A short-term wind power forecasting method

Abstract

The present invention discloses a method for predicting short-term wind power, wherein the first historical wind power data set is subjected to a multidimensional variational mode decomposition model to obtain an intrinsic mode function; the optimal hyperparameter combination is obtained based on the model performance evaluation of the de-redundancy model; the intrinsic mode function is de-redundant based on the optimal hyperparameter combination to obtain a target intrinsic mode function; a second historical wind power data set represented in the frequency domain is obtained based on the first historical wind power data set; an adaptive previous preset number of cycles and corresponding wind power data are obtained from the second historical wind power data set; the corresponding wind power data is decomposed into seasonal components, trend components and residual components, and weighted summation is performed to obtain a target seasonal component, a target residual component and a target trend component; the target seasonal component, the target residual component, the target trend component and the target intrinsic mode function are fused and the predicted short-term wind power is obtained through the model. The present invention improves the accuracy of short-term wind power prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and time series prediction, and relates to but is not limited to a method for predicting short-term wind power. Background Art

[0002] As wind power accounts for an increasing proportion of the global energy structure, how to accurately predict short-term wind power has become an important issue for optimizing wind farm management, power system dispatching, and power market transactions. The short-term wind power is measured in hours, such as the wind power forecast after 1 hour. Since wind power is affected by many factors, including short-term weather changes, air pressure, wind speed, and long-term seasonal changes, traditional machine learning methods have difficulty capturing complex nonlinear patterns and have low prediction accuracy over long time spans and in complex scenarios. Therefore, prediction methods based on deep learning have become a research hotspot.

[0003] Among the related technologies, the hybrid model prediction technology based on deep learning is mainly used. The typical solution is to train on the hybrid model architecture, extract the time series characteristics of wind power through the hybrid model, and then use the fully connected layer for the final prediction. However, wind power is affected by many factors, including short-term fluctuations and long-term seasonal changes, and it is difficult to deal with these different levels of influence at the same time. In addition, wind power is affected by many factors, including short-term weather changes, air pressure, wind speed and other factors. The existing prediction methods cannot effectively extract the deep coupling relationship between wind power sequences and meteorological factors, and thus cannot accurately predict short-term wind power.

[0004] Therefore, how to efficiently and accurately predict short-term wind power has become an urgent problem to be solved. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method for predicting short-term wind power, which at least solves the problem that the accuracy of short-term wind power prediction in related technologies is not high.

[0006] According to a first aspect of an embodiment of the present invention, a method for predicting short-term wind power is provided, comprising:

[0007] Inputting the first historical wind power data set into a multidimensional variational modal decomposition model to obtain a preset number of intrinsic mode functions; each intrinsic mode function corresponds to a component of different frequency and scale in the first historical wind power data set;

[0008] Obtaining an optimal hyperparameter combination from different hyperparameter combinations based on the model performance evaluation of the de-redundancy model; and obtaining a target intrinsic mode function after removing redundant intrinsic mode functions based on the optimal hyperparameter combination;

[0009] Convert the first historical wind power data set from the time domain to the frequency domain to obtain a second historical wind power data set represented in the frequency domain; and obtain an adaptive wind power data set corresponding to a preset number of previous cycles and each cycle from the second historical wind power data set;

[0010] Decomposing the wind power data corresponding to each cycle into a seasonal component, a trend component and a residual component; and performing weighted summation on the seasonal component, the trend component and the residual component to obtain a target seasonal component, a target residual component and a target trend component;

[0011] The target seasonal component, the target residual component, the target trend component and the target intrinsic mode function are fused to obtain fused features; and the fused features are input into a long sequence information perception model to obtain predicted short-term wind power.

[0012] According to a second aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, wherein the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect.

[0013] According to a third aspect of an embodiment of the present invention, there is provided a computer storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the first aspect is implemented.

