A method and storage medium for ultra-short-term wind power prediction based on wavelet decomposition

Abstract

The present invention relates to an ultra-short-term wind power prediction method and storage medium based on wavelet decomposition, wherein the method comprises: obtaining a time series of historical wind power data corresponding to a wind farm at a time when ultra-short-term wind power is to be predicted and a time series of historical meteorological data corresponding to the wind farm; performing wavelet decomposition on the time series of historical wind power data, inputting each wavelet component and the time series of meteorological data into a pre-established ultra-short-term prediction model for wind farm power generation, and obtaining predicted values of each wavelet component at the time to be predicted; and performing wavelet reconstruction on the predicted values of each wavelet component to obtain an ultra-short-term wind power prediction value at the time to be predicted. The technical solution provided by the present invention first performs wavelet decomposition on the time series of wind power data, and then substitutes the wavelet components into a pre-established ultra-short-term prediction model for wind farm power generation, thereby obtaining an ultra-short-term wind power prediction value, thereby improving the accuracy of ultra-short-term wind power prediction.

Description

Technical Field

[0001] The present invention relates to the field of power prediction, and in particular to an ultra-short-term wind power prediction method and storage medium based on wavelet decomposition. Background Art

[0002] While renewable energy sources such as wind power provide electricity to the power system, the inherent randomness and intermittency of their energy sources significantly impact the safe and reliable operation of the power grid. Effective wind power forecasting provides a crucial basis for renewable energy generation scheduling. However, large wind power forecast errors can hinder peak and frequency regulation in the power system, hindering stable grid operation and reducing the economic efficiency of grid operation. With the rapid development of artificial intelligence (AI) technology, intelligent algorithms, such as deep learning, have played a significant role in load forecasting and renewable energy generation forecasting. Deep learning techniques, with their deep architecture of multi-layered nonlinear mapping, can fully learn the connections and patterns between input features. Time series analysis is a typical wind power forecasting method, essentially relying on the integration of historical information for reasonable extrapolation. Long-short-term memory networks, a combination of deep learning and time series, are frequently used for ultra-short-term wind power forecasting.

[0003] However, wind power, as data with frequent fluctuations, usually contains high-frequency components generated by a large number of high-frequency random disturbances, and these high-frequency components are unpredictable. The existing technology does not consider the adverse effects of high-frequency components on wind power forecasting, resulting in poor accuracy of ultra-short-term wind power forecasting. Summary of the Invention

[0004] The present application provides an ultra-short-term wind power prediction method and storage medium based on wavelet decomposition, so as to at least solve the technical problem of poor accuracy of ultra-short-term wind power prediction in related technologies.

[0005] The first embodiment of the present application proposes an ultra-short-term wind power prediction method based on wavelet decomposition, the method comprising:

[0006] Obtaining a time series of historical wind power data corresponding to the wind farm at the time when the ultra-short-term wind power is to be predicted and a time series of historical meteorological data corresponding to the wind farm;

[0007] Performing wavelet decomposition on the historical wind power data time series to obtain wavelet components corresponding to the historical wind power data time series;

[0008] Inputting the wavelet components and the time series of historical meteorological data into a pre-established ultra-short-term prediction model for wind farm power generation to obtain the predicted values of the wavelet components at the time to be predicted;

[0009] The wavelet component prediction values are subjected to wavelet reconstruction to obtain the ultra-short-term wind power prediction value at the time to be predicted.

[0010] Preferably, the meteorological data is meteorological data that is strongly correlated with the wind power data.

[0011] Preferably, before inputting the wavelet components and the historical meteorological data time series into a pre-established ultra-short-term prediction model for wind farm power generation, the method further includes:

[0012] The wavelet components and the data in the historical meteorological data time series are standardized.

[0013] Preferably, the process of establishing the ultra-short-term prediction model for wind farm power generation includes:

[0014] Obtain the wind power data time series and wind speed data time series of the wind farm within the historical period;

[0015] Performing wavelet decomposition on the wind power data time series and the meteorological data time series to obtain a low-frequency trend component and a high-frequency component of the wind power data time series, and a low-frequency trend component and a high-frequency component of the meteorological data time series;

[0016] The low-frequency trend component and high-frequency component of the wind power data time series, and the low-frequency trend component and high-frequency component of the meteorological data time series are input into the optimized long short-term memory network, and trained using a weight-sharing learning method to obtain a trained wind farm power generation ultra-short-term prediction model.

[0017] Furthermore, the optimization of the long short-term memory network includes:

[0018] The bird flock algorithm is used to jointly optimize the number of time steps, initial learning rate, momentum, and number of hidden layer units in the initial long short-term memory network to obtain the optimized long short-term memory network.

