Method and system for multi-step prediction of future wind speed based on automatic reservoir neural network

Abstract

The present invention discloses a method and system for multi-step prediction of future wind speed based on automatic reservoir neural network, which realizes accurate and fast multi-step prediction of future information, maintains high robustness to noise and system time variation and avoids overfitting problem. Its technical scheme is: for short-time high-dimensional wind speed data, based on delayed embedding theory, using space-time information conversion, using observed high-dimensional dynamics as reservoir, mapping high-dimensional wind speed data to future information of target variable, and automatic reservoir neural network realizes multi-step prediction of target variable by solving a pair of conjugate space-time information interaction equations.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method and system for multi-step prediction of future wind speed based on an automatic reservoir neural network. Background Art

[0002] Prediction of wind speed is always fascinating and attractive. However, due to the nonlinearity of wind speed system and insufficient information, multi-step prediction of nonlinear dynamical system such as wind speed based on short-term time series is a challenging task. Although many methods including statistical regression (e.g. ARIMA12, robust regression), exponential smoothing and machine learning (e.g. long-short term memory network: LSTM) have been applied to the predictability problem, most existing methods require sufficient observations or long-term time series, and thus cannot reliably predict the wind speed changes of short-term time series with small samples. On the other hand, neural networks including recurrent neural networks (RNN) and LSTM can theoretically learn the nonlinear dynamics of wind speed from observed data and have been used in time series prediction based on phase space embedding strategies. However, when only short-term time series are available for training the network, the above methods usually suffer from overfitting problems. In addition, training neural networks consumes a lot of time and computational resources, such as the large amount of feedback from neurons leading to gradient vanishing problems, which also hinders the application of traditional neural network methods in many practical systems.

[0003] At present, there are few prediction methods for short-term high-dimensional wind speed data, but the prediction of such short-term wind speed series is natural and important. First, short-term high-dimensional wind speed data is widely available in the real world. Second, due to the time-varying non-stationary characteristics of wind speed systems in the real world, the latest short-term series can more accurately describe the future time characteristics of complex systems than the long-term series containing distant historical information. Therefore, even if long-term data are measured, truly effective predictions mainly depend on their recent short-term data. Third, since the measurable short-term high-dimensional data has rich information, the dynamics of these variables are intertwined and can be used for prediction. In fact, assuming that the steady state is contained in a low-dimensional manifold, for many high-dimensional wind speed systems that meet practical conditions, the spatiotemporal information (STI) transformation can be theoretically obtained by the delayed embedding theory. This theory can transform the spatial information of high-dimensional data into the temporal information of any target variable. On the basis of the STI transformation, by constructing multiple STI mappings to form a step-by-step prediction of short-term high-dimensional time series, a random distribution embedding (RDE) framework is developed to predict the distribution of values. In addition, there are methods to perform multi-step-ahead predictions by adopting a multi-layer neural network as an STI transformation. However, for short-term high-dimensional wind speed forecasting, there are still two unresolved issues: one is the computational cost and the other is robustness. Summary of the invention

[0004] A brief summary of one or more aspects is given below to provide a basic understanding of these aspects. This summary is not an exhaustive overview of all conceived aspects, and is neither intended to identify the key or critical elements of all aspects nor to define the scope of any or all aspects. Its only purpose is to give some concepts of one or more aspects in a simplified form as a prelude to a more detailed description that will be given later.

[0005] The purpose of the present invention is to solve the above-mentioned problems and provide a method and system for multi-step prediction of future wind speed based on automatic reservoir neural network, which not only realizes accurate and fast multi-step prediction of future information, but also maintains high robustness to noise and system time variation and avoids overfitting problems.

[0006] The technical solution of the present invention is: the present invention discloses a method for multi-step prediction of future wind speed based on automatic reservoir neural network, the method comprising:

[0007] Step 1: According to the characteristics of the input wind speed data, construct short-term high-dimensional data, and determine the target variable to be predicted, the known time series length, and the length to be predicted from the short-term high-dimensional data;

[0008] Step 2: Construct a high-dimensional short time series matrix, a delay embedding matrix, and a spatiotemporal information STI conjugate equation, wherein the spatiotemporal information STI conjugate equation contains a coefficient matrix;

[0009] Step 3: Randomly set the weights of the neural network F, and update the coefficient matrix in the spatiotemporal information STI conjugate equation by using the neural network F as the reservoir for reservoir calculation;

[0010] Step 4: Based on the spatiotemporal information STI conjugate equation of ARNN constructed in step 2, the coefficient matrix in the spatiotemporal information STI conjugate equation is solved by using data encoding and decoding, and finally the predicted value of the target variable to be predicted is obtained.