[0014] According to the solution provided by an embodiment of the present invention, a first historical wind power data set is input into a multidimensional variational modal decomposition model to obtain a preset number of intrinsic mode functions; each intrinsic mode function corresponds to a component of different frequencies and scales in the first historical wind power data set; an optimal hyperparameter combination is obtained from different hyperparameter combinations based on a model performance evaluation of a de-redundancy model; and a target intrinsic mode function after removing redundant intrinsic mode functions is obtained based on the optimal hyperparameter combination; the first historical wind power data set is converted from a time domain to a frequency domain to obtain a second historical wind power data set represented in the frequency domain; and from the second historical wind power data set, a frequency domain representation of the first historical wind power data set is obtained. Centrally acquire the adaptive pre-set number of cycles and the corresponding wind power data sets for each cycle; decompose the wind power data corresponding to each cycle into seasonal components, trend components and residual components; and perform weighted summation on the seasonal components, trend components and residual components to obtain the target seasonal components, target residual components and target trend components; fuse the target seasonal components, target residual components, target trend components and target intrinsic mode functions to obtain the fused features; and input the fused features into the long sequence information perception model to obtain the predicted short-term wind power. In this process, the advantages of two decomposition technologies, namely, the multidimensional variational mode decomposition model and the decomposition of the wind power data corresponding to each cycle, are utilized to comprehensively analyze the historical wind power data sets from the perspectives of frequency domain and time domain, which not only improves the modeling capability of the complex time series mode of wind power, but also can more accurately analyze the deep coupling relationship between wind power and influencing factors, providing more refined feature support for short-term wind power prediction, thereby significantly improving the prediction accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0016] Figure 1 A schematic diagram of a flow chart of a short-term wind power prediction method provided by an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0020] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0021] Those skilled in the art will appreciate that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the field to which the embodiments of the present invention belong. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0022] Figure 1 A schematic flow chart of a short-term wind power prediction method provided in an embodiment of the present invention. The short-term wind power prediction method provided in an embodiment of the present invention can be executed by an electronic device, such as a computer, a server, etc.

[0023] like Figure 1 As shown, a short-term wind power prediction method includes:

[0024] S101. Input a first historical wind power data set into a multidimensional variational modal decomposition model to obtain a preset number of intrinsic mode functions; each intrinsic mode function corresponds to a component of different frequency and scale in the first historical wind power data set.

[0025] In an embodiment of the present invention, the first historical wind power data set may be a data set collected in hours. For example, when predicting wind power 1 hour in the future, the first historical wind power data set may be data collected 1 hour ago. The multidimensional variational mode decomposition model is used to extract components of different frequencies and scales in the first historical wind power data set, reflecting the local characteristics and change trends of the wind power data. That is, its main goal is to extract a predefined preset number (K) of intrinsic mode functions (IMF) from the first historical wind power data set containing C data channels. Assume that the mathematical expressions of the first historical wind power data set X and the decomposed IMF are as shown in the following formulas (1) and (2):

[0026] (1)

[0027] (2)

[0028] In the above formulas (1) and (2), C represents the number of data channels of the first historical wind power data set, K represents the number of IMFs decomposed under each channel, t represents the sampling time of the first historical wind power data set, and there are N sampling points in total. is the value of the cth data channel at time t. IMF represents the set of intrinsic mode functions, which is obtained by performing variational mode decomposition on the first historical wind power data set X. is the value of the cth data channel of the kth IMF at time t.

[0029] Furthermore, the first historical wind power data set is input into the multidimensional variational mode decomposition model, and a preset number of intrinsic mode functions can be obtained by combining formulas (1) and (2), wherein each intrinsic mode function corresponds to a component of different frequency and scale in the first historical wind power data set.

[0030] S102. Model performance evaluation based on the de-redundancy model obtains the optimal hyperparameter combination from different hyperparameter combinations; and based on the optimal hyperparameter combination, obtains the target intrinsic mode function after removing redundant intrinsic mode functions.