[0019] Furthermore, the long short-term memory network consists of an output layer, a hidden layer and an input layer.

[0020] Furthermore, the process of acquiring meteorological data that is strongly correlated with the wind power data includes:

[0021] Obtaining the joint probability density function between wind power data and various meteorological data, the marginal probability density function of wind power data, and the marginal probability density function of meteorological data;

[0022] Determining a mutual information value between the wind power data and each meteorological data based on a joint probability density function between the wind power data and each meteorological data, a marginal probability density function of the wind power data, and a marginal probability density function of the meteorological data;

[0023] The meteorological data whose mutual information value is greater than a preset mutual information threshold is regarded as meteorological data that is strongly correlated with the wind power data.

[0024] Furthermore, the standardization of the wavelet components and the data in the historical meteorological data time series includes:

[0025] The z-score standardization method is used to standardize the data in the wavelet components and historical meteorological data time series.

[0026] Preferably, performing wavelet reconstruction on the wavelet component prediction values to obtain the ultra-short-term wind power prediction value at the time to be predicted includes:

[0027] The predicted values of the wavelet components are added and denormalized, and the denormalized result is used as the ultra-short-term wind power prediction value at the time to be predicted.

[0028] The second embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the prediction method of the first aspect of the present application is implemented.

[0029] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects:

[0030] The present invention provides an ultra-short-term wind power prediction method and storage medium based on wavelet decomposition, wherein the method includes: obtaining a time series of historical wind power data corresponding to a wind farm at a time when ultra-short-term wind power is to be predicted and a time series of historical meteorological data corresponding to the wind farm; performing wavelet decomposition on the time series of historical wind power data to obtain wavelet components corresponding to the time series of wind power data; inputting the wavelet components and the time series of historical meteorological data into a pre-established ultra-short-term prediction model for wind farm power generation to obtain predicted values of the wavelet components at the time to be predicted; and performing wavelet reconstruction on the predicted values of the wavelet components to obtain the ultra-short-term wind power prediction value at the time to be predicted. The technical solution provided by the present invention first performs wavelet decomposition on the time series of historical wind power data, and then substitutes the wavelet components into a pre-established ultra-short-term prediction model for wind farm power generation to obtain the ultra-short-term wind power prediction value, thereby improving the accuracy of ultra-short-term wind power prediction.

[0031] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0033] Figure 1 This is a flow chart of an ultra-short-term wind power prediction method based on wavelet decomposition provided according to one embodiment of the present application;

[0034] Figure 2 1 is a waveform diagram of a wavelet function and a scaling function provided according to an embodiment of the present application;

[0035] Figure 3 1 is a schematic diagram of a unit structure of a long short-term memory network provided according to one embodiment of the present application;

[0036] Figure 4 2 is a schematic diagram of a multi-task learning mechanism with a hard sharing mechanism according to an embodiment of the present application;

[0037] Figure 5 This is a detailed flow chart of an ultra-short-term wind power prediction method based on wavelet decomposition provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0038] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0039] The present application proposes a method and storage medium for ultra-short-term wind power prediction based on wavelet decomposition, wherein the method includes: obtaining a time series of historical wind power data corresponding to a wind farm at a time when ultra-short-term wind power is to be predicted and a time series of historical meteorological data corresponding to the wind farm; performing wavelet decomposition on the time series of historical wind power data to obtain wavelet components corresponding to the time series of historical wind power data; inputting the wavelet components and the time series of historical meteorological data into a pre-established ultra-short-term prediction model for wind farm power generation to obtain predicted values of the wavelet components at the time to be predicted; and performing wavelet reconstruction on the predicted values of the wavelet components to obtain the ultra-short-term wind power predicted value at the time to be predicted. The technical solution provided by the present invention first performs wavelet decomposition on the time series of historical wind power data, and then substitutes the wavelet components into a pre-established ultra-short-term prediction model for wind farm power generation to obtain the ultra-short-term wind power predicted value, thereby improving the accuracy of ultra-short-term wind power prediction.

[0040] The following describes an ultra-short-term wind power prediction method and storage medium based on wavelet decomposition according to an embodiment of the present application with reference to the accompanying drawings.

[0041] Example 1

[0042] Figure 1 A flow chart of an ultra-short-term wind power prediction method based on wavelet decomposition provided in an embodiment of the present disclosure is shown in FIG. Figure 1 As shown, the method includes:

[0043] Step 1: Obtain a time series of historical wind power data corresponding to the wind farm at the time when the ultra-short-term wind power is to be predicted and a time series of historical meteorological data corresponding to the wind farm; the time series of historical wind power data and the time series of historical meteorological data corresponding to the wind farm are N time periods before the time when the ultra-short-term wind power is to be predicted.

