A wind power simulation method, system, computer equipment, and storage medium based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert models.

Abstract

This application relates to the field of wind power simulation and provides a method, system, computer equipment, and storage medium for wind power simulation based on ensemble empirical mode decomposition and residual multi-scale temporal convolutional sparse expert models. The method includes: acquiring historical power datasets and meteorological datasets of the target wind farm; performing sliding window processing to construct an initial training sample set; using ensemble empirical mode decomposition technology to decompose the data into a series of intrinsic mode function sequences and a residual sequence to obtain the ensemble empirical mode decomposition result; using a preset similarity algorithm to filter the intrinsic mode function sequences and update the target training sample set; constructing a residual multi-scale temporal convolutional sparse expert model and training the model based on the target training sample set; and inputting the spliced ​​sequence to be simulated into the trained model to output the wind power simulation result. The implementation of this method improves the effectiveness of wind power simulation.

Description

Technical Field

[0001] This application relates to the field of wind power simulation technology, and in particular to a wind power simulation method, system, computer equipment, and storage medium based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model. Background Technology

[0002] As an environmentally friendly and sustainable clean energy source, wind power is playing an increasingly important role in modern power systems. However, due to the intermittent, random, and volatile nature of wind energy, its large-scale grid integration can pose challenges to the stability and security of the power system. Wind power simulation technology is an important technical means to mitigate these problems.

[0003] Currently, methods for wind power simulation can be mainly divided into three categories: physical methods, statistical methods, and deep learning-based methods. Physical methods do not require a large amount of historical data, but they are computationally expensive and difficult to achieve high accuracy. Statistical methods have the advantages of simple implementation and fast computation speed, but they are difficult to handle nonlinear and complex wind data, and the simulation results often fail to meet practical needs. In recent years, deep learning-based methods have gradually become an important method for wind power simulation due to their powerful feature extraction capabilities. However, existing deep learning methods struggle to capture the multi-scale features of wind power data when dealing with non-stationary time series such as wind power data, and they also fail to fully learn the changing patterns of short-term and long-term dependencies in wind power data, resulting in insufficient effectiveness of wind power simulation. Summary of the Invention

[0004] This application provides a wind power simulation method, system, computer equipment, and storage medium based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model. More specifically, this application provides a wind power simulation method, a wind power simulation system, computer equipment, computer storage medium, and computer program product based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model, thereby improving the effectiveness of wind power simulation.

[0005] In a first aspect, embodiments of this application provide a wind power simulation method based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert models, including:

[0006] Obtain the historical power dataset and meteorological dataset of the target wind farm, and perform sliding window processing on the historical power dataset and the meteorological dataset to construct an initial training sample set;

[0007] Using ensemble empirical mode decomposition (EMD), the historical power sequence in the initial training sample set is decomposed into a series of intrinsic mode function sequences and a residual sequence to obtain the ensemble empirical mode decomposition result;

[0008] The intrinsic mode function sequence is filtered using a preset similarity algorithm to obtain a filtered sequence. The initial training sample set is then updated based on the filtered sequence to obtain the target training sample set.

[0009] A residual multi-scale temporal convolutional sparse expert model is constructed, and the residual multi-scale temporal convolutional sparse expert model is trained based on the target training sample set;

[0010] The spliced ​​sequence to be simulated is input into the trained residual multi-scale temporal convolutional sparse expert model to output the intrinsic mode function simulation results and residual simulation results.

[0011] The simulation results of the intrinsic mode functions and the simulation results of the residuals are added together to obtain the simulation results of wind power.

[0012] Optionally, in some embodiments of this application, the step of performing sliding window processing on the historical power dataset and the meteorological dataset to construct an initial training sample set includes:

[0013] The historical power dataset and the meteorological dataset are traversed based on a preset sliding window length to determine the input features and target power value of each sample.

[0014] The initial training sample set is constructed based on the input features and target power value of each sample.

[0015] Optionally, in some embodiments of this application, the step of using ensemble empirical mode decomposition (EMD) to decompose the historical power sequence in the initial training sample set into a series of intrinsic mode function (IMF) sequences and a residual sequence to obtain the EMD result includes:

[0016] For any repeated cycle, random white noise is added to the historical power sequence to obtain a sequence with added white noise; empirical mode decomposition is performed on the sequence with added white noise to obtain the preliminary decomposition results for each repeated cycle.

[0017] The eigenmode functions of the same order in the preliminary decomposition results are averaged to obtain the integrated empirical mode decomposition results of the corresponding order.

[0018] Optionally, in some embodiments of this application, performing empirical mode decomposition on the sequence after adding white noise to obtain preliminary decomposition results for each repetition round includes:

[0019] Using the signal extremum detection method, the local extremum points of the sequence after adding white noise are found. Based on the local extremum points, upper and lower envelopes are constructed by spline interpolation to obtain the envelope mean curve.

[0020] The difference between the sequence after adding white noise and the envelope mean curve is used as a new signal, and the decomposition is repeated until a preset decomposition stopping condition is met to determine the preliminary decomposition result.

[0021] Optionally, in some embodiments of this application, the step of filtering the intrinsic mode function sequence using a preset similarity algorithm to obtain a filtered sequence includes:

[0022] The similarity value between the intrinsic mode function sequence and the historical power sequence is calculated using a preset similarity algorithm;

[0023] Based on the similarity values, the intrinsic mode function sequence with the highest similarity to the historical power sequence is determined as the sequence of the target order; sequences from the first order to the target order are retained as the filtered sequences.