[0031] In an embodiment of the present invention, a de-redundancy model is used to remove redundant intrinsic mode functions containing a large amount of repeated information from a preset number of intrinsic mode functions to obtain a target intrinsic mode function. The de-redundancy model may be an elastic network. Among possible different hyperparameter combinations, an optimal hyperparameter combination can be obtained from possible different hyperparameter combinations based on the model performance evaluation of the de-redundancy model by means of a grid search, and further based on the optimal hyperparameter combination, a redundancy elimination operation is performed in a preset number of intrinsic mode functions to obtain a target intrinsic mode function.

[0032] S103, converting the first historical wind power data set from the time domain to the frequency domain to obtain a second historical initial wind power data set represented in the frequency domain; and obtaining an adaptive previous preset number of cycles and wind power data corresponding to each cycle from the second historical wind power data set.

[0033] Fast Fourier Transform (FFT) is an efficient algorithm of Discrete Fourier Transform (DFT) used to convert a time series from the time domain to the frequency domain. Through FFT, the different frequency components contained in the time series can be decomposed and the intensity of each frequency component can be understood.

[0034] In an embodiment of the present invention, the first historical wind power data set is converted from the time domain to the frequency domain by fast Fourier transform to obtain a second historical wind power data set represented in the frequency domain. This process automatically identifies the main cycle by fast Fourier transform, avoids the limitations of human settings, and can dynamically adapt to seasonal changes in the first historical wind power data set. After obtaining the second historical wind power data set, the adaptive first preset number (first k) cycles are calculated according to the following formulas (3), (4) and (5):

[0035] (3)

[0036] (4)

[0037] (5)

[0038] In the above equations (3), (4) and (5), the FFT function and the Amp function represent the calculation of FFT and amplitude values, and A represents the calculated amplitude of each frequency. Considering the sparsity of the frequency domain and avoiding the noise caused by meaningless high frequencies, only the first k amplitude values ​​are selected and the most important frequencies are obtained. , these selected frequencies also correspond to k period lengths ,in is the calculation period, represents the target wind power in the second historical wind power data set, T represents the length of the target wind power, and f * is a wildcard, which means that all frequencies f are selected within a certain range, ranging from 1 to [T / 2], where T is the length of the second historical wind power data set. is the frequency represented by the i-th amplitude value, Indicates rounding up, that is, for T and The value is rounded up.

[0039] Furthermore, wind power data corresponding to each period is obtained from the second historical wind power data set.

[0040] S104, decomposing the wind power data corresponding to each cycle into a seasonal component, a trend component and a residual component; and performing weighted summation on the seasonal component, the trend component and the residual component to obtain a target seasonal component, a target residual component and a target trend component;

[0041] In some embodiments of the present invention, each cycle has corresponding wind power data, and a multidimensional seasonal decomposition method is applied to decompose the wind power data corresponding to each cycle into seasonal components, trend components and residual components. This process enables the multi-scale smoothing technology to effectively extract multiple seasonal components, improve the accuracy of trend and residual separation, and avoid the information loss problem that may occur in traditional methods.

[0042] Furthermore, the obtained seasonal components, trend components and residual components can be weighted and summed respectively, that is, the seasonal components, trend components and residual components of all cycles are weighted and combined to obtain the target seasonal component, target residual component and target trend component.

[0043] S105, fusing the target seasonal component, the target residual component, the target trend component and the target intrinsic mode function to obtain fused features; and inputting the fused features into the long sequence information perception model to obtain the predicted short-term wind power.

[0044] In some embodiments of the present invention, the target seasonal component, the target residual component, the target trend component and the target intrinsic mode function are fused according to the channel to obtain the fused features. The features obtained after the two data decomposition techniques are fused, which can fully mine and utilize the multi-dimensional time-frequency features in the wind power data and dynamically capture the complex seasonal and nonlinear changes.

