A method and system for ultra-short-term wind speed prediction

Abstract

The present invention proposes an ultra-short-term wind speed prediction method and system, including: establishing a long short-term memory neural network LSTM model based on improved adaptive noise complete set empirical mode decomposition ICEEMDAN and sparrow search algorithm SSA optimization. By using ICEEMDAN to decompose the original wind speed sequence into a series of components from high frequency to low frequency, the complexity of the wind speed is reduced. Wind speed sequences at different scales are obtained. On this basis, the optimized long short-term memory neural network is used for prediction, and the accuracy of wind speed prediction is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind speed prediction, and in particular relates to an ultra-short-term wind speed prediction method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] At present, wind power generation plays a vital role in the global renewable energy field. However, with the continuous expansion of wind power installed capacity, the volatility, intermittency and randomness of wind energy have brought challenges to the safe and stable operation of the power grid. In wind power generation, wind speed is one of the most important influencing factors. Therefore, accurate wind speed prediction plays an important role in the operating efficiency, grid stability, economic benefits and operation and maintenance of wind power systems. However, due to the volatility and complexity of wind speed, building a high-precision wind speed prediction model is still an unresolved problem.

[0004] Traditional methods basically predict wind speed directly, such as using the BP neural network optimized by the particle swarm algorithm PSO. The PSO algorithm is a swarm intelligence algorithm that simulates the collective behavior of organisms such as bird flocks or fish schools. Each particle represents a solution and searches for the optimal solution by constantly adjusting its position. Particles guide the search process by sharing information to better explore the search space. The PSO-BP algorithm combines these two methods. It uses the PSO algorithm to optimize the initial weights and biases of the neural network, and further optimizes and adjusts them through the back propagation algorithm. The PSO algorithm can help the BP algorithm avoid falling into the local optimal solution and speed up the convergence of the network. In addition, the RNN recurrent neural network is also commonly used, but these solutions are all trained by directly inputting the original wind speed data to obtain the prediction results.

[0005] In recent years, there have been some solutions that decompose the original wind speed data and then input it into the model for prediction, such as using variational mode decomposition (VMD) and empirical mode decomposition (EMD).

[0006] However, the above solution still has technical problems:

[0007] 1. It is difficult to select VMD decomposition parameters. The number of modes needs to be determined in advance, and the selected parameters have a great influence on the results. At the same time, VMD decomposition is very sensitive to noise and interference in the signal, and noise may be decomposed into additional modal components.

[0008] 2. EMD decomposition has a major disadvantage: modal aliasing, which makes it impossible to accurately decompose the signal into separate components, which may lead to inaccurate decomposition results and mutual interference between modes. Summary of the invention

[0009] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides an ultra-short-term wind speed prediction method, which greatly improves the accuracy of wind speed prediction.

[0010] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0011] In a first aspect, an ultra-short-term wind speed prediction method is disclosed, comprising:

[0012] Decompose the original wind speed data to obtain a series of wind speed sequences from high frequency to low frequency;

[0013] The decomposed wind speed series are predicted separately using the long short-term memory network (LSTM) model optimized by the Sparrow Search Algorithm (SSA);

[0014] The wind speed sequences predicted separately are superimposed and reconstructed to obtain the final prediction result.

[0015] As a further technical solution, when decomposing the original wind speed data, the improved adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN) algorithm is used to decompose the original wind speed data. The specific process is as follows:

[0016] Define w as the wind speed signal to be discretized and introduce E in the EMD algorithm k (·) and τ(·), where E k (·) is the k-order modal component of the original signal, τ(·) is used to generate the local mean of the signal, and λ (i) To indicate the Gaussian white noise to be added;

[0017] Construct a noise signal and obtain a set of residuals;

[0018] When k = 1, the first group of IMF modal components is calculated, and the residual is γ1 = (τ(w (i) )). Using the average value of the local mean γ1+α1E(λ (i) ) Calculate the second-order residual and IMF components;

[0019] When the kth order, the corresponding IMF components are obtained;

[0020] Until the decomposition is completed, all modal components and residuals are obtained.