[0024] Optionally, in some embodiments of this application, updating the initial training sample set according to the filtered sequence to obtain the target training sample set includes:

[0025] The residual sequence is updated based on the filtered sequence to obtain the updated residual sequence;

[0026] The initial training sample set is updated based on the filtered sequence and the updated residual sequence to obtain the target training sample set.

[0027] Optionally, in some embodiments of this application, before the step of inputting the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model, the method further includes:

[0028] Obtain the power sequence to be simulated for the target wind farm;

[0029] The power sequence to be simulated is decomposed using integrated empirical mode decomposition technology, and the decomposition results are filtered using a preset similarity algorithm to obtain the intrinsic mode function sequence and the residual sequence to be simulated.

[0030] The intrinsic mode function sequence to be simulated, the residual sequence to be simulated, and the meteorological sequence are spliced ​​together to form a multivariate input sequence, so as to determine the spliced ​​sequence to be simulated.

[0031] Secondly, embodiments of this application provide a wind power simulation system based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert models, which has the function of implementing the wind power simulation method based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert models provided in the first aspect above. The function can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware.

[0032] In one possible design, the system includes:

[0033] The data acquisition module is used to acquire the historical power dataset and meteorological dataset of the target wind farm, and to perform sliding window processing on the historical power dataset and the meteorological dataset to construct an initial training sample set;

[0034] The signal decomposition module is used to decompose the historical power sequence in the initial training sample set into a series of intrinsic mode function sequences and a residual sequence using integrated empirical mode decomposition technology, so as to obtain the integrated empirical mode decomposition result;

[0035] The similarity filtering module is used to filter the intrinsic mode function sequence using a preset similarity algorithm to obtain a filtered sequence, and to update the initial training sample set according to the filtered sequence to obtain the target training sample set.

[0036] The model building and training module is used to build a residual multi-scale temporal convolutional sparse expert model and train the residual multi-scale temporal convolutional sparse expert model based on the target training sample set.

[0037] The model simulation module is used to input the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model, so as to output the simulation results of intrinsic mode functions and residual simulation results.

[0038] The power simulation module is used to add the simulation results of the intrinsic mode functions and the simulation results of the residuals to obtain the wind power simulation results.

[0039] In another aspect, this application provides a computer device including at least one connected processor and a memory, wherein the memory is used to store program code, and the processor is used to call the program code in the memory to execute the methods described in the above aspects.

[0040] In another aspect, embodiments of this application provide a computer storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the above aspects.

[0041] In another aspect, this application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods described in the above aspects.

[0042] Compared with traditional technologies, the technical solution of this application improves the processing effect of non-stationary sequences by constructing samples through sliding windows, sequence decomposition and screening, and dedicated model construction and training. It also reduces the redundant computational cost of model training, enhances the ability to extract complex time series features, thereby improving the effectiveness of wind power simulation and enhancing the stability of power grid operation. Attached Figure Description

[0043] Figure 1 This is a flowchart of one embodiment.

[0044] Figure 2 This is a schematic diagram of a residual multi-scale temporal convolutional sparse expert model in one embodiment.

[0045] Figure 3 This is a structural block diagram of a system in one embodiment.

[0046] Figure 4 This is an internal structural diagram of a computer device in one embodiment.

[0047] Figure 5 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0048] The terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The division of modules appearing in the embodiments of this application is only a logical division. In actual applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interface, and the indirect coupling or communication connection between modules may be electrical or other similar forms. None of these are limited in the embodiments of this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0049] Figure 1 This is a flowchart illustrating one embodiment, such as... Figure 1 As shown in the embodiments of this application, the wind power simulation method based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model includes:

[0050] S1. Obtain the historical power dataset and meteorological dataset of the target wind farm. Perform sliding window processing on the historical power dataset and meteorological dataset to construct the initial training sample set.

[0051] Among them, the historical power dataset refers to the set of power-related data that has occurred at the target wind farm; the meteorological dataset refers to the set of meteorological-related data corresponding to the target wind farm.

[0052] Among them, sliding window processing refers to the processing method of traversing data according to a preset window length.

[0053] The initial training sample set refers to the set of samples used for the initial training of the model.

[0054] S2. Using ensemble empirical mode decomposition (EMD) technology, the historical power sequence in the initial training sample set is decomposed into a series of intrinsic mode function sequences and a residual sequence to obtain the ensemble empirical mode decomposition result.

[0055] Among them, the Ensemble Empirical Mode Decomposition (EEMD) technique refers to a signal decomposition technique proposed in this application that improves Empirical Mode Decomposition (EMD) and can solve the mode aliasing problem; the EEMD result refers to the set of sequences obtained after decomposition.

[0056] Among them, the historical power sequence refers to the sequence formed by arranging historical power data over time.

[0057] Among them, the Intrinsic Mode Function (IMF) sequence refers to the set of single-component, locally stationary oscillating sequences obtained from the decomposition; the residual sequence refers to the trend sequence remaining after decomposition.

[0058] S3. Use a preset similarity algorithm to filter the intrinsic mode function sequence to obtain the filtered sequence. Update the initial training sample set according to the filtered sequence to obtain the target training sample set.