[0045] Furthermore, the fused features are input into the long sequence information perception model (Informer model) to obtain the predicted short-term wind power, where the short-term wind power is the wind power in hours.

[0046] Among them, in order to further reduce unnecessary computational overhead and accurately capture long-term dependencies in the sequence, the Informer model was selected as the predictor for point prediction. Informer effectively solves the high complexity problem faced by traditional Transformer when processing long sequences through an improved attention mechanism (sparse self-attention mechanism for long sequences). Its core advantage lies in the Encoder-Decoder architecture, in which the encoder uses a long sequence sparse self-attention mechanism to reduce computational complexity and memory usage, and effectively processes long sequence data. The decoder generates a weighted attention combination based on future time steps to predict wind power. The specific process is:

[0047] (1) Mapping the time series of fused features to a high-dimensional space is usually done through an encoder module. In order to capture temporal information, the fused features will contain time encoding (such as periodic encoding, representing information such as hours and dates).

[0048] (2) In the encoder, the fused features are processed through a long sequence sparse self-attention mechanism through a hybrid modeling of global and local information. This mechanism can capture the dependencies between different time steps in the data sequence. The long sequence sparse self-attention mechanism allows the model to learn different subspace features in parallel, enhancing the model's ability to model long sequences. In this process, the model calculates the query, key, and value of the input sequence, and then calculates the weights between each time step through the long sequence sparse self-attention mechanism, thereby weighted merging information.

[0049] (3) After each layer of long sequence sparse self-attention module, the data is further processed by a feed-forward neural network to enhance the expressive power. The network usually includes one or more fully connected layers.

[0050] (4) In the decoder, the Informer model continues to use the long sequence sparse self-attention mechanism to process the features of the encoder output and generate short-term wind power through the time step of target prediction.

[0051] It can be understood that in an embodiment of the present invention, the first historical wind power data set is input into a multidimensional variational mode decomposition model to obtain a preset number of intrinsic mode functions; each intrinsic mode function corresponds to a component of different frequencies and scales in the first historical wind power data set; the model performance evaluation based on the de-redundancy model obtains the optimal hyperparameter combination in different hyperparameter combinations; and the target intrinsic mode function after removing the redundant intrinsic mode functions is obtained based on the optimal hyperparameter combination; the first historical wind power data set is converted from the time domain to the frequency domain to obtain a second historical wind power data set represented in the frequency domain; and the wind power data set corresponding to the first preset number of cycles and each cycle of the adaptive wind power data set is obtained from the second historical wind power data set; the wind power data corresponding to each cycle is decomposed into a seasonal component, a trend component and a residual component; and the seasonal component, the trend component and the residual component are weighted and summed respectively to obtain a target seasonal component, a target residual component and a target trend component; the target seasonal component, the target residual component, the target trend component and the target intrinsic mode function are fused to obtain fused features; and the fused features are input into the long sequence information perception model to obtain the predicted short-term wind power. In this process, the advantages of two decomposition techniques, namely the multidimensional variational modal decomposition model and the decomposition of wind power data corresponding to each cycle, are utilized to comprehensively analyze the historical wind power data set from the frequency domain and time domain perspectives. This not only improves the modeling capability of complex time series patterns of wind power, but also can more accurately analyze the deep coupling relationship between wind power and influencing factors, providing more refined feature support for short-term wind power forecasting, thereby significantly improving the prediction accuracy and robustness.

[0052] In an embodiment of the present invention, the model performance evaluation based on the de-redundancy model in S102 to obtain the optimal hyperparameter combination from different hyperparameter combinations can be implemented through S102A to S102B, which is specifically explained through the following steps.

[0053] S102A, traverse different hyperparameter combinations, and obtain the error evaluation results of the de-redundancy model under each set of hyperparameter combinations through a cross-validation method.

[0054] S102B. The hyperparameter combination with the smallest error evaluation result is taken as the optimal hyperparameter combination.

