A scenario-based short-term wind speed prediction method and system for green data centers

Abstract

The present invention discloses a scenario-based short-term wind speed prediction method and system for green data centers, the method comprising the following steps: collecting historical weather data, clustering the historical weather data, and obtaining several weather scenarios; constructing Copula functions of wind speed and other meteorological factors in each scenario under different weather scenarios, and determining the optimal input variables of wind speed and other meteorological factors based on the Copula function; constructing an LSTM neural network model, and training the LSTM neural network model based on historical weather data and optimal input variables to obtain a wind speed prediction model; obtaining weather factor data of the scenario to be predicted, and using the wind speed prediction model to predict the wind speed data of the next period. The present invention effectively solves the problem that a single scenario prediction method is difficult to accurately capture the law of wind speed changes when facing complex and changeable weather parameters by introducing technical means such as fuzzy clustering, cluster number evaluation, and Copula function correlation analysis.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind speed prediction, and in particular relates to a scenario-based short-term wind speed prediction method and system applied to a green data center. Background Art

[0002] In the energy industry and other fields, the accuracy and reliability of wind speed prediction are crucial to optimizing operations and improving efficiency. In the energy structure of green data centers, clean energy, especially wind energy, as an efficient and environmentally friendly form of energy, occupies a pivotal position. However, the inherent intermittent and unpredictable characteristics of wind energy pose a significant challenge to its stable application in the power supply of data centers. In view of this, wind speed prediction technology plays a vital role in the operation of green data centers. Through wind speed prediction, data center managers can gain insight into the fluctuation trend of wind energy supply in advance, and then implement refined power dispatching strategies and energy storage planning. When wind energy resources are abundant, increase the proportion of wind power supply and reduce dependence on fossil fuels; and in the period of wind energy shortage, quickly start backup power or flexibly adjust power demand to ensure the continuity and stability of data center operation. In addition, wind speed prediction also promotes the optimization of resource allocation, enabling data centers to accurately configure energy storage facilities and backup power supplies, avoid capital redundancy and resource waste, and guide the site selection and layout of data centers to tilt towards wind energy-rich areas to maximize the utilization of wind energy resources. From an economic perspective, wind energy, as a low-cost clean energy, combined with effective wind speed prediction and power management, can help significantly reduce the operating costs of data centers, reduce carbon emissions, and enhance the environmental image and social responsibility of enterprises. Wind speed prediction has become an indispensable technical support for green data centers to achieve sustainable development and move towards a more environmentally friendly and efficient future, and it is essential to promote the data center industry to actively respond to global climate change challenges. Therefore, the development of green data centers in the future should continue to strengthen the research and development and application of wind speed prediction technology to effectively deal with the intermittent and uncertain nature of wind power generation and promote the green transformation and sustainable development of the data center industry.

[0003] The name of a similar technical patent is: A method for ultra-short-term wind speed prediction, and the application number is 202211696874.6. The patent proposes an ultra-short-term wind speed prediction method, which clusters the wind speed modeling sequence to obtain training data sets of different categories; according to the different categories of the training data sets, a multi-step prediction model based on an extreme learning machine is established to perform multi-step wind speed prediction and output a prediction sequence. However, this method does not introduce an accurate cluster number evaluation indicator, and the random selection of the number of clusters affects the subsequent prediction accuracy. In addition, this method does not take into account other weather parameters that affect wind speed, which in turn affects the accuracy and reliability of the prediction.

[0004] The name of a similar technology patent is: A wind speed prediction method based on BP algorithm, and the application number is 202110142012.8. The patent proposes a wind speed prediction method based on BP algorithm, which divides the wind speed data set into a training set and a test set, sets features and labels respectively, and feeds the BP neural network for training to obtain a high-precision wind speed prediction result. This method does not perform clustering on the data set, and the prediction accuracy cannot be guaranteed when the data is heterogeneous. Other weather parameters that affect wind speed are also not considered. Summary of the invention

[0005] The present invention aims to solve the deficiencies of the prior art and provides the following solutions:

[0006] A scenario-based short-term wind speed prediction method for a green data center includes the following steps:

[0007] Collect historical weather data, cluster the historical weather data, and obtain several weather scenes;

[0008] Under different weather scenarios, constructing Copula functions of wind speed and other meteorological factors in each scenario, and determining optimal input variables of wind speed and other meteorological factors based on the Copula functions;

[0009] Constructing an LSTM neural network model, and training the LSTM neural network model based on the historical weather data and the optimal input variables to obtain a wind speed prediction model;

[0010] The weather factor data of the scene to be predicted is obtained, and the wind speed prediction model is used to predict the wind speed data of the next period.