[0059] The preset similarity algorithm refers to a pre-defined calculation method used to measure the degree of sequence similarity. For example, the preset similarity algorithm is the Tanimoto similarity algorithm.

[0060] Here, screening refers to the operation of selecting key sequences based on similarity; the filtered sequence refers to the key intrinsic mode function sequence retained after similarity screening.

[0061] The target training sample set refers to the updated sample set used for the final training of the model.

[0062] S4. A residual multi-scale temporal convolutional sparse expert model is constructed, and the residual multi-scale temporal convolutional sparse expert model is trained based on the target training sample set.

[0063] Among them, the Residual Multi-scale Temporal Convolutional Mixture of Experts (RM-TCN-MoE) refers to a model with residual connections and multi-scale feature extraction capabilities, also known as the Residual Multi-scale Temporal Convolutional Sparse Expert Network or the Wind Power Simulation Model.

[0064] S5 inputs the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model to output the simulation results of intrinsic mode functions and residual simulation results.

[0065] The spliced ​​sequence to be simulated refers to the multivariate sequence formed by splicing related sequences to be simulated.

[0066] Among them, the intrinsic mode function simulation result refers to the sequence of intrinsic mode functions corresponding to the power to be simulated as simulated by the model; the residual simulation result refers to the residual sequence corresponding to the power to be simulated as simulated by the model.

[0067] S6. Add the simulation results of the intrinsic mode function and the residual simulation results to obtain the simulation results of wind power.

[0068] Among them, the wind power simulation result refers to the wind power value of the simulated period in the final output of the model.

[0069] Compared to traditional technologies, in this embodiment, the historical power dataset and meteorological dataset of the target wind farm are first acquired, and a sliding window process is performed to construct an initial training sample set. Then, ensemble empirical mode decomposition (EMD) is used for decomposition, and a preset similarity algorithm is used for filtering to update the target training sample set. Subsequently, a residual multi-scale temporal convolutional sparse expert model is constructed and trained based on the target training sample set. Finally, the spliced ​​sequence to be simulated is input into the trained model to output the wind power simulation result. The technical solution of this embodiment improves the processing effect of non-stationary sequences, reduces the redundant computational cost of model training, and enhances the ability to extract complex temporal features through sliding window sample construction, sequence decomposition and filtering, and dedicated model construction and training. This improves the effectiveness of wind power simulation and enhances the stability of power grid operation.

[0070] This application provides a wind power simulation method based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model, which can also be called a wind power simulation method based on EEMD and RM-TCN-MoE models. For ease of description in the embodiments, it can also be simply referred to as the simulation method of this application.

[0071] In a specific embodiment, in step S1, the key features affecting wind power are mainly determined by historical power and meteorological features. Therefore, this embodiment can obtain the historical power data and corresponding meteorological data of the target wind farm, which constitute the historical power dataset and meteorological dataset, respectively, as the basis for subsequent construction and training of residual multi-scale temporal convolutional sparse expert model.

[0072] In one example, the historical power dataset contains variables including date, time, and historical power values; the meteorological dataset contains variables including date, time, floor height, wind speed, wind direction, humidity, temperature, air pressure, and air density.

[0073] During the model training phase, a sliding window process is applied to the historical power dataset and meteorological dataset to construct training samples and obtain the initial training sample set.

[0074] Specifically, the sliding window length is set to 7 days. That is, historical data of any consecutive 7 days (including data in the historical power dataset and meteorological dataset) are used as input features to simulate the target power value of the day after the time window. In this way, all time points of the historical power dataset and meteorological dataset are traversed to generate the initial training sample set.

[0075] For example, historical power data and meteorological data from January 1 to January 7 can be used as input features, and historical power data from January 8 can be used as the target power value to construct a set of samples in the initial training sample set.

[0076] In the model application and testing phase of steps S5 to S6, historical power data and meteorological data of the most recent 7 consecutive days are used as model inputs to simulate the power output of the next day.

[0077] Optionally, in some embodiments of this application, a sliding window process is performed on the historical power dataset and the meteorological dataset to construct an initial training sample set, including: traversing the historical power dataset and the meteorological dataset based on a preset sliding window length to determine the input features and target power value of each sample; and constructing an initial training sample set based on the input features and target power value of each sample.

[0078] The preset sliding window length refers to the pre-set time length of the sliding window, for example, a sliding window length of 7 days.

[0079] Here, traversal refers to the operation of processing each data point in the dataset sequentially; input features refer to the sample data used as input to the model; and target power value refers to the power value to be simulated for each sample.

[0080] In this embodiment, the sample data is determined by traversing the data through a preset window length, which improves the rationality of the initial training sample set and provides a high-quality dataset for model training.

[0081] In one specific embodiment, in step S2, EEMD is used to decompose the historical power sequence to obtain the IMF sequence.

[0082] Integrated Empirical Mode Decomposition (EEMD) is an improved signal decomposition technique based on Empirical Mode Decomposition (EMD). It solves the common mode aliasing problem in EMD and preserves the physical meaning of nonlinear signals to the greatest extent possible. Its core idea is to actively add white noise to the EMD and then extract the intrinsic mode function (IMF) sequence. White noise has a uniform frequency distribution, which can locate signal components of different scales to their corresponding frequency ranges, reducing interference with local extrema.