[0055] In an embodiment of the present invention, model performance evaluation can be achieved by error evaluation. By traversing different possible hyperparameter combinations, the hyperparameter combinations are applied to the de-redundancy model, and the error of the de-redundancy model under each hyperparameter combination can be evaluated by a 5-fold cross-validation method to obtain multiple error evaluation results. The smallest target error evaluation result among the multiple error evaluation results is determined, and the hyperparameter combination corresponding to the target error evaluation result is used as the optimal hyperparameter combination.

[0056] In some embodiments of the present invention, S101 further includes S10 to S11, which is explained by the following steps.

[0057] S10, normalizing a preset number of intrinsic mode functions to obtain processed intrinsic mode functions;

[0058] S11. Centralize the target variable wind power in the first historical wind power data set to obtain a processed target wind power.

[0059] In some embodiments of the present invention, in order to scale different features to the same scale and eliminate dimensional differences, thereby improving model performance and stability, etc., a preset number of intrinsic mode functions are normalized to obtain processed intrinsic mode functions.

[0060] Furthermore, the target variable wind power is centrally processed, that is, the target variable wind power is subtracted from its mean value, so that the data is distributed around the zero mean to obtain the processed target wind power.

[0061] In an embodiment of the present invention, obtaining the target intrinsic mode function after removing redundant intrinsic mode functions based on the optimal hyperparameter combination in S102 can be implemented through S102a to S102b, which is explained through the following steps.

[0062] S102a, using the optimal hyperparameter combination, the processed target wind power, the processed intrinsic mode function and the de-redundancy model to perform data fitting to obtain the optimal regression coefficient vector.

[0063] S102b, taking the intrinsic mode functions with zero regression coefficients in the optimal regression coefficient vector as redundant intrinsic mode functions; and removing the redundant intrinsic mode functions to obtain the target intrinsic mode function.

[0064] In some embodiments of the present invention, the processed intrinsic mode function and the processed target wind power and the found optimal hyperparameter combination are used to apply the de-redundancy model (which can be an elastic network) algorithm for data fitting. This process will calculate the optimal regression coefficient vector A, in which each element ai represents the importance of the i-th processed intrinsic mode function to the wind power prediction. The larger the absolute value of the coefficient, the more important the processed intrinsic mode function is; the processed intrinsic mode function with a coefficient close to 0 is considered insignificant and is eliminated. Formula (6) for data fitting using the de-redundancy model (which can be an elastic network) algorithm is as follows:

[0065] (6)

[0066] In the above formula (6), z is the processed intrinsic mode function, is the optimal regression coefficient vector, β is the regression coefficient vector, y is the target wind power after processing, P is the number of intrinsic mode functions after processing, j is the number of intrinsic mode functions after processing, is the penalty intensity parameter, which is a non-negative regularization parameter. The de-redundancy model contains Norm and The norm penalty term, To control Norm and The ratio of norm regularization, is the jth element of the regression coefficient vector β, for The square value of .

[0067] In some embodiments of the present invention, S104 may be implemented through the following S1041 to S1042, which is explained through the following steps.

[0068] S1041. Perform weighted summation on the seasonal components of each cycle to obtain a target seasonal component.

[0069] S1042. Perform weighted summation on the trend components of each cycle to obtain a target trend component; and determine a target residual component through the second historical wind power data set, the target seasonal component, and the target trend component.

[0070] In some embodiments of the present invention, the seasonal component and the trend component of each cycle are weighted and summed respectively by the following formulas (7) and (8) to obtain the target seasonal component and the target trend component, and the target residual component is further calculated by the target seasonal component and the target trend component.

[0071] (7)

[0072] (8)

[0073] In the above formulas (7) and (8), is the target seasonal component, is the target trend component, For the i Cycle p , For the i Cycle p The corresponding seasonal components, For the i Cycle p The corresponding trend component.