[0011] Preferably, the clustering method comprises:

[0012] Using the FDM clustering method to cluster the historical weather data;

[0013] The XB evaluation index is introduced to divide the historical weather data into several weather scenes.

[0014] Preferably, the clustering processing method includes:

[0015] S1.1.1. The historical weather data X = {x 1 , x 2 , ..., x k} is divided into c category, x k Represents the k-th dimension of data to be classified;

[0016] S1.1.2. Initialize the cluster centers based on the classified data:

[0017] V={v1 ,v 2 ,...,v e}

[0018] Among them, V represents the set of c cluster centers, v c represents the cth cluster center;

[0019] S1.1.3. Calculate the membership matrix:

[0020]

[0021] U={u ik}

[0022] Among them, k represents the dimension to which the data point belongs, U represents the membership matrix, and u ik Represents data x k The membership degree to class i, v i represents the i-th cluster center, v j represents the jth cluster center, d ik represents the Euclidean distance from the k-th dimension data to the i-th cluster center, and m represents the weighted index;

[0023] S1.1.4. Update the cluster center based on the membership matrix:

[0024]

[0025] S1.1.5. Based on the cluster center and the membership matrix, the FDM clustering is used to obtain the minimized objective function:

[0026]

[0027] Repeat S1.1.3 and S1.1.4 until the minimization objective function converges and clustering is completed.

[0028] Preferably, the method of dividing the weather scenes comprises:

[0029] Introduce the improved XB index to select the optimal number of clusters and determine the clustering effectiveness index V zwj :

[0030]

[0031] Where d represents the Euclidean distance between two cluster centers;

[0032] The historical weather data is classified into several scenes using the clustering validity index, and the clustering results are arranged in order from small to large according to the cluster center, thereby completing the scene division of the multi-dimensional random variable.

[0033] Preferably, the method for obtaining the optimal input variable includes:

[0034] Constructing Copula functions of wind speed and other meteorological factors in each scenario, and estimating the optimal parameters of each Copula function using maximum likelihood estimation method;

[0035] Selecting an optimal Copula function from the Copula functions using an empirical Copula function;

[0036] Calculate the Spearman coefficient and Kendall coefficient of wind speed and other meteorological factors under different weather scenarios based on the optimal Copula function;

[0037] The optimal input variables are determined based on the Spearman coefficient and the Kendall coefficient.

[0038] Preferably, the method for selecting the optimal Copula function includes:

[0039] Construct the empirical Copula function:

[0040]

[0041] Where I represents the empirical function, F(x i ) and G(y i ) represents the marginal distribution function of meteorological factors;

[0042] Calculate the Euclidean distance between the CDF value of each Copula function and the CDF value of the empirical Copula function:

[0043]

[0044] Among them, d Eu represents the Euclidean distance, C n Represents the CDF value of the Copula function, C p Represents the CDF value of the empirical Copula function;

[0045] When the Euclidean distance is minimum, the optimal Copula function is obtained.

[0046] Preferably, the Spearman coefficient and the Kendall coefficient are:

[0047]

[0048] Among them, ρ s is the Spearman coefficient, τ ij is the Kendall coefficient.

[0049] Preferably, the method for determining the optimal input variable includes: drawing a graph with the number of variables as the horizontal axis and the Spearman coefficient and the Kendall coefficient as the vertical axis, and determining whether there is a clear inflection point in the graph. If so, selecting data with a large comprehensive correlation coefficient value as the optimal input variable.