[0083] Using the historical power sequence as the original power sequence in the decomposition process, step S2 may include the following sub-steps: S21, adding random white noise to the original power sequence. S22, performing Empirical Mode Decomposition (EMD) on the sequence after adding white noise. S23, repeatedly adding new white noise to the original power sequence and repeating the above steps (S21 to S22). S24, averaging the IMF sequences of the same order among the IMF and residual sequences obtained from multiple decompositions to obtain the EEMD decomposition result of the corresponding order.

[0084] Specifically, S21, random white noise is added to the original power sequence, which is mathematically expressed as:

[0085]

[0086] in, It is a white noise sequence.

[0087] S22, Empirical Mode Decomposition (EMD) is performed on the sequence after adding white noise.

[0088] S23, add K new white noise to the original power sequence, and repeat steps S21 and S22 to obtain the preliminary decomposition results for each repeated round.

[0089] The number of times white noise is added, K, is a preset value, typically ranging from 100 to 500. A larger number of noise additions can achieve better decomposition results, but also requires more computation time. To balance computational efficiency and decomposition accuracy, K=100 is preferably chosen.

[0090] S24, in the IMF and residual sequences (preliminary decomposition results) obtained from these K decompositions, the average of the IMF sequences of the same order is taken to obtain the EEMD decomposition result (ensemble empirical mode decomposition result) of the corresponding order, which is mathematically expressed as:

[0091]

[0092] Optionally, in some embodiments of this application, the ensemble empirical mode decomposition technique is used to decompose the historical power sequence in the initial training sample set into a series of intrinsic mode function sequences and a residual sequence to obtain the ensemble empirical mode decomposition result. This includes: adding random white noise to the historical power sequence for any repeated round to obtain a sequence with added white noise; performing empirical mode decomposition on the sequence with added white noise to obtain the preliminary decomposition result for each repeated round; and averaging the intrinsic mode functions of the same order in the preliminary decomposition result to obtain the ensemble empirical mode decomposition result of the corresponding order.

[0093] Here, "repetition rounds" refers to the number of times white noise is added and decomposed. For example, the maximum number of repetition rounds is set to K. "Random white noise" refers to a random noise signal with a uniform frequency distribution. "Sequence after adding white noise" refers to the signal after adding white noise to the original historical power sequence.

[0094] The preliminary decomposition result refers to the sequence obtained after adding white noise and decomposing it through multiple rounds.

[0095] Among them, eigenmode functions of the same order refer to eigenmode functions of the same order obtained from decomposition in different repeated rounds.

[0096] In this embodiment, by adding white noise multiple times for decomposition and averaging, the mode mixing problem is further alleviated, the accuracy of the integrated empirical mode decomposition results is improved, and the original signal characteristics are better preserved.

[0097] In a specific embodiment, S22 specifically includes: steps S221 to S223, specifically including:

[0098] S221, using the signal extreme point detection method, find The local maxima and minima are constructed using spline interpolation. The upper and lower envelopes are obtained, and thus the envelope mean curve is obtained.

[0099]

[0100] in, The upper envelope, The lower envelope, This is the envelope mean curve.

[0101] S222, the difference between the original signal and the envelope mean curve is used as the new signal, and repeated decomposition is performed based on step S221.

[0102] For example, targeting (The sequence obtained from the 0th decomposition) will With envelope mean curve The difference The sequence obtained during the first decomposition ;

[0103] For the sequence obtained in the first decomposition The corresponding envelope mean curve is obtained by solving the problem. The difference between the two The sequence obtained during the second decomposition The mathematical expression for this process is:

[0104]

[0105] And so on, for the sequence obtained from the (k-1)th decomposition The corresponding envelope mean curve is obtained by solving the problem. The difference between the two The sequence obtained at the k-th decomposition The mathematical expression for this process is:

[0106]

[0107] in, and Let represent the sequences obtained at the (k-1)th and kth decompositions, respectively. for The envelope mean curve. Check. Does it meet the IMF conditions, i.e., when When the number of zeros crossed differs from the number of extreme points by no more than 1, a first-order IMF sequence can be obtained. ,Right now

[0108]

[0109] S223, order Repeat steps (S221) and (S222) n times until the preset decomposition stopping condition is met. Ultimately, the original sequence can be decomposed into an n-order IMF sequence and a residual sequence.

[0110]

[0111] in, Represents the i-th order IMF sequence. Residual sequences can be represented by... Calculations show that The order of the IMF sequence is represented by n, which refers to the number of times steps (S221) and (S222) are repeated until the preset decomposition stopping condition is met.

[0112] Based on the descriptions of steps S221 and S222, by repeating steps S221 and S222 n times, an n-order IMF sequence can be obtained.

[0113] The preset decomposition stopping conditions include one of the following: 1. The residual sequence is a monotonic function; 2. The number of extreme points in the residual sequence is less than a preset threshold (e.g., less than 2); 3. The order of the decomposed IMF sequence reaches the preset maximum decomposition order.

[0114] Optionally, in some embodiments of this application, empirical mode decomposition is performed on the sequence after adding white noise to obtain preliminary decomposition results for each repeated round, including: using a signal extremum detection method to find local extrema of the sequence after adding white noise; constructing upper and lower envelopes based on the local extrema through spline interpolation to obtain the envelope mean curve; using the difference between the sequence after adding white noise and the envelope mean curve as a new signal, and performing repeated decomposition until a preset decomposition stop condition is met to determine the preliminary decomposition results.

