension factor matrix is ​​updated with the first update value to obtain an updated moment dimension factor matrix; Determine a second update value of the day dimension factor matrix using the first update value and the initial values ​​of the factor matrices of the dimensions other than the time dimension, and update the initial value of the day dimension factor matrix using the second update value to obtain an updated day dimension factor matrix; Determine a third update value of the monthly dimension factor matrix using the first update value, the second update value, the initial value of the monthly dimension factor matrix, and the initial value of the annual dimension factor matrix, and update the initial value of the monthly dimension factor matrix using the third update value to obtain an updated monthly dimension factor matrix; Determine a fourth update value of the year dimension factor matrix using the first update value, the second update value, the third update value, and the initial value of the year dimension factor matrix, and update the initial value of the year dimension factor matrix using the fourth update value to obtain an updated year dimension factor matrix; Determine whether the preset convergence condition is met according to the updated factor matrices of each dimension, and if so, stop iterative training to generate a wind power prediction model; If not, the first update value, the second update value, the third update value and the fourth update value are respectively used as the initial values ​​of the moment dimension factor matrix, the day dimension factor matrix, the month dimension factor matrix and the year dimension factor matrix at the next iteration.

8. A wind farm control device based on tensor decomposition as claimed in claim 7, characterized in that: The preset convergence condition includes when the error function of the initial wind power prediction tensor model is minimum or when the maximum number of iterations is reached.

9. A terminal device, characterized in that: The invention comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a wind farm control method based on tensor decomposition as claimed in any one of claims 1 to 4 is implemented.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a wind farm control method based on tensor decomposition as claimed in any one of claims 1 to 4.

Citation Information

Cited By

  • Power control method and device of wind power plant, terminal equipment and storage medium

    CN119695857A