[0021] As a further technical solution, the LSTM model optimized using SSA is as follows:

[0022] The SSA algorithm is used to optimize the three hyperparameters of LSTM and establish an ultra-short-term wind speed prediction model based on SSA-LSTM. The main process is as follows:

[0023] 1) Divide the wind speed data into training data and test data, and then decompose them using the ICEEMDAN method;

[0024] 2) Initialize LSTM model parameters;

[0025] 3) The SSA algorithm determines the maximum number of iterations, population size, and search boundary;

[0026] 4) Use the SSA algorithm to improve the LSTM model parameters to achieve the optimal value. This iterative process can improve the prediction accuracy of the model and thus improve the wind speed prediction.

[0027] 5) Use the optimized parameters to predict the decomposed test set wind speed series respectively.

[0028] 6) The prediction results are superimposed and reconstructed to obtain the final prediction results.

[0029] In a second aspect, an ultra-short-term wind speed prediction system is disclosed, comprising:

[0030] The decomposition module is configured to: decompose the original wind speed data;

[0031] The prediction module is configured to: use the LSTM model optimized by SSA to predict the decomposed wind speed series separately;

[0032] The reconstruction module is configured to: superimpose and reconstruct the wind speed sequences predicted separately to obtain the final prediction result.

[0033] One or more of the above technical solutions have the following beneficial effects:

[0034] The sub-technical scheme of this embodiment is aimed at the wind speed prediction in the wind power generation system, and a long short-term memory neural network LSTM model based on the improved adaptive noise complete set empirical mode decomposition ICEEMDAN and the sparrow search algorithm SSA optimization is established. By using ICEEMDAN to decompose the original wind speed sequence into a series of components from high frequency to low frequency, the complexity of the wind speed is reduced. Wind speed sequences at different scales are obtained. On this basis, the optimized long short-term memory neural network is used for prediction, and the accuracy of wind speed prediction is greatly improved.

[0035] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0037] Figure 1 This is a schematic diagram of the structure of a long short-term memory neural network according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the prediction model flow in an embodiment of the present invention;

[0039] Figure 3 Decomposition results of the embodiments of the present invention;

[0040] Figure 4 Schematic diagram of prediction results of an embodiment of the present invention. DETAILED DESCRIPTION

[0041] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0042] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0043] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0044] Embodiment 1

[0045] This embodiment discloses an ultra-short-term wind speed prediction method, comprising:

[0046] First, the ICEEMDAN algorithm is used to decompose the original wind speed data. Then, the LSTM model optimized by SSA is used to predict the decomposed wind speed sequences separately. Finally, the predicted wind speed sequences are superimposed and reconstructed to obtain the final prediction results.

[0047] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0048] ICEEMDAN:

[0049] Define w as the wind speed signal to be discretized. The E in the EMD algorithm is introduced. k (·) and τ(·). In the formula, E k (·) is the k-order modal component of the original signal, and τ(·) is used to generate the local mean of the signal. (i) To represent the Gaussian white noise to be added.

[0050] 1) Construct a noise signal

[0051] w (i) =w+α0E(λ (i) ) (1)

[0052] Among them, α0=υ0std(w(t)) / std(E(λ (i) )), represents the expected signal-to-noise ratio at the first decomposition, υ0 is the set white noise amplitude, and std represents the standard deviation. Then we get a set of residuals

[0053] γ1=(τ(w (i) )) (2)

[0054] 2) When k = 1, calculate the first group of IMF modal components

[0055] imf1=w-γ1 (3)

[0056] 3) Using the average value of the local mean γ1+α1E(λ (i) ) Calculate the second-order residual and IMF components

[0057] imf2=γ1-γ2=γ1-(τ(γ1+α1E(λ (i) ))) (4)

[0058] 4) When the kth order

[0059] γ k =(τ(γ k-1 +α k-1 E(λ (i) ))) (5)

[0060] The corresponding IMF components

[0061] imf k =γ k-1 -γ k (6)

[0062] 5) Until the decomposition is completed, all modal components and residuals are obtained. The components obtained in this part can be used as the decomposed wind speed sequence for the next step of prediction.