[0115] Among them, the signal extremum point detection method refers to the method of detecting the local maximum and minimum points in a signal; the local extremum point refers to the point corresponding to the maximum or minimum value within a local range in the signal.

[0116] Among them, spline interpolation refers to the numerical values ​​that construct a smooth curve through interpolation points; upper and lower envelopes refer to the curves that enclose all local extreme points of the signal; and the envelope mean curve refers to the curve formed by the average values ​​of the upper and lower envelopes.

[0117] The new signal refers to the difference between the original signal and the envelope mean curve.

[0118] The preset decomposition stop condition refers to the pre-defined criteria for stopping the decomposition.

[0119] In this embodiment, the techniques of extreme point detection, envelope construction, and repeated decomposition ensure the standardization of empirical mode decomposition, improve the accuracy of the preliminary decomposition results, and provide a reliable foundation for subsequent integrated decomposition.

[0120] In a specific embodiment, step S3 can also be described as: using Tanimoto similarity to filter the intrinsic mode function sequence to obtain the key IMF sequence (filtered sequence), and updating the residual sequence based on the key IMF sequence.

[0121] Specifically, after obtaining the IMF sequence using EEMD, the IMF sequence can be filtered to obtain an IMF sequence that retains more features of the original power sequence, which can then be used as input for subsequent wind power simulation models.

[0122] In one example, step S3 may include the following sub-steps:

[0123] S31, Calculate the Tanimoto similarity (numerical similarity). The Tanimoto similarity considers both the direction and norm similarity of the vectors, thus enabling simultaneous measurement of the similarity in amplitude and pulse shape between the obtained IMF sequence and the original wind power sequence. The calculation method is as follows:

[0124]

[0125] in, Represents the original power sequence. Let i represent the i-th order IMF sequence.

[0126] Among them, inner product term Used to consider vector direction, used to reflect pulse shape similarity; magnitude term It is used to consider vector norms and reflect magnitude similarity.

[0127] S32, determine the relationship with the original power sequence The IMF sequence with the highest similarity is the k-th order IMF sequence (the sequence of the target order); the first k order IMF sequences (i.e., the IMF sequences from the 1st to the kth order) are retained for subsequent processing.

[0128] S33, Update the residual sequence based on the first k-order IMF sequence retained in step S32, using the following update formula:

[0129]

[0130] Optionally, in some embodiments of this application, the intrinsic mode function sequence is filtered using a preset similarity algorithm to obtain a filtered sequence, including: calculating the similarity value between the intrinsic mode function sequence and the historical power sequence using the preset similarity algorithm; determining the intrinsic mode function sequence with the highest similarity to the historical power sequence as the sequence of the target order based on the similarity value; and retaining the sequences from the first order to the target order as the filtered sequence.

[0131] The preset similarity algorithm is Tanimoto, and the similarity value refers to the numerical value calculated by the similarity algorithm to measure the degree of sequence similarity.

[0132] Here, the sequence of target order refers to the sequence of intrinsic mode functions that has the highest similarity to the original historical power sequence. For example, the sequence of target order is the sequence of intrinsic mode functions that has the highest similarity to the original power sequence. The IMF sequence with the highest similarity is the k-th order IMF sequence.

[0133] The sequence from the first order to the target order refers to the sequence of all intrinsic mode functions from the first order to the target order, such as the sequence of IMFs from the first order to the kth order.

[0134] In this embodiment, by calculating similarity and retaining high-similarity sequences, core feature sequences are accurately selected, redundant information is eliminated, the effectiveness of feature sequences is improved, and the cost of subsequent model training is reduced.

[0135] In a specific embodiment, step S4 can also be described as: constructing the RM-TCN-MoE model, splicing the selected IMF sequence, residual sequence and meteorological data sequence to form a multivariate input sequence, and using it as the input to the RM-TCN-MoE model to train the RM-TCN-MoE model.

[0136] Among them, the RM-TCN-MoE model can capture multi-scale local features of wind power data, and the integration of sparse expert networks enables the model to capture more refined local patterns and enhance the modeling ability of complex temporal behaviors through expert activation.

[0137] Regarding the input and output of the RM-TCN-MoE model, in this embodiment, a multidimensional feature tensor is constructed using the selected IMF sequence, residual sequence, and meteorological data sequence, which is then used as the input for training the RM-TCN-MoE model. Specifically, the IMF sequence, residual sequence, and meteorological data sequence at the same time step are concatenated along the feature dimension to form a multivariate input sequence; the model output is the simulated IMF sequence value and the simulated residual sequence value corresponding to the wind power sequence to be simulated.

[0138] In a specific embodiment, the construction process of the RM-TCN-MoE model in step S4 is as follows.

[0139] Figure 2 This is a schematic diagram of a residual multi-scale temporal convolutional sparse expert model in one embodiment. See [link to diagram]. Figure 2 In its specific implementation, the residual multi-scale temporal convolutional sparse expert model is also known as the RM-TCN-MoE model, which mainly includes a residual connection module, a temporal convolution module, and a sparse expert network module.

[0140] The connections between the various modules and the data flow in the model application scenarios are as follows: Figure 2 As shown, it will not be elaborated further here.