[0050] Preferably, the method for obtaining the wind speed prediction model includes:

[0051] Constructing an LSTM neural network model, dividing the historical weather data into a training set and a test set;

[0052] Determine the optimal input variables under each weather scenario, set 24-hour weather parameters as features of the training set, and set the wind speed in the last 15 minutes as a label of the training set;

[0053] Based on the training set and the optimal input variables, training the LSTM neural network model in respective scenarios;

[0054] The LSTM neural network model is tested using the test set, and the root mean square error RMSE and the coefficient of determination R of the test set prediction results are calculated. 2 , if the root mean square error RMS and the determination coefficient R 2 If all meet the preset values, the model training is completed and the wind speed prediction model is obtained.

[0055] The present invention also provides a scenario-based short-term wind speed prediction system for a green data center, wherein the prediction system applies any of the above-mentioned prediction methods and comprises: a scenario clustering module, a variable confirmation module, a model training module and a prediction module;

[0056] The scene clustering module is used to collect historical weather data, cluster the historical weather data, and obtain several weather scenes;

[0057] The variable confirmation module is used to construct the Copula function of wind speed and other meteorological factors in each scene under different weather scenes, and determine the optimal input variables of wind speed and other meteorological factors based on the Copula function;

[0058] The model training module is used to construct an LSTM neural network model, and train the LSTM neural network model based on the historical weather data and the optimal input variables to obtain a wind speed prediction model;

[0059] The prediction module is used to obtain weather factor data of the scene to be predicted, and use the wind speed prediction model to predict the wind speed data of the next period.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The scenario-based short-term wind speed prediction algorithm proposed in the present invention has shown significant improvement in prediction accuracy. Under the same model architecture and data processing conditions, the present invention effectively solves the problem that a single scenario prediction method is difficult to accurately capture the law of wind speed changes when facing complex and changeable weather parameters by introducing technical means such as fuzzy clustering, cluster number evaluation, and Copula function correlation analysis; through fuzzy clustering, the present invention divides the original weather data into several scenarios with similar characteristics, so that the wind speed prediction in each scenario is more targeted and accurate. At the same time, through the new cluster number evaluation index, the present invention can automatically determine the optimal number of clusters, avoiding the subjectivity and uncertainty brought about by artificially setting the number of clusters; in addition, the correlation analysis method based on the Copula function enables the present invention to accurately identify key variables that have an important impact on wind speed prediction, further improving the performance of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0063] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0064] Figure 2 Schematic diagram of the structure of the LSTM neural network model of an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] Embodiment 1

[0068] In this embodiment, if Figure 1 As shown, a scenario-based short-term wind speed prediction method for a green data center includes the following steps:

[0069] S1. Collect historical weather data, cluster the historical weather data, and obtain several weather scenarios.

[0070] Clustering methods include:

[0071] S1.1. Use the FDM clustering method to cluster historical weather data. The clustering method includes:

[0072] S1.1.1. Historical weather data X = {x 1 , x 2 , ..., x k} is divided into c category, x k Represents the k-th dimension of data to be classified;

[0073] S1.1.2. Initialize the cluster centers based on the classified data:

[0074] V={v 1 ,v 2 ,...,v c}

[0075] Among them, V represents the set of c cluster centers, v c represents the cth cluster center;

[0076] S1.1.3. Calculate the membership matrix:

[0077]

[0078] U={u ik}

[0079] Among them, k represents the dimension to which the data point belongs, U represents the membership matrix, and u ik Represents data x k The membership degree to class i, v i represents the i-th cluster center, v j represents the jth cluster center, d ik represents the Euclidean distance from the k-th dimension data to the i-th cluster center, m represents the weighted index, and in this embodiment, m is 2;

[0080] S1.1.4. Update cluster centers based on membership matrix:

[0081]

[0082] S1.1.5. Based on the cluster center and membership matrix, FDM clustering is used to obtain the minimized objective function:

[0083]

[0084] Repeat S1.1.3 and S1.1.4 until the minimization objective function converges and clustering is completed.

[0085] S1.2. Introduce the XB evaluation index and divide the historical weather data into several weather scenarios.