[0011] According to an embodiment of the method for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, step 1 further comprises:

[0012] Given a high-dimensional wind speed time series with length m and dimension n The wind speed y of the target monitoring station to be predicted is x1,x2,…,x n The wind speed monitoring stations with similar geographical locations measure any one of the wind speeds, that is, y = x k , k represents the subscript of the target monitoring station, which is any one from 1 to n;

[0013] Select the wind speed monitoring station variable with the greatest correlation with the target variable y: tSelect relevant variables or eliminate irrelevant variables to improve the performance of ARNN. For a given high-dimensional wind speed time series X t , calculate the time series With {y 1 ,y 2 ,…,y m}, select the D variables {x1,x2,…,x D}(D≤n).

[0014] According to an embodiment of the method for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, step 2 further comprises:

[0015] For the high-dimensional short time series matrix X composed of D related variables as follows t

[0016]

[0017] Through the processing of the randomly given neural network F, the matrix X t Convert to dimensional variable F(X t )=F1(Xt),…,FD(Xt)′, and the following ARNN spatiotemporal information STI conjugate equation is obtained:

[0018]

[0019] Among them, Y L×m is the delay embedding matrix, I L×L is the identity matrix, the coefficient matrix is unknown, and the future information is the target variable y, that is, {y m+1 ,y m+2 ,…,y m+L-1};

[0020] Construct the delayed embedding matrix as follows:

[0021]

[0022] Among them, L is the number of delayed embeddings, and L-1 is the number of prediction steps.

[0023] According to an embodiment of the method for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, step 3 further comprises:

[0024] from Random Selection variables, solve the following equation:

[0025]

[0026]

[0027]

[0028] in, is the coefficient matrix A sub-matrix of is the coefficient matrix A submatrix of ;

[0029] Update the coefficient matrix by the following criteria

[0030] If the initial element b ij Empty, use the equation directly Solution Replace b ij ; If the initial element b ij Not empty, let

[0031] According to an embodiment of the method for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, step 4 further comprises:

[0032] Solve the determination coefficient matrices A and B: Repeat step 3 above and update the matrices iteratively When an iteration meets the set convergence condition, the coefficient matrix Finally, it was determined that according to the following formula, It is determined that:

[0033]

[0034] in, and [Y L×m |I L×L ] is the augmented matrix;

[0035] When the coefficient matrices A and B are known, solve the unknown part of the target variable y.

[0036] The present invention also discloses a system for multi-step prediction of future wind speed based on an automatic reservoir neural network, the system comprising:

[0037] The target variable construction module constructs short-term high-dimensional data according to the characteristics of the input wind speed data, and determines the target variable to be predicted, the known time series length, and the length to be predicted from the short-term high-dimensional data;

[0038] Conjugate equation construction module, which constructs high-dimensional short time series matrix, delay embedding matrix, and spatiotemporal information STI conjugate equation, where the spatiotemporal information STI conjugate equation contains a coefficient matrix;

[0039] The coefficient matrix update module randomly sets the weights of the neural network F and updates the coefficient matrix in the spatiotemporal information STI conjugate equation by using the neural network F as the reservoir for reservoir calculation;

[0040] The target variable prediction module, based on the constructed ARNN's spatiotemporal information STI conjugate equation, uses data encoding and decoding to solve the coefficient matrix in the spatiotemporal information STI conjugate equation, and finally obtains the predicted value of the target variable to be predicted.

[0041] According to an embodiment of the system for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, the target variable construction module is further configured as follows:

[0042] Given a high-dimensional wind speed time series with length m and dimension n The wind speed y of the target monitoring station to be predicted is x1,x2,…,x n The wind speed monitoring stations with similar geographical locations measure any one of the wind speeds, that is, y = x k , k represents the subscript of the target monitoring station, which is any one from 1 to n;

[0043] Select the wind speed monitoring station variable with the greatest correlation with the target variable y: t Select relevant variables or eliminate irrelevant variables to improve the performance of ARNN. For a given high-dimensional wind speed time series X t , calculate the time series With {y 1 ,y 2 ,…,y m}, select the D variables {x1,x2,…,x D}(D≤n).