[0141] (1) Residual connection modules can effectively alleviate the common problems of gradient vanishing and gradient exploding in deep neural networks. This connection method allows information to jump and be passed between different layers of the network, avoiding the gradual decay or loss of information and gradients in deep networks. Its mathematical form is as follows:

[0142]

[0143] in, This represents the input vector of the residual connection module. This represents the residual mapping function, which includes convolutional layers and nonlinear mappings. This represents the weights of each layer of the neural network in the residual mapping function.

[0144] (2) The temporal convolution module can perform efficient parallel computation using convolution operations and can effectively capture multi-scale features in sequence data. Its mathematical form can be expressed as:

[0145]

[0146] in, This represents the output of the dilated convolution. Represents the input sequence. Indicates the convolution kernel weights. Indicates the kernel size. This represents the expansion factor, which is the positional spacing of the convolution kernel weights.

[0147] (3) The sparse expert network module utilizes a gating network to select the expert model, where the expert model is a temporal convolutional module with different parameter settings. The gating network can assign different weights to the outputs of different temporal convolutional modules and perform weighted summation, thereby improving the overall performance of the model. Its mathematical form is as follows:

[0148]

[0149] in, This represents the output of the sparse expert network module. Represents the input sequence. This indicates that the gating network is assigned to the first The weights of each temporal convolutional module, Indicates the first The output of each temporal convolutional module This represents the K temporal convolutional modules with the largest weights selected by the gating network.

[0150] Among them, the gating network that assigns weights to different temporal convolutional modules needs to be obtained through model training, that is, the assigned weights are determined by the output of the gating network.

[0151] For example, suppose the short-time expert model has a convolutional kernel length of 3 and 2 layers; the medium-time expert model has a convolutional kernel length of 5 and 4 layers; and the long-time expert model has a convolutional kernel length of 3 and 5 layers, with expansion factors of [1, 2, 4, 8, 16] for each layer. When encountering short-term disturbances such as gusts or equipment response delays, the gating network assigns higher weights to the short-time expert model; when encountering disturbances such as diurnal wind speed variations, the gating network assigns higher weights to the medium-time expert model; and when encountering disturbances such as weather system migration or the passage of cold and warm fronts, the gating network assigns higher weights to the long-time expert model. By combining the outputs of different expert models through weighted summation, the overall simulation effect of the model can be improved.

[0152] In a specific embodiment, step S5 can also be described as follows: using the historical power data and meteorological data of the most recent seven consecutive days as model input, and using the trained RM-TCN-MoE model to simulate the IMF sequence and residual sequence of the power sequence to be simulated.

[0153] In the model application phase, step S5 specifically includes: decomposing and filtering the power sequence to be simulated (such as historical power data for the most recent seven consecutive days) as in steps S2 and S3 to obtain the intrinsic mode function (IMF) sequence and the residual sequence to be simulated. Then, the IMF sequence, the residual sequence, and the meteorological sequence are concatenated to obtain the concatenated multivariate sequence (the concatenated sequence to be simulated) input into the model. The model output includes the simulated IMF sequence (IMF simulation results) and the residual sequence (residual simulation results). The meteorological data sequence in the input data is used to enhance the model's simulation effect and does not need to be simulated.

[0154] It should be noted that the data processing procedures in the model application stage (steps S5 to S6) and the model training stage (steps S1 to S4) of this application are consistent. For example, the ensemble empirical mode decomposition process in step S2, the process of filtering the intrinsic mode function sequence using a preset similarity algorithm in step S3, and the process of concatenating multivariate data when training the model using the target training sample set in step S4 are still applicable in the model application stage (steps S5 to S6). The implementation details of the same technologies used in the model application stage (steps S5 to S6) and the model training stage (steps S1 to S4) of this application can be referred to each other, and will not be repeated here.

[0155] Optionally, in some embodiments of this application, updating the initial training sample set based on the filtered sequence to obtain the target training sample set includes: updating the residual sequence based on the filtered sequence to obtain the updated residual sequence; and updating the initial training sample set based on the filtered sequence and the updated residual sequence to obtain the target training sample set.

[0156] The updated residual sequence refers to the residual sequence after adjustment by combining the filtered sequence.

[0157] For example, the historical power sequence in the initial training sample set can be replaced with the filtered sequence and the updated residual sequence as new sample data, while retaining the meteorological data in the meteorological dataset and the target power value, thus obtaining a meteorological dataset. This facilitates the concatenation of the filtered IMF sequence, residual sequence, and meteorological data sequence during the model application phase to obtain the concatenated sequence to be simulated.

[0158] In this embodiment, the residual sequence and training sample set are updated by filtering the sequence, so that the sample set is closer to the core features of the original signal, thereby improving the quality of the sample set and facilitating the improvement of simulation efficiency in the subsequent model application stage.

[0159] Optionally, in some embodiments of this application, before the step of inputting the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model, the method further includes: obtaining the power sequence to be simulated for the target wind farm; decomposing the power sequence to be simulated using integrated empirical mode decomposition technology, and filtering the decomposition results using a preset similarity algorithm to obtain the intrinsic mode function sequence to be simulated and the residual sequence to be simulated; splicing the intrinsic mode function sequence to be simulated, the residual sequence to be simulated, and the meteorological sequence to form a multivariate input sequence to determine the spliced ​​sequence to be simulated.