[0044] It should be noted that the meteorological data is meteorological data that is strongly correlated with the wind power data.

[0045] The process of acquiring meteorological data that is strongly related to the wind power data includes:

[0046] Obtaining the joint probability density function between wind power data and various meteorological data, the marginal probability density function of wind power data, and the marginal probability density function of meteorological data;

[0047] The mutual information value between the wind power data and each meteorological data is determined based on the joint probability density function between the wind power data and each meteorological data, the marginal probability density function of the wind power data and the marginal probability density function of the meteorological data; wherein, the mutual information value between the wind power data and each meteorological data is determined using the formula Determine the mutual information value between the wind power data and each meteorological data, fX (x), f Y (y) is the marginal probability density function, f X,Y (x, y) is the joint probability density function of the two input features, and MI(X; Y) is the mutual information value between X and Y. The higher the mutual information value, the closer the corresponding meteorological factor is to wind power.

[0048] The meteorological data whose mutual information value is greater than a preset mutual information threshold is regarded as meteorological data that is strongly correlated with the wind power data.

[0049] Step 2: performing wavelet decomposition on the historical wind power data time series to obtain wavelet components corresponding to the historical wind power data time series;

[0050] It should be noted that the Mallet algorithm is used to decompose the time series into a low-frequency trend component and a high-frequency component.

[0051] Furthermore, the time series is decomposed by Daubechies (db) 4 wavelet to obtain the low-frequency trend component and the high-frequency component corresponding to the time series.

[0052] The db4 wavelet has no definite formula. Its wavelet function is determined by the scaling function and the filter coefficient. The waveforms of the db4 wavelet function ψ(t) and the scaling function φ(t) are as follows: Figure 2 shown.

[0053] Assuming X(t) is a square integrable signal, the continuous wavelet transform formula of X(t) is as follows:

[0054]

[0055] Where, It is called the base wavelet, and α and β are the scale parameter and displacement parameter of the wavelet transform respectively. Because the continuous wavelet decomposition has the problem of information redundancy and the computational complexity is large, which does not meet the requirements of rapid wind power prediction, this patent performs discrete wavelet transform on X(t), and makes The formula is as follows:

[0056]

[0057] In the formula, WT X is the signal after discrete wavelet transformation, X(t) is the input signal, v j,k (t) is the discrete wavelet, j and k are the scale parameter and displacement parameter of discrete wavelet transform respectively.

[0058] In the embodiment of the present disclosure, before step 3, the method further includes:

[0059] The wavelet components and the data in the historical meteorological data time series are standardized.

[0060] The z-score standardization method is used to standardize the wavelet components and the data in the historical meteorological data time series.

[0061] For example, using the formula Standardization is performed, and each feature after standardization satisfies the normal distribution. X is the actual value in the time series, and X' is the value after the actual value in the time series is standardized. is the average value of each data in the time series, X i is the actual value at the i-th moment in the time series, and N is the total number of moments in the time series.

[0062] Step 3: Inputting the wavelet components and the time series of historical meteorological data into a pre-established ultra-short-term prediction model for wind farm power generation to obtain the predicted values of the wavelet components at the time to be predicted;

[0063] It should be noted that the process of establishing the ultra-short-term prediction model for wind farm power generation includes:

[0064] Obtain the wind power data time series and wind speed data time series of the wind farm within the historical period;

[0065] Performing wavelet decomposition on the wind power data time series and the meteorological data time series to obtain a low-frequency trend component and a high-frequency component of the wind power data time series, and a low-frequency trend component and a high-frequency component of the meteorological data time series;

[0066] The low-frequency trend component and high-frequency component of the wind power data time series, and the low-frequency trend component and high-frequency component of the meteorological data time series are input into the optimized long short-term memory network, and trained using a weight-sharing learning method to obtain a trained wind farm power generation ultra-short-term prediction model.

[0067] Furthermore, the optimization of the long short-term memory network includes:

[0068] 1) The bird flock algorithm is used to jointly optimize the number of time steps, initial learning rate, momentum, and number of hidden layer units in the initial long short-term memory network to obtain the optimized long short-term memory network.

[0069] The long short-term memory network consists of an output layer, a hidden layer and an input layer.