[0063] ICEEMDAN decomposition technology better reveals the structure and characteristics of signals through features such as multi-scale decomposition, adaptive noise suppression and physical interpretability.

[0064] 2. Sparrow Search Algorithm SSA

[0065] As an optimization algorithm, the sparrow search algorithm has the advantages of strong optimization ability and fast convergence speed. It can be used to optimize the hyperparameters of the deep learning model LSTM used below, making the wind speed predicted by LSTM more accurate.

[0066] Before introducing the sparrow search algorithm in detail, let's first introduce a few rules:

[0067] 1) Explorers in a population usually have high energy reserves and their mission is to find food-rich areas. They provide guidance to all followers about foraging areas and directions. The energy reserves of individual sparrows directly affect their adaptability in the algorithm.

[0068] 2) When a natural enemy appears, the sparrow will send out an alarm signal. If the alarm value exceeds the safety threshold, the explorer will guide the follower to other safe foraging areas.

[0069] 3) The algorithm assigns sparrows identities based on their ability to locate dominant food sources. Although the identities of sparrows change dynamically, the overall ratio of explorers to followers in the population remains constant.

[0070] 4) Sparrows with higher energy reserves act as explorers. In contrast, followers with lower energy reserves tend to look for food in different places to increase their energy levels.

[0071] 5) During foraging, followers will constantly track explorers with higher energy reserves. They may observe and compete with explorers to obtain more food resources, thereby increasing their own foraging rate.

[0072] 6) When a predator poses a threat, sparrows at the periphery of the population will quickly move to a safer area, while sparrows in the center will move randomly to deal with the threat.

[0073] By following these rules, the algorithm controls the sparrows' behavior and interactions during their foraging process.

[0074] In SSA, the best individual in the group gets food first during the search process. As an explorer, it can obtain a larger foraging search range than the follower. In each iteration, the explorer's position is updated as follows:

[0075]

[0076] In the above formula, X i,j It represents the position of the i-th sparrow in the j-th dimension. max Represents the maximum number of iterations of the algorithm. Parameter Q is a random number that follows a normal distribution. L is a column matrix where each element is 1. The range of α is (0,1). The ranges of the warning value R2 and the safety value ST are [0,1] and [0.5,1] respectively.

[0077] The above formula is one of the steps in the SSA principle of the optimization algorithm. The explorer searches for the optimal position globally, aiming to optimize the parameters of LSTM so that LSTM can predict wind speed more accurately.

[0078] When conditions indicate the absence of predators, it means that explorers can conduct a global search without taking immediate safety measures. However, when some sparrows detect the presence of predators, all sparrows must take appropriate actions to ensure their own safety.

[0079] As mentioned above, during the foraging process, some followers will constantly observe the explorer. Once the explorer finds a more favorable food source, these followers will abandon their existing positions and actively compete for the newly found food. If the competition is successful, they can immediately obtain the food. However, if they fail to obtain the food, they must continue to look for further foraging opportunities according to rule (4).

[0080] The position of the follower is updated according to the following formula:

[0081]

[0082] A + =A T (AA T ) -1 (9)

[0083] Among them, X c It is the best location for explorers, X w is the worst position in the area where the population is located. A is a column matrix with element size 1 or -1. This formula aims to optimize the parameters of LSTM through the process of global search for the optimal position by the explorer, so that LSTM can predict wind speed more accurately.

[0084] The algorithm assumes that a significant percentage of individuals in the population (ranging from 10% to 20%) are highly aware of the potential danger posed by predators. These individuals are placed at random positions in the population.