[0160] The power sequence to be simulated refers to the wind power sequence for future periods that need to be simulated.

[0161] The intrinsic mode function sequence to be simulated refers to the intrinsic mode function sequence obtained after decomposition and screening; the residual sequence to be simulated refers to the residual sequence obtained after decomposition and screening.

[0162] Among them, the meteorological sequence refers to the meteorological data sequence corresponding to the period to be simulated.

[0163] Among them, the multivariate input sequence refers to the input sequence formed by splicing together multiple related sequences to be simulated.

[0164] In this embodiment, the sequence to be simulated is decomposed, filtered, and spliced ​​to make the input format to be simulated consistent with the training samples, thereby improving the standardization and accuracy of the simulation and ensuring that the model effectively outputs simulation results.

[0165] As can be seen from the above embodiments, one technical implementation flow of the simulation method of this application includes steps S1 to S6. In one technical description, the simulation method of this application includes: step S1, obtaining the historical power dataset and meteorological dataset of the target wind farm; step S2, using ensemble empirical mode decomposition (EEMD) to decompose the historical power sequence into a series of intrinsic mode function (IMF) sequences and a residual sequence; step S3, using Tanimoto Similarity is used to filter the intrinsic mode function sequences to obtain key IMF sequences, and the residual sequences are updated based on the key IMF sequences. In step S4, the selected IMF sequences, residual sequences, and meteorological data sequences are concatenated to form a multivariate input sequence, which is used as the input to the residual multi-scale temporal convolutional sparse expert network (RM-TCN-MoE) to train the RM-TCN-MoE model. In step S5, the historical power data and meteorological data of the most recent seven consecutive days are used as model input. The IMF sequences, residual sequences, and meteorological data sequences of the historical power sequence to be simulated are concatenated to obtain the multivariate input sequence to be simulated. The multivariate input sequence to be simulated is input into the trained RM-TCN-MoE model to simulate and obtain the simulation results of the IMF sequences and residual sequences. In step S6, the simulation results of the IMF sequences and residual sequences are added together to obtain the final wind power simulation result. The specific steps and implementation principles are the same as those in the above embodiment and will not be repeated here.

[0166] The simulation method of this application decomposes the wind power sequence into IMF sequence through EEMD, then uses Tanimoto similarity to screen out key IMF sequences, and then constructs and trains an RM-TCN-MoE model that can extract multi-scale features to improve the effectiveness of the wind power simulation method. This solves the technical problem that existing wind power simulation methods are unable to handle non-stationary signals, resulting in insufficient simulation effectiveness.

[0167] It should be noted that any technical feature in any of the above embodiments provided in this application is also applicable to any of the following embodiments provided in this application. For example, the technical features in the relevant embodiments of the above methods are also applicable to the relevant embodiments of the following devices, systems or equipment, and similar details will not be repeated hereafter.

[0168] Figure 3 Here is a structural block diagram of the system in one embodiment, with reference to Figure 3 The wind power simulation system based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert models, also known as the wind power simulation system based on EEMD and RM-TCN-MoE models, includes:

[0169] The data acquisition module 301 is used to acquire the historical power dataset and meteorological dataset of the target wind farm, and to perform sliding window processing on the historical power dataset and meteorological dataset to construct an initial training sample set.

[0170] The signal decomposition module 302 is used to decompose the historical power sequence in the initial training sample set into a series of intrinsic mode function sequences and a residual sequence using the integrated empirical mode decomposition technique, so as to obtain the integrated empirical mode decomposition result.

[0171] The similarity filtering module 303 is used to filter the intrinsic mode function sequence using a preset similarity algorithm to obtain the filtered sequence, and update the initial training sample set according to the filtered sequence to obtain the target training sample set.

[0172] The model building and training module 304 is used to build a residual multi-scale temporal convolutional sparse expert model and train the residual multi-scale temporal convolutional sparse expert model based on the target training sample set.

[0173] The model simulation module 305 is used to input the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model, so as to output the simulation results of intrinsic mode functions and residual simulation results.

[0174] The power simulation module 306 is used to add the intrinsic mode function simulation results and the residual simulation results to obtain the wind power simulation results.

[0175] In this embodiment of the application, based on, as follows Figure 3 The connection relationships between the various modules or units shown in the diagram improve the effectiveness of wind power simulation through the cooperation between these modules or units.

[0176] In another embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, it includes a processor, memory, input / output interfaces, and a communication interface. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface is connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores relevant data. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. The computer program can be executed by the processor to implement the various methods described in the above embodiments.

[0177] In yet another embodiment, a computer device is provided, such as a terminal, whose internal structure diagram may be as follows: Figure 5 As shown, it includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. The computer program can be executed by the processor to implement the various methods described in the above embodiments.

[0178] Those skilled in the art will understand that Figure 4 and Figure 5 The structure shown is only a block diagram of a part of the structure related to the present application and does not constitute a limitation on the computer device on which the present application is applied. It may also include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, in order to realize the function of the computer device.

[0179] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0180] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the systems, devices, equipment, modules or units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0181] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, devices, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0182] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0183] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0184] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0185] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium, or a semiconductor medium (e.g., a solid-state drive), etc.