[0044] According to an embodiment of the system for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, the conjugate equation building module is further configured as:

[0045] For the high-dimensional short time series matrix X composed of D related variables as follows t

[0046]

[0047] Through the processing of the randomly given neural network F, the matrix X t Convert to dimensional variable F(X t )=F1(Xt),…,FD(Xt)′, and the following ARNN spatiotemporal information STI conjugate equation is obtained:

[0048]

[0049] Among them, Y L×m is the delay embedding matrix, I L×L is the identity matrix, the coefficient matrix is unknown, and the future information is the target variable y, that is, {y m+1 ,y m+2 ,…,y m+L-1};

[0050] Construct the delayed embedding matrix as follows:

[0051]

[0052] Among them, L is the number of delayed embeddings, and L-1 is the number of prediction steps.

[0053] According to an embodiment of the system for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, the coefficient matrix updating module is further configured as follows:

[0054] from Random Selection variables, solve the following equation:

[0055]

[0056]

[0057]

[0058] in, is the coefficient matrix A sub-matrix of is the coefficient matrix A submatrix of ;

[0059] Update the coefficient matrix by the following criteria

[0060] If the initial element b ij Empty, use the equation directly Solution Replace b ij ; If the initial element b ij Not empty, let

[0061] According to an embodiment of the system for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention, the target variable prediction module is further configured as follows:

[0062] Solve to determine the coefficient matrices A and B: Repeat the above coefficient matrix update module processing and update the matrix in an iterative manner When an iteration meets the set convergence condition, the coefficient matrix Finally, it was determined that according to the following formula, It is determined that:

[0063]

[0064] in, and [Y L×m |I L×L ] is the augmented matrix;

[0065] When the coefficient matrices A and B are known, solve the unknown part of the target variable y.

[0066] Compared with the prior art, the present invention has the following beneficial effects: the method and system of the present invention are based on the delayed embedding theory, use the space-time information conversion, take the observed high-dimensional dynamics as the reservoir, map the high-dimensional wind speed data to the future information of the target variable, and accurately predict the target variable in multiple steps. The automatic reservoir neural network not only realizes the accurate and fast multi-step prediction of future information by solving a pair of conjugate space-time information interaction equations at the same time, but also maintains high robustness and avoids the overfitting problem. By taking the observed dynamics itself as the reservoir, the automatic reservoir neural network explores the inherent low-dimensional dynamics of the target system while ensuring the high efficiency of the reservoir calculation. In addition, a large number of calculations reflect that even when the wind speed data is subject to large noise interference and the system is time-varying, the calculation results of ARNN are still relatively stable. The automatic reservoir neural network has been successfully applied to real wind speed data sets and exhibited excellent prediction results, which shows that the algorithm has great potential in the development of artificial intelligence and machine learning.

[0067] In general, the present invention has the following characteristics and technical effects:

[0068] 1) The present invention is aimed at a segment of wind speed data measured by multiple monitoring stations in close geographical locations at the same time as the original input. Since the wind speed system in the real world has time-varying non-stationary characteristics, the latest short-term series can more accurately describe the future time characteristics of the complex system than the long-term series containing distant historical information.

[0069] 2) The present invention combines reservoir calculation reservoir structure and spatiotemporal information STI conversion, uses the dynamics of observed high-dimensional data as a repository, and utilizes the intrinsic dynamics of the observation / target system; through STI conversion, the spatial information of high-dimensional data is converted into the temporal / dynamic information of any target variable, thereby solving the problem of small sample size. The main form and conjugate form of the STI equation are used to encode and decode the temporal dynamics of high-dimensional wind speed data.

[0070] 3) The present invention avoids the problem of overfitting. Training traditional neural networks consumes a lot of time and computing resources. For example, a large amount of feedback from neurons leads to gradient vanishing, which makes it difficult to apply traditional neural network methods on a large scale in many practical systems. Based on the reservoir structure, ARNN only needs to train fewer parameters and therefore does not require too many samples.