[0085]

[0086] Among them, X b Represents the global optimal position, α plays a crucial role in determining the magnitude of each move. It is calculated by drawing a random number from a normal distribution with mean 0 and variance 1, thus ensuring that the step size varies randomly, making the optimization process exploratory and adaptive. μ represents a random number less than or equal to 1. f i corresponds to the fitness value of each sparrow, and f u and f d Represent the best and worst fitness values ​​in the population respectively. ε is a constant that ensures that the denominator is not zero.

[0087] Long Short-Term Memory Neural Network LSTM:

[0088] As above Figure 1 As shown. In order to solve the problem that historical information in recurrent neural networks cannot be transmitted for a long time, LSTM introduces a gating mechanism (Gated Cell) to achieve the ability to learn long time series. LSTM consists of three gates: input gate, forget gate, and output gate. These gates allow LSTM to selectively process and retain relevant information in its memory cells. The input gate enables LSTM to determine which information is important. It selectively learns and stores valuable information based on the input received. The forget gate allows LSTM to decide which information should be forgotten or discarded from the storage cell. This gate helps prevent the accumulation of unnecessary or outdated information, allowing LSTM to focus on relevant data. Finally, the output gate controls the flow of information from the storage cell to the LSTM output. It adjusts the output based on LSTM's learning of the input sequence.

[0089] The model of the present invention, ICEEMDAN-SSA-LSTM:

[0090] The training effect of LSTM is mainly affected by three hyperparameters: the number of hidden layer units, the regularization coefficient, and the initial learning rate. In order to improve the accuracy of wind speed prediction using LSTM, the SSA algorithm is used to optimize the three hyperparameters of LSTM, and an ultra-short-term wind speed prediction model based on SSA-LSTM is established. Figure 2 The main process is as follows:

[0091] 1) The wind speed data is divided into training data and test data, and then the original sequence is decomposed using the ICEEMDAN method. The decomposition results are shown in Figure 3 .

[0092] 2) Initialize the LSTM model parameters, including learning rate, number of hidden layer nodes, and regularization parameters.

[0093] 3) Initialize the maximum number of iterations, population size and search boundary of SSA. The above parameters are given by empirical values, and the optimal values ​​are manually found through multiple attempts.

[0094] 4) Use the SSA algorithm to improve the LSTM model parameters. The SSA algorithm can continuously and randomly try the parameters of the LSTM algorithm, and update each time it finds a better parameter. This process continues until the number of iterations is reached or the accuracy requirements are met, so that the LSTM parameters reach the optimal value. This iterative process can improve the prediction accuracy of the model, thereby improving wind speed prediction.

[0095] 5) Input the wind speed sequences into the LSTM model respectively, and use the optimized LSTM model to predict the decomposed test set wind speed sequences respectively.

[0096] 6) Add the separately predicted wind speed sequences to obtain the final prediction result, see Appendix Figure 4 shown.

[0097] Compared with traditional methods, ICEEMDAN decomposes the wind speed sequence from high frequency to low frequency. This step extracts the local features and frequency components in the signal, which has a better prediction effect for random wind and sudden change wind. In addition, the LSTM neural network uses the SSA algorithm with better convergence performance for optimization, which improves the prediction accuracy.

[0098] Embodiment 2

[0099] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0100] Embodiment 3

[0101] The purpose of this embodiment is to provide a computer-readable storage medium.

[0102] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0103] Embodiment 4

[0104] The purpose of this embodiment is to provide an ultra-short-term wind speed prediction system, including:

[0105] The decomposition module is configured to: decompose the original wind speed data;

[0106] The prediction module is configured to: use the LSTM model optimized by SSA to predict the decomposed wind speed series separately;

[0107] The reconstruction module is configured to: superimpose and reconstruct the wind speed sequences predicted separately to obtain the final prediction result.