[0186] The technical solutions provided by the embodiments of this application have been described in detail above. Specific examples have been used in the embodiments of this application to illustrate the principles and implementation methods of the embodiments of this application. The description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of the embodiments of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments of this application. Therefore, the content of this specification should not be construed as a limitation on the embodiments of this application.

Claims

1. A wind power simulation method based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model, characterized in that, The method includes: Obtain the historical power dataset and meteorological dataset of the target wind farm, and perform sliding window processing on the historical power dataset and the meteorological dataset to construct an initial training sample set; Using ensemble empirical mode decomposition (EMD), the historical power sequence in the initial training sample set is decomposed into a series of intrinsic mode function sequences and a residual sequence to obtain the ensemble empirical mode decomposition result; The intrinsic mode function sequence is filtered using a preset similarity algorithm to obtain a filtered sequence. The initial training sample set is then updated based on the filtered sequence to obtain the target training sample set. A residual multi-scale temporal convolutional sparse expert model is constructed, and the residual multi-scale temporal convolutional sparse expert model is trained based on the target training sample set; The spliced ​​sequence to be simulated is input into the trained residual multi-scale temporal convolutional sparse expert model to output the intrinsic mode function simulation results and residual simulation results. The simulation results of the intrinsic mode functions and the simulation results of the residuals are added together to obtain the simulation results of wind power.

2. The method according to claim 1, characterized in that, The process of applying a sliding window to the historical power dataset and the meteorological dataset to construct an initial training sample set includes: The historical power dataset and the meteorological dataset are traversed based on a preset sliding window length to determine the input features and target power value of each sample. The initial training sample set is constructed based on the input features and target power value of each sample.

3. The method according to claim 1, characterized in that, The method utilizes ensemble empirical mode decomposition (EMD) to decompose the historical power sequence in the initial training sample set into a series of intrinsic mode function (IMF) sequences and a residual sequence, yielding the EMD result, including: For any repeated cycle, random white noise is added to the historical power sequence to obtain a sequence with added white noise; empirical mode decomposition is performed on the sequence with added white noise to obtain the preliminary decomposition results for each repeated cycle. The eigenmode functions of the same order in the preliminary decomposition results are averaged to obtain the integrated empirical mode decomposition results of the corresponding order.

4. The method according to claim 3, characterized in that, The empirical mode decomposition of the sequence after adding white noise to obtain preliminary decomposition results for each repeated round includes: Using the signal extremum detection method, the local extremum points of the sequence after adding white noise are found. Based on the local extremum points, upper and lower envelopes are constructed by spline interpolation to obtain the envelope mean curve. The difference between the sequence after adding white noise and the envelope mean curve is used as a new signal, and the decomposition is repeated until a preset decomposition stopping condition is met to determine the preliminary decomposition result.

5. The method according to claim 1, characterized in that, The step of filtering the intrinsic mode function sequence using a preset similarity algorithm to obtain the filtered sequence includes: The similarity value between the intrinsic mode function sequence and the historical power sequence is calculated using a preset similarity algorithm; Based on the similarity values, the intrinsic mode function sequence with the highest similarity to the historical power sequence is determined as the sequence of the target order; sequences from the first order to the target order are retained as the filtered sequences.

6. The method according to claim 1, characterized in that, The step of updating the initial training sample set according to the filtered sequence to obtain the target training sample set includes: The residual sequence is updated based on the filtered sequence to obtain the updated residual sequence; The initial training sample set is updated based on the filtered sequence and the updated residual sequence to obtain the target training sample set.

7. The method according to claim 1, characterized in that, Before the step of inputting the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model, the method further includes: Obtain the power sequence to be simulated for the target wind farm; The power sequence to be simulated is decomposed using integrated empirical mode decomposition technology, and the decomposition results are filtered using a preset similarity algorithm to obtain the intrinsic mode function sequence and the residual sequence to be simulated. The intrinsic mode function sequence to be simulated, the residual sequence to be simulated, and the meteorological sequence are spliced ​​together to form a multivariate input sequence, so as to determine the spliced ​​sequence to be simulated.

8. A wind power simulation system based on integrated empirical mode decomposition and residual multi-scale temporal convolutional sparse expert model, characterized in that, The system includes: The data acquisition module is used to acquire the historical power dataset and meteorological dataset of the target wind farm, and to perform sliding window processing on the historical power dataset and the meteorological dataset to construct an initial training sample set. The signal decomposition module is used to decompose the historical power sequence in the initial training sample set into a series of intrinsic mode function sequences and a residual sequence using integrated empirical mode decomposition technology, so as to obtain the integrated empirical mode decomposition result; The similarity filtering module is used to filter the intrinsic mode function sequence using a preset similarity algorithm to obtain a filtered sequence, and to update the initial training sample set according to the filtered sequence to obtain the target training sample set. The model building and training module is used to build a residual multi-scale temporal convolutional sparse expert model and train the residual multi-scale temporal convolutional sparse expert model based on the target training sample set. The model simulation module is used to input the spliced ​​sequence to be simulated into the trained residual multi-scale temporal convolutional sparse expert model, so as to output the simulation results of intrinsic mode functions and residual simulation results. The power simulation module is used to add the simulation results of the intrinsic mode functions and the simulation results of the residuals to obtain the wind power simulation results.

9. A computer device, characterized in that, The computer device includes: At least one processor and memory; The memory is used to store program code, and the processor is used to call the program code stored in the memory to execute the method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, It includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images