[0071] 4) The multi-step future prediction method for wind speed at multiple monitoring stations provided by the present invention is accurate, robust and energy-saving. The present invention conducts experiments on time-varying systems under noise interference, proving that the ARNN method has a good prediction effect on wind speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The above features and advantages of the present invention can be better understood after reading the detailed description of the embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily drawn to scale, and components with similar related properties or features may have the same or similar reference numerals.

[0073] Figure 1 A flow chart of an embodiment of a method for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention is shown.

[0074] Figure 2 A schematic diagram showing an embodiment of a system for multi-step prediction of future wind speed based on an automatic reservoir neural network of the present invention. DETAILED DESCRIPTION

[0075] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Note that the aspects described below in conjunction with the accompanying drawings and specific embodiments are only exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0076] Figure 1 The flowchart of an embodiment of the method for multi-step prediction of future wind speed based on Auto-Reservoir Neural Network (ARNN) of the present invention is shown. Figure 1 , the implementation steps of the prediction method of this embodiment are described in detail as follows. For ease of understanding, in the specific description of each step of this embodiment, the wind speed measured by 155 wind speed monitoring stations in the Wakkanai area of ​​Japan is used as an example for explanation.

[0077] Step 1: According to the characteristics of the input wind speed data, construct short-time high-dimensional data, determine the target variable to be predicted, the known time series length, and the length to be predicted.

[0078] Given a high-dimensional wind speed time series with length m and dimension n The wind speed y of the target monitoring station to be predicted is x1,x2,…,x n The wind speed monitoring stations with similar geographical locations measure any one of the wind speeds, that is, y = x k , k represents the subscript of the target monitoring station, which is any one from 1 to n.

[0079] Select the wind speed monitoring station variables with the greatest correlation with the target variable y: Selecting relevant variables or eliminating irrelevant variables (not related to the target variable y) from the high-dimensional wind speed time series will significantly improve the performance of ARNN. For a given high-dimensional wind speed time series, calculate the time series With {y 1 ,y 2 ,…,y m}, select the D variables {x1,x2,…,x D}(D≤n).

[0080] In one example, a high-dimensional dynamical system with 155 dimensions is constructed: The system contains 155-dimensional time series variables Arbitrarily select one of the time series variables as the target variable y. P(t) is a parameter. The 155-dimensional variables of the coupled system and the target variable y have a large correlation. We take the known time series length m = 110, and the length to be predicted is L-1 = 45.

[0081] Step 2: Construct a high-dimensional short time series matrix, a delay embedding matrix, and a spatiotemporal information STI conjugate equation, wherein the spatiotemporal information STI conjugate equation contains a coefficient matrix.

[0082] The high-dimensional short time series matrix in step 2 represents the input; the delayed embedding matrix contains the future prediction interval of the target variable and represents the output; the spatiotemporal information STI conjugate equation builds a bridge between the two and realizes the conversion from time to space.

[0083] The process of establishing the ARNN spatiotemporal information STI conjugate equation is as follows.

[0084] The input is D related variables as follows (1), that is, a high-dimensional short time series matrix X t .

[0085]

[0086] Through the processing of the randomly given neural network F, X t Convert to dimensional variable F(X t )=F1(Xt),…,FD(Xt)′, so we can get the following ARNN spatiotemporal information STI conjugate equation:

[0087]

[0088] Among them, I L×L is the identity matrix. In the above equation (2), the coefficient matrix is unknown, the future information is y, that is, {y m+1 ,y m+2 ,…,y m+L-1}.

[0089] Construct the delayed embedding matrix Y as follows:

[0090]

[0091] In this step, L is the number of delayed embeddings, L-1 is the number of prediction steps, and the coefficient matrix and is set to an empty matrix (to be updated in subsequent steps 3 and 4).

[0092] In one example, a four-layer neural network F is randomly given. After the nonlinear transformation of F, the 155-dimensional time series variable X t Convert to Dimension (e.g. 100-dimensional) variable So we can get the following ARNN spatiotemporal information STI conjugate equation:

[0093]

[0094] Among them, I 46×46 is the identity matrix. In this step, A 46×100 and B 100×46 is set to an empty matrix.

[0095] Step 3: Randomly set the weights of the neural network F and update the coefficient matrix in the spatiotemporal information STI conjugate equation by taking the neural network F as the reservoir for reservoir calculation.

[0096] In step 3, the idea of ​​reservoir calculation is used to avoid fitting of large-scale parameters, making small sample prediction possible and thus reducing the computational time.