[0086] The method of dividing several weather scenes includes: introducing the improved XB index to select the optimal number of clusters, determining the clustering effectiveness index V zwj :

[0087]

[0088] Among them, d represents the Euclidean distance between two cluster centers. The historical weather data are classified into several scenarios using the clustering effectiveness index, and the clustering results are arranged in ascending order according to the cluster centers to complete the scenario division of multidimensional random variables.

[0089] S2. Under different weather scenarios, construct the Copula function of wind speed and other meteorological factors in each scenario, and determine the optimal input variables of wind speed and other meteorological factors based on the Copula function.

[0090] Methods for obtaining optimal input variables include:

[0091] S2.1. Construct the Copula functions of wind speed and other meteorological factors in each scenario, and use the maximum likelihood estimation method to estimate the optimal parameters of each Copula function.

[0092] In this embodiment, the Copula functions are Gaussian-Copula, t-Copula, Clayton-Copula, Frank-Copula and Gumbel-Copula functions respectively; the maximum likelihood estimation method is used to find the optimal parameters of the copula function parameters by maximizing the likelihood function of the joint probability density function (PDF) under given data.

[0093] S2.2. Use the empirical Copula function to select the optimal Copula function from the Copula functions.

[0094] Methods for selecting the optimal Copula function include: constructing an empirical Copula function:

[0095]

[0096] Wherein, I represents the empirical function. In this embodiment, the value of I is 0 or 1. When F(x i )≤u, I=1, otherwise I=0, F(x i ) and G(y i) represents the marginal distribution function of meteorological factors; calculate the Euclidean distance between the CDF value of each Copula function and the CDF value of the empirical Copula function:

[0097]

[0098] Among them, d Eu represents the Euclidean distance, C n Represents the CDF value of the Copula function, C p Represents the CDF value of the empirical Copula function; when the Euclidean distance is minimum, the optimal Copula function is obtained.

[0099] S2.3. Based on the optimal Copula function, the Spearman coefficient and Kendall coefficient of wind speed and other meteorological factors under different weather scenarios are calculated, and the average value of the sum of the absolute values ​​of the two correlation coefficients is taken as the final correlation evaluation standard.

[0100] The Spearman coefficient and Kendall coefficient are:

[0101]

[0102] Among them, ρ s is the Spearman coefficient, τ ij is the Kendall coefficient. The range of the two correlation coefficients is (-1, 1). A positive value indicates a positive correlation, and a negative value indicates a negative correlation. The larger the absolute value, the stronger the correlation.

[0103] S2.4. Determine the optimal input variables based on the Spearman coefficient and Kendall coefficient.

[0104] The method for determining the optimal input variables includes: drawing a graph with the number of variables as the horizontal axis and the Spearman coefficient and the Kendall coefficient as the vertical axis, and judging whether there is a clear inflection point in the graph. If so, selecting the data with a large comprehensive correlation coefficient value as the optimal input variable.

[0105] S3. Build an LSTM neural network model and train it based on historical weather data and optimal input variables to obtain a wind speed prediction model.

[0106] Methods for obtaining a wind speed prediction model include:

[0107] S3.1. Build an LSTM neural network model and divide the historical weather data into training and test sets;

[0108] In this embodiment, if Figure 2 As shown, the constructed LSTM neural network model includes: input gate i t、Forget gate t , output gate o t , cell state t , candidate state and memory h t :

[0109] i t =σ(W t ·[h t-1 ,x t ]+b i )

[0110] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0111] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0112]

[0113] h t =o t ×tanh(C t )

[0114] Among them, σ represents the sigmoid activation function; tanh represents the tanh activation function; W represents the weight matrix to be trained; b represents the training bias matrix; h t-1 represents the memory at time t-1; x t Represents the input parameter at time t.

[0115] S3.2. Determine the optimal input variables for each weather scenario, set the 24-hour weather parameters as the features of the training set, and the wind speed in the last 15 minutes as the label of the training set.

[0116] S3.3. Based on the training set and the optimal input variables, train the LSTM neural network model in each scenario.

[0117] S3.4. Use the test set to test the LSTM neural network model and calculate the root mean square error RMSE and determination coefficient R of the test set prediction results 2 , if the root mean square error RMS and the coefficient of determination R 2 If all meet the preset values, the model training is completed and the wind speed prediction model is obtained.