[0108] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0109] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0110] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for ultra-short-term wind speed prediction, characterized in that: include: The ICEEMDAN algorithm is used to decompose the original wind speed data to obtain a series of wind speed sequences from high frequency to low frequency. The specific process is as follows: definition To discretize the wind speed signal, the EMD algorithm is introduced. and , where is the k-order modal component of the original signal, Used to generate the local mean of the signal and use To indicate the Gaussian white noise to be added; Construct a noise signal and obtain a set of residuals; When k=1, the first group of IMF modal components is calculated using the average of the local means. Calculate the second-order residual and IMF components; When the kth order, the corresponding IMF components are obtained; Until the decomposition is completed, all modal components and residuals are obtained; the decomposed wind speed series are predicted separately using the LSTM model optimized by SSA; The wind speed sequences predicted separately are superimposed and reconstructed to obtain the final prediction result; After ICEEMDAN decomposition, a series of wind speed sequences from high frequency to low frequency are obtained, and the local characteristics and frequency components in the signal are extracted, which has a better prediction effect for random wind and mutation wind.

2. The ultra-short-term wind speed prediction method according to claim 1, characterized in that: Using SSA to optimize the LSTM model is as follows: The SSA algorithm is used to optimize the three hyperparameters of LSTM, and an ultra-short-term wind speed prediction model based on SSA-LSTM is established. The three hyperparameters are: the number of hidden layer units, the regularization coefficient and the initial learning rate.

3. The ultra-short-term wind speed prediction method according to claim 2, characterized in that: The main process of establishing an ultra-short-term wind speed prediction model based on SSA-LSTM is as follows: 1) The wind speed data is divided into training data and test data, and then decomposed using the ICEEMDAN method; 2) Initialize LSTM model parameters; 3) The SSA algorithm determines the maximum number of iterations, population size, and search boundary; 4) Use the SSA algorithm to improve the LSTM model parameters to achieve the optimal value; 5) Use the optimized parameters to predict the decomposed test set wind speed series respectively; 6) Superimpose and reconstruct the prediction results to obtain the final prediction results.

4. The ultra-short-term wind speed prediction method according to claim 1, characterized in that: LSTM consists of It consists of three gates: input gate, forget gate and output gate; The input gate enables LSTM to determine what information is important. It selectively learns and stores valuable information based on the input it receives. The forget gate allows the LSTM to decide which information should be forgotten or discarded from the memory cell; The output gate controls the flow of information from the memory cell to the output of the LSTM.

5. An ultra-short-term wind speed prediction system, characterized in that it includes: The decomposition module is configured to: use the ICEEMDAN algorithm to decompose the original wind speed data to obtain a series of wind speed sequences from high frequency to low frequency. The specific process is: definition To discretize the wind speed signal, the EMD algorithm is introduced. and , where is the k-order modal component of the original signal, Used to generate the local mean of the signal and use To indicate the Gaussian white noise to be added; Construct a noise signal and obtain a set of residuals; When k=1, the first group of IMF modal components is calculated using the average of the local means. Calculate the second-order residual and IMF components; When the kth order, the corresponding IMF components are obtained; Until the decomposition is completed, all modal components and residuals are obtained; the prediction module is configured to: use the LSTM model optimized by SSA to predict the decomposed wind speed series separately; The reconstruction module is configured to: superimpose and reconstruct the wind speed sequences predicted respectively to obtain a final prediction result; After ICEEMDAN decomposition, a series of wind speed sequences from high frequency to low frequency are obtained, and the local characteristics and frequency components in the signal are extracted, which has a better prediction effect for random wind and mutation wind.

6. The ultra-short-term wind speed prediction system according to claim 5, characterized in that: SSA is used to optimize the LSTM model in the prediction module, specifically: The SSA algorithm is used to optimize the three hyperparameters of LSTM, and an ultra-short-term wind speed prediction model based on SSA-LSTM is established. The three hyperparameters are: the number of hidden layer units, the regularization coefficient and the initial learning rate.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 4 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 4 are performed.

Images