[0097] The coefficient matrix B is updated by the following dropout scheme. Random Selection variables. Solve the following equation:

[0098]

[0099]

[0100]

[0101] in, is the coefficient matrix A sub-matrix of is the coefficient matrix The coefficient matrix is ​​updated by the following criteria.

[0102] Guidelines Content:

[0103] If the initial element b ij If it is empty, directly use the solution of (4) Replace b ij ; If the initial element b ij Not empty, let Here, b ij is the (i,j)th element of matrix B. That is, the update method is:

[0104]

[0105] in, Yes ij The updated value, is the current value (the value before the update).

[0106] In one example, the matrix B is updated by a dropout scheme. From [F1(X t ) … F 100 (X t )] T Randomly select 30 variables. Solve the following equation

[0107]

[0108] in, is the coefficient matrix B 100×46 Update B by the following criteria: 100×46 .

[0109] If the initial element b ij If it is empty, directly use the solution of (4) Replace b ij ; If the initial element b ij Not empty, let Here, b ij is the (i,j)th element of matrix B. The update method is:

[0110]

[0111] in, Yes ij The updated value, is the current value (the value before the update).

[0112] Step 4: Based on the spatial-temporal information STI conjugate equation of ARNN constructed in step 2, the coefficient matrices A and B in the spatial-temporal information STI conjugate equation are solved by using data encoding and decoding, and finally the predicted value of the target variable is obtained.

[0113] The predicted value of the target variable in step 4 is obtained by solving the multivariate equation system of the coefficient matrix, which is simple, fast and effective.

[0114] Solve to determine matrices A and B. Repeat step 3 above and update the matrix iteratively After a sufficient number of iterations, the convergence condition is met. Finally, it is determined. According to the following formula (7), is determined.

[0115]

[0116] in, and [Y L×m |I L×L ] is the augmented matrix.

[0117] According to formula (2), When all are known, solve the unknown part of the target variable y {y m+1 ,y m +2 ,…,y m+L-1}, which is the wind speed forecast value of the target monitoring station in the next L-1 steps.

[0118] Solution The process is as follows:

[0119] Formula (4) is equivalent to the following matrix equation:

[0120]

[0121] Formula (8) is expanded as follows:

[0122]

[0123] Note that {y 1 ,y 2 ,…,y m} is a known sequence. For any s, the first m-L+1 equations of formula (9) contain L unknowns. When 2L-1≤m, the number of unknowns is not greater than the number of equations. According to the least squares method, the first m-L+1 equations can be used to solve the element

[0124] In one example, repeat step 3 to update the matrix B in an iterative manner. 100×46 =(bij ) 100×46 After a sufficient number of iterations, the convergence condition is met, B 100×46 Finally, it was determined. According to the following formula, A 46×100 =(a ij ) 46×100 is determined.

[0125] A 46×100 ·[F(X)|B 100×46 ]=[Y 46×50 |I 46×46 ],

[0126] where [F(X)|B 100×46 ] and [Y 46×50 |I 46×46 ] is the augmented matrix.

[0127] In A 46×100 ,B 100×46 When all are known, solve the unknown part of the target variable y {y 51 ,y 52 ,…,y 68}.

[0128] Figure 2 The principle of an embodiment of the system for multi-step prediction of future wind speed based on automatic reservoir neural network of the present invention is shown. Figure 2 The system of this embodiment includes: a target variable construction module, a conjugate equation construction module, a coefficient matrix update module, and a target variable prediction module.

[0129] The target variable construction module is used to construct short-term high-dimensional data according to the characteristics of the input wind speed data, and to determine the target variable to be predicted, the known time series length, and the length to be predicted from the short-term high-dimensional data.

[0130] The target variable building block is further configured as:

[0131] Given a high-dimensional wind speed time series with length m and dimension n The wind speed y of the target monitoring station to be predicted is x1,x2,…,x n The wind speed monitoring stations with similar geographical locations measure any one of the wind speeds, that is, y = x k , k represents the subscript of the target monitoring station, which is any one from 1 to n;

[0132] Select the wind speed monitoring station variable with the greatest correlation with the target variable y: t Select relevant variables or eliminate irrelevant variables to improve the performance of ARNN. For a given high-dimensional wind speed time series X t , calculate the time series With {y 1 ,y 2 ,…,y m}, select the D variables {x1,x2,…,x D}(D≤n).