[0074] In some embodiments of the present invention, S1042 may be implemented through S201, which is explained by the following steps.

[0075] S201. Perform a difference operation using the second historical wind power data set and the target seasonal component and the target trend component to obtain a target residual component.

[0076] In some embodiments of the present invention, the target seasonal component and the target trend component are respectively subtracted from the second historical data set to obtain the target residual component, as shown in the following formula (9):

[0077] (8)

[0078] In the above formula, R is the target residual component, and Y is the second historical wind power data set.

[0079] In the embodiment of the present invention, the following is a simulation experiment provided by the present invention, which can further illustrate the advantages of the short-term wind power prediction method provided by the present invention, specifically:

[0080] 1. Experimental conditions

[0081] The present invention conducts experiments on a 12th Gen Intel(R) Core(TM) i7-12650H CPU, an NVIDIAGeForce RTX 4060 Laptop GPU, and a Windows 11 operating system using the deep learning framework Tensorflow-GPU 2.6.

[0082] The method compared in the experiment is as follows: short-term wind power prediction is performed by using the following model, and is compared with the short-term wind power prediction method provided by the embodiment of the present invention.

[0083] CNN-LSTM (M1): A hybrid model of convolutional neural network and long short-term memory network to capture local features and temporal features.

[0084] Autoformer (M2): It tackles prediction problems by combining a deep factorization architecture with an autocorrelation mechanism.

[0085] Improved Transformer (M3): Based on the fusion of multi-dimensional modal decomposition and seasonal decomposition technology proposed in the embodiment of the present invention, the Transformer architecture is used as the predictor.

[0086] 2. Experimental Dataset

[0087] The experimental data set uses one year of wind power data from an onshore wind farm, with a time resolution of 15 minutes and a total of 35,040 data observations. In addition, the data set also includes wind speed and direction collected at different heights (10 meters, 30 meters, 50 meters and 70 meters), as well as climate factors such as ambient temperature, air pressure and humidity outside the cabin. 70% of the data is used for training and 30% for testing.

[0088] 3. Experimental evaluation indicators

[0089] Four classic indicators are used to evaluate the performance of the short-term wind power prediction method provided by CNN-LSTM, Autoformer, improved Transformer and the present invention, including mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE) and R2 (coefficient of determination). For MAE, MAPE and RMSE, the smaller the value, the better the performance of the prediction method. R2 fluctuates in the range of 0 to 1, and the larger the value, the better the performance. Since the wind power sequence contains more zero values, the calculation method of the MAPE formula is modified, that is, MSE (mean square error) divided by the average wind power.

[0090] 4. Comparative Experiment

[0091] In order to verify the effectiveness of the method of the present invention, a comparative experiment was conducted on the above algorithms to predict the wind power in the next hour. The results are shown in Table 1.

[0092] Table 1 Wind power prediction evaluation indicators

[0093]

[0094] As can be seen from Table 1, since the present invention proposes a short-term wind power prediction method that utilizes the advantages of two decomposition techniques and comprehensively analyzes wind power data from both frequency domain and time domain perspectives, compared with the other three methods, it not only improves the modeling capability of complex time series patterns of wind power, but also can more accurately analyze the deep coupling relationship between wind power and influencing factors, providing more refined feature support for short-term wind power prediction, thereby significantly improving prediction accuracy and robustness.

[0095] Reference Figure 2 , shows a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the electronic device.

[0096] like Figure 2As shown, the electronic device may include: a processor (processor) 502, a communication interface (Communications Interface 504, a memory (memory) 506, and a communication bus 508.

[0097] in:

[0098] The processor 502 , the communication interface 504 , and the memory 506 communicate with each other via a communication bus 508 .

[0099] The communication interface 504 is used to communicate with other electronic devices or servers.

[0100] The processor 502 is used to execute the program 510, and specifically can execute the relevant steps in the above method embodiment.