[0118] In this embodiment, the root mean square error RMS and the determination coefficient R 2The calculation formula is:

[0119]

[0120] Among them, Obs i and For i They represent the true value and predicted value of wind speed at time i respectively.

[0121] S4. Obtain weather factor data for the scene to be predicted, and use the wind speed prediction model to predict the wind speed data for the next period.

[0122] Embodiment 2

[0123] In this embodiment, a scenario-based short-term wind speed prediction system applied to a green data center includes: a scenario clustering module, a variable confirmation module, a model training module and a prediction module.

[0124] The scenario clustering module is used to collect historical weather data, cluster the historical weather data, and obtain several weather scenarios; the variable confirmation module is used to construct the Copula function of wind speed and other meteorological factors in each scenario under different weather scenarios, and determine the optimal input variables of wind speed and other meteorological factors based on the Copula function; the model training module is used to build an LSTM neural network model, and train the LSTM neural network model based on historical weather data and optimal input variables to obtain a wind speed prediction model; the prediction module is used to obtain the weather factor data of the scenario to be predicted, and use the wind speed prediction model to predict the wind speed data for the next period of time.

[0125] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A scenario-based short-term wind speed prediction method, characterized in that: The following steps are involved: Collect historical weather data, cluster the historical weather data, and obtain several weather scenes; Under different weather scenarios, constructing Copula functions of wind speed and other meteorological factors in each scenario, and determining optimal input variables of wind speed and other meteorological factors based on the Copula functions; Constructing an LSTM neural network model, and training the LSTM neural network model based on the historical weather data and the optimal input variables to obtain a wind speed prediction model; Obtain weather factor data for the scene to be predicted, and use the wind speed prediction model to predict wind speed data for the next period of time; The clustering method comprises: Using the FDM clustering method to cluster the historical weather data; Introducing the XB evaluation index, dividing the historical weather data into several weather scenarios; The method of dividing the weather scenes comprises: Introduce the improved XB index to select the optimal number of clusters and determine the clustering effectiveness index V zwj : Among them, d represents the Euclidean distance between the two cluster centers, k represents the dimension to which the data point belongs, and u ik Represents data x k The membership degree of class i, c represents the type of historical weather data, v i represents the i-th cluster center, d ik represents the Euclidean distance from the k-th dimension data to the i-th cluster center, and m represents the weighted index; The historical weather data is classified into several scenes using the clustering validity index, and the clustering results are arranged in order from small to large according to the cluster center, thereby completing the scene division of the multi-dimensional random variable.

2. A scenario-based short-term wind speed prediction method according to claim 1, characterized in that: The clustering method comprises: S1.1.

1. The historical weather data X = {x1, x2, ..., x k } is divided into c category, x k Represents the k-th dimension of data to be classified; S1.1.

2. Initialize the cluster centers based on the classified data: V={v1,v2,...,v c } Among them, V represents the set of c cluster centers, v c represents the cth cluster center; S1.1.

3. Calculate the membership matrix: In={in ik } Among them, U represents the membership matrix, v j represents the jth cluster center; S1.1.

4. Update the cluster center based on the membership matrix: S1.1.

5. Based on the cluster center and the membership matrix, the FDM clustering is used to obtain the minimized objective function: Repeat S1.1.3 and S1.1.4 until the minimization objective function converges and clustering is completed.

3. The scenario-based short-term wind speed prediction method according to claim 2, characterized in that: The method for obtaining the optimal input variables includes: Constructing Copula functions of wind speed and other meteorological factors in each scenario, and estimating the optimal parameters of each Copula function using maximum likelihood estimation method; Selecting an optimal Copula function from the Copula functions using an empirical Copula function; Calculate the Spearman coefficient and Kendall coefficient of wind speed and other meteorological factors under different weather scenarios based on the optimal Copula function; The optimal input variables are determined based on the Spearman coefficient and the Kendall coefficient.

4. The scenario-based short-term wind speed prediction method according to claim 3, characterized in that: The method of selecting the optimal Copula function includes: Construct the empirical