[0133] The conjugate equation construction module is used to construct a high-dimensional short time series matrix, a delay embedding matrix, and a spatiotemporal information STI conjugate equation, wherein the spatiotemporal information STI conjugate equation contains a coefficient matrix.

[0134] The conjugate equation building block is further configured as:

[0135] For the high-dimensional short time series matrix X composed of D related variables as follows t

[0136]

[0137] Through the processing of the randomly given neural network F, the matrix X t Convert to dimensional variable F(X t )=F1(Xt),…,FD(Xt)′, and the following ARNN spatiotemporal information STI conjugate equation is obtained:

[0138]

[0139] Among them, I L×L is the identity matrix, the coefficient matrix is unknown, and the future information is the target variable y, that is, {y m+1 ,y m+2 ,…,y m+L-1};

[0140] Construct the delayed embedding matrix as follows:

[0141]

[0142] L is the number of delayed embeddings, and L-1 is the number of prediction steps.

[0143] The coefficient matrix update module is used to randomly set the weights of the neural network F, and to update the coefficient matrix in the spatiotemporal information STI conjugate equation by taking the neural network F as the reservoir for reservoir calculation.

[0144] The coefficient matrix update module is further configured as:

[0145] from Random Selection variables, solve the following equation:

[0146]

[0147]

[0148]

[0149] in, is the coefficient matrix A sub-matrix of is the coefficient matrix A submatrix of ;

[0150] Update the coefficient matrix by the following criteria

[0151] If the initial element b ij Empty, use the equation directly Solution Replace b ij ; If the initial element b ij Not empty, let

[0152] The target variable prediction module is used to solve the coefficient matrix in the spatiotemporal information STI conjugate equation based on the constructed ARNN, ​​using data encoding and decoding, and finally obtain the predicted value of the target variable to be predicted.

[0153] The target variable prediction module is further configured as:

[0154] Solve to determine the coefficient matrices A and B: Repeat the above coefficient matrix update module processing and update the matrix in an iterative manner When an iteration meets the set convergence condition, the coefficient matrix Finally, it was determined that according to the following formula, It is determined that:

[0155]

[0156] in, and [Y L×m |I L×L ] is the augmented matrix;

[0157] When the coefficient matrices A and B are known, solve the unknown part of the target variable y.

[0158] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0159] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0160] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0161] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0162] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media may be any available medium that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, wherein disk often reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0163] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for multi-step prediction of future wind speed based on automatic reservoir neural network, characterized in that: Methods include: Step 1: According to the characteristics of the input wind speed data, construct short-term high-dimensional data, and determine the target variable to be predicted, the known time series length, and the length to be predicted from the short-term high-dimensional data; Step 2: Construct a high-dimensional short time series matrix, a delay embedding matrix, and a spatiotemporal information STI conjugate equation, wherein the spatiotemporal information STI conjugate equation contains a coefficient matrix; Step 3: Randomly set the weights of the neural network F, and update the coefficient matrix in the spatiotemporal information STI conjugate equation by using the neural network F as the reservoir for reservoir calculation; Step 4: Based on the spatiotemporal information STI conjugate equation of ARNN constructed in step 2, the coefficient matrix in the spatiotemporal information STI conjugate equation is solved by using data encoding and decoding, and finally the predicted value of the target variable to be predicted is obtained; Wherein, step 1 further comprises: Given a high-dimensional wind speed time series with length m and dimension n The wind speed y of the target monitoring station to be predicted is x1,x2,…,x n The wind speed monitoring stations with similar geographical locations measure any one of the wind speeds, that is, y = x k , k represents the subscript of the target monitoring station, which is any one from 1 to n; Select the wind speed monitoring station variable with the greatest correlation with the target variable y: t Select relevant variables or eliminate irrelevant variables to improve the performance of ARNN. For a given high-dimensional wind speed time series X t , calculate the time series With {y 1 ,y 2 ,…,y m }, select the D variables {x1,x2,…,x D }, where D≤n; Wherein, step 2 further comprises: For the high-dimensional wind speed time series matrix X composed of D related variables as follows t Through the processing of the randomly given neural network F, the matrix X t Convert to Dimension variables The following ARNN spatiotemporal information STI conjugate equation is obtained: Among them, Y L×m is the delay embedding matrix, I L×L is the identity matrix, the coefficient matrix is unknown, and the future information is the wind speed y at the target monitoring station, that is, {y m+1 ,y m+2 ,…,y m+L-1 }; Construct the delayed embedding matrix as follows: Where L is the number of delayed embeddings, and L-1 is the number of prediction steps; Wherein, step 3 further comprises: from Randomly select k variables, where Solve the following equation: in, is the coefficient matrix A sub-matrix of is the coefficient matrix A submatrix of ; Update the coefficient matrix by the following criteria If the initial element b ij Empty, use the equation directly [F(X 1 )F(X 2 )…F(X m )] k×m Solution Replace b ij ; If the initial element b ij Not empty, let 2. The method for multi-step prediction of future wind speed based on automatic reservoir neural network according to claim 1 is characterized in that: Step 4 further includes: Solve the determination coefficient matrices A and B: Repeat step 3 above and update the matrices iteratively When an iteration meets the set convergence condition, the coefficient matrix Finally, it was determined that according to the following formula, It is determined that: in, and [Y L×m |I L×L ] is the augmented matrix; When the coefficient matrices A and B are known, solve the unknown part of the target variable y.