[0101] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0102] The processor 502 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0103] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0104] The program 510 may be specifically used to enable the processor 502 to execute operations corresponding to the methods described in the above method embodiments.

[0105] The specific implementation of each step in program 510 can refer to the corresponding description of the corresponding steps and units in the above method embodiment, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above method embodiment, which will not be repeated here.

[0106] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present invention can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present invention.

[0107] The method according to the embodiment of the present invention described above may be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and will be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0108] Those of ordinary skill in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present invention.

[0109] The above implementation methods are only used to illustrate the embodiments of the present invention, and are not limitations of the embodiments of the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the embodiments of the present invention. The patent protection scope of the embodiments of the present invention should be defined by the claims.

Claims

1. A method for predicting short-term wind power, characterized in that: include: Inputting the first historical wind power data set into a multidimensional variational modal decomposition model to obtain a preset number of intrinsic mode functions; each intrinsic mode function corresponds to a component of different frequency and scale in the first historical wind power data set; Normalizing the preset number of intrinsic mode functions to obtain processed intrinsic mode functions; Centralizing the target variable wind power in the first historical wind power data set to obtain a processed target wind power; The model performance evaluation based on the de-redundancy model obtains the optimal hyperparameter combination from different hyperparameter combinations; and uses the optimal hyperparameter combination, the processed target wind power, the processed intrinsic mode function and the de-redundancy model to perform data fitting to obtain the optimal regression coefficient vector; Taking the intrinsic mode function with a regression coefficient of zero in the optimal regression coefficient vector as a redundant intrinsic mode function; and removing the redundant intrinsic mode function to obtain a target intrinsic mode function; Convert the first historical wind power data set from the time domain to the frequency domain to obtain a second historical wind power data set represented in the frequency domain; and obtain an adaptive previous preset number of cycles and wind power data corresponding to each cycle from the second historical wind power data set; Decompose the wind power data corresponding to each period into seasonal components, trend components and residual components; and performing weighted summation on the seasonal component, trend component and residual component respectively to obtain a target seasonal component, a target residual component and a target trend component; The target seasonal component, the target residual component, the target trend component and the target intrinsic mode function are fused to obtain fused features; and the fused features are input into a long sequence information perception model to obtain predicted short-term wind power; The optimal regression coefficient vector can be obtained by the following formula: ； In the above formula, z is the intrinsic mode function after the processing, is the optimal regression coefficient vector, is the regression coefficient vector, y is the target wind power after the processing, P is the number of the intrinsic mode functions after the processing, j is the number of the intrinsic mode functions after the processing, is the penalty intensity parameter, which is a non-negative regularization parameter. The de-redundancy model contains Norm and The norm penalty term, To control Norm and The ratio of norm regularization, is the jth element of the regression coefficient vector β, for The square value of and is the optimal hyperparameter combination.

2. The method according to claim 1, characterized in that The model performance evaluation includes error evaluation; The model performance evaluation based on the de-redundancy model obtains the optimal hyperparameter combination from different hyperparameter combinations, including: Traversing the different hyperparameter combinations, and obtaining the error evaluation results of the de-redundancy model under each set of hyperparameter combinations through a cross-validation method; The hyperparameter combination with the smallest error evaluation result is taken as the optimal hyperparameter combination.

3. The method according to claim 1, characterized in that The seasonal component, the trend component and the residual component are weighted and summed respectively to obtain a target seasonal component, a target residual component and a target trend component, including: Performing weighted summation on the seasonal components of each cycle to obtain a target seasonal component; The trend components of each cycle are weighted and summed to obtain a target trend component; and the target residual component is determined by the second historical wind power data set, the target seasonal component and the target trend component.

4. The method according to claim 3, characterized in that The determining the target residual component by using the second historical wind power data set, the target seasonal component and the target trend component comprises: The target residual component is obtained by performing a difference operation using the second historical wind power data set and the target seasonal component and the target trend component.

Images