[0070] For example, the construction of the model specifically includes:

[0071] Construct a long short-term memory network and jointly optimize the four important hyperparameters of the long short-term memory network through the bird's flock algorithm to achieve the best prediction effect;

[0072] Among them, the long short-term memory network can be regarded as an improved model of the ordinary recurrent neural network (RNN), which solves the problems of gradient explosion and gradient disappearance in ordinary RNN networks, can perform long-term learning, and make full use of historical information. The long short-term memory network consists of an output layer, a hidden layer, and an input layer. Unlike the RNN with a single hidden layer, the LSTM stores information in a control unit outside the normal information flow of the RNN, and introduces a new state unit C. The memory of the previous sequence is controlled by the forget gate f. The working memory is used for output, and the output gate controls the part to be written in the current memory. The input gate i is responsible for controlling the current state information h to be written to the memory unit. t-1 and the current input X t Current status information h t-1 and the current input X t The unit structure of the long short-term memory network is determined by linear combination and a nonlinear activation function. Figure 3 As shown; the calculation of each variable in the figure can be: f t =σ(W f ·[h t-1 ,x t ]+b f ),i t =σ(W i ·[h t-1 ,x t ]+b i ), h t =σ(W o [h t-1 ,x t ]+b0)*tanh(C t ), Where: W f is the weight matrix used to control the behavior of the forget gate, W C is the output vector of the connected neuron activation function, W i is the weight matrix of the input gate connection, W o To connect X t The weight matrix of h t-1 is the output signal of the hidden layer at time t-1, h t is the output signal of the hidden layer at time t; x t is the output of the input layer at time t, b f , b i , b cAnd b0 is the bias vector; σ, tanh are activation functions; C t , C * t , C t-1 It is a temporary state, jointly deciding whether to forget the previous information; f t is the current memory module state value, i t The impact of current data input on the state of the memory module.

[0073] Furthermore, the hyperparameters optimized by the bird flock algorithm specifically include:

[0074] Since the four parameters of the long short-term memory network, namely the number of time steps, initial learning rate, momentum, and number of hidden layer units, jointly affect the prediction effect, the bird flock algorithm is used to jointly optimize these four parameters.

[0075] The Bird Flock Algorithm (BFA) is a swarm intelligence optimization algorithm based on the foraging, vigilance, and flight behaviors of birds. Compared to traditional optimization algorithms like genetic algorithms, it boasts faster convergence and greater robustness. The algorithm's fitness function is the root mean square error of the prediction results from the training data. The BFA parameters are: dimension 4, maximum number of iterations 60, population size 30, individual cognition coefficient 1.6, group cognition coefficient 1.6, and migration period 3.

[0076] As a search algorithm, the Bird's Flock Algorithm (BFA) optimizes the four hyperparameter values, resulting in slight fluctuations. To minimize uncertainty, this example uses the average of five optimization results as the basis for selecting parameters for the LSTM network. After optimization using the BFA, the parameter values are assigned to the LSTM network. Testing with iterations ranging from 50 to 300 revealed that 200 iterations was the optimal training method.

[0077] 2) Using a weight-sharing learning method to train the optimized long short-term memory network includes:

[0078] The multi-task learning mechanism is used to perform multi-task parallel learning to obtain an ultra-short-term prediction model for wind farm power generation.

[0079] For example, the high-frequency and low-frequency components D obtained by decomposition are used as the input of each long short-term memory network LSTM respectively. A weight-sharing learning method is adopted. Each component is input into the corresponding long short-term memory network for prediction. Then, the predicted values of each component are summed and denormalized to obtain the actual predicted value. The error analysis of the obtained predicted value is performed until the error is less than the preset error threshold and the training is stopped.

[0080] Furthermore, the hard sharing mechanism in the multi-task learning mechanism has the characteristics that multiple subtasks share the same feature sharing layer and the feature parameters are exactly the same. For the ultra-short-term prediction model of wind farm power generation to be constructed, which has many parameters and complex structure, it is less likely to cause overfitting problems and the model generalization ability is also stronger. Figure 4 Shown is a diagram of the multi-task learning mechanism of the hard sharing mechanism.

[0081] The multi-task learning mechanism consists of a set of tasks y t , where t∈M and the data set Where N is the number of tasks and the number of data samples in the dataset, then the prediction function f is defined t (x,Θ s ,Θ t ):x-→y t It is worth noting that the equal-weight loss function of the multi-input multi-output network model, this application uses auxiliary tasks to improve the generalization ability of the main task, the overall optimization loss function of the multi-task learning mechanism can be defined as follows: Where: Θ s ,Θ t They are parameters shared between different tasks and parameters related to the task. j i is the mutual information value of the jth component of the i-th task sequence with respect to the corresponding component of the main task sequence, w j i is the weight coefficient of the i-th task of component j, L j i is the i-th task loss function of the j-th component, L j total is the total loss function of the multi-task learning mechanism of the jth component of this application.