3. A system for multi-step prediction of future wind speed based on automatic reservoir neural network, characterized in that the system include: The target variable construction module constructs short-term high-dimensional data according to the characteristics of the input wind speed data, and determines the target variable to be predicted, the known time series length, and the length to be predicted from the short-term high-dimensional data; Conjugate equation construction module, which constructs high-dimensional short time series matrix, delay embedding matrix, and spatiotemporal information STI conjugate equation, where the spatiotemporal information STI conjugate equation contains a coefficient matrix; The coefficient matrix update module randomly sets the weights of the neural network F and updates the coefficient matrix in the spatiotemporal information STI conjugate equation by using the neural network F as the reservoir for reservoir calculation; The target variable prediction module, based on the constructed ARNN's spatiotemporal information STI conjugate equation, uses data encoding and decoding to solve the coefficient matrix in the spatiotemporal information STI conjugate equation, and finally obtains the predicted value of the target variable to be predicted; Among them, the target variable construction module is further configured as: Given a high-dimensional wind speed time series with length m and dimension n The wind speed y of the target monitoring station to be predicted is x1,x2,…,x n The wind speed monitoring stations with similar geographical locations measure any one of the wind speeds, that is, y = x k , k represents the subscript of the target monitoring station, which is any one from 1 to n; Select the wind speed monitoring station variable with the greatest correlation with the target variable y: t Select relevant variables or eliminate irrelevant variables to improve the performance of ARNN. For a given high-dimensional wind speed time series X t , calculate the time series With {y 1 ,y 2 ,…,y m }, select the D variables {x1,x2,…,x D }, where D≤n; Among them, the conjugate equation building block is further configured as: For the high-dimensional wind speed time series matrix X composed of D related variables as follows t Through the processing of the randomly given neural network F, the matrix X t Convert to Dimension variables The following ARNN spatiotemporal information STI conjugate equation is obtained: Among them, Y L×m is the delay embedding matrix, I L×L is the identity matrix, the coefficient matrix is unknown, and the future information is the wind speed y at the target monitoring station, that is, {y m+1 ,y m+2 ,…,y m+L-1 }; Construct the delayed embedding matrix as follows: Where L is the number of delayed embeddings, and L-1 is the number of prediction steps; Among them, the coefficient matrix update module is further configured as: from Randomly select k variables, where Solve the following equation: in, is the coefficient matrix A sub-matrix of is the coefficient matrix A submatrix of ; Update the coefficient matrix by the following criteria If the initial element b ij Empty, use the equation directly Solution Replace b ij ; If the initial element b ij Not empty, let 4. The system for multi-step prediction of future wind speed based on automatic reservoir neural network according to claim 3 is characterized in that: The target variable prediction module is further configured as: Solve to determine the coefficient matrices A and B: Repeat the above coefficient matrix update module processing and update the matrix in an iterative manner When an iteration meets the set convergence condition, the coefficient matrix Finally, it was determined that according to the following formula, It is determined that: in, and [Y L×m |I L×L ] is the augmented matrix; When the coefficient matrices A and B are known, solve the unknown part of the target variable y.

Images