[0082] In the process of solving a problem, the multi-task learning mechanism can use shared layers to learn and obtain auxiliary coupling information provided by other related subtasks, thereby achieving the purpose of improving the model output accuracy and enhancing the generalization ability.

[0083] Among them, this application takes the historical wind power time series as the main task, and uses meteorological information such as wind speed and wind direction as auxiliary tasks to improve the prediction accuracy of wind power forecasting.

[0084] Step 4: Perform wavelet reconstruction on the predicted values of each wavelet component to obtain the ultra-short-term wind power prediction value at the time to be predicted.

[0085] In the embodiment of the present disclosure, step 4 includes:

[0086] The predicted values of the wavelet components are added and denormalized, and the denormalized result is used as the ultra-short-term wind power prediction value at the time to be predicted.

[0087] In the embodiment of the present disclosure, the detailed flow chart of the above method is as follows: Figure 5 shown.

[0088] In summary, the embodiment of the present disclosure provides an ultra-short-term wind power prediction method based on wavelet decomposition, which first performs wavelet decomposition on the time series of historical wind power data, and then substitutes the wavelet components into a pre-established ultra-short-term prediction model of wind farm power generation, thereby obtaining an ultra-short-term wind power prediction value, thereby improving the accuracy of ultra-short-term wind power prediction.

[0089] Example 2

[0090] In order to implement the above embodiments, the present disclosure also proposes a computer-readable storage medium.

[0091] The computer device provided in this embodiment stores a computer program, which implements the method in Example 1 when executed by a processor.

[0092] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0093] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0094] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for ultra-short-term wind power prediction based on wavelet decomposition, characterized in that: The method comprises: Obtaining a time series of historical wind power data corresponding to the wind farm at a time when ultra-short-term wind power is to be predicted and a time series of historical meteorological data corresponding to the wind farm, wherein the meteorological data is meteorological data that is strongly correlated with the wind power data; Performing wavelet decomposition on the historical wind power data time series to obtain wavelet components corresponding to the historical wind power data time series; Inputting the wavelet components and the time series of historical meteorological data into a pre-established ultra-short-term prediction model for wind farm power generation to obtain the predicted values of the wavelet components at the time to be predicted; Performing wavelet reconstruction on the predicted values of each wavelet component to obtain an ultra-short-term wind power prediction value at the time to be predicted; The process of establishing the ultra-short-term prediction model for wind farm power generation includes: Obtain the wind power data time series and wind speed data time series of the wind farm within the historical period; Performing wavelet decomposition on the wind power data time series and the meteorological data time series to obtain a low-frequency trend component and a high-frequency component of the wind power data time series, and a low-frequency trend component and a high-frequency component of the meteorological data time series; The low-frequency trend component and high-frequency component of the wind power data time series, and the low-frequency trend component and high-frequency component of the meteorological data time series are input into the optimized long short-term memory network, and trained using a weight-sharing learning method to obtain a trained wind farm power generation ultra-short-term prediction model.

2. The method according to claim 1, wherein Before inputting the wavelet components and the historical meteorological data time series into the pre-established ultra-short-term prediction model for wind farm power generation, the method further includes: The wavelet components and the data in the historical meteorological data time series are standardized.

3. The method according to claim 1, wherein The optimization of the long short-term memory network includes: The bird flock algorithm is used to jointly optimize the number of time steps, initial learning rate, momentum, and number of hidden layer units in the initial long short-term memory network to obtain the optimized long short-term memory network.

4. The method according to claim 1, wherein The long short-term memory network consists of an output layer, a hidden layer and an input layer.

5. The method according to claim 1, wherein The process of acquiring meteorological data that is strongly correlated with the historical wind power data includes: Obtaining the joint probability density function between wind power data and various meteorological data, the marginal probability density function of wind power data, and the marginal probability density function of meteorological data; Determining a mutual information value between the wind power data and each meteorological data based on a joint probability density function between the wind power data and each meteorological data, a marginal probability density function of the wind power data, and a marginal probability density function of the meteorological data; The meteorological data whose mutual information value is greater than a preset mutual information threshold is regarded as meteorological data that is strongly correlated with the wind power data.

6. The method according to claim 2, wherein The standardization of the wavelet components and the data in the historical meteorological data time series includes: The z-score standardization method is used to standardize the data in the wavelet components and historical meteorological data time series.

7. The method according to claim 1, wherein The wavelet reconstruction of the wavelet component prediction values to obtain the ultra-short-term wind power prediction value at the time to be predicted includes: The predicted values of the wavelet components are added and denormalized, and the denormalized result is used as the ultra-short-term wind power prediction value at the time to be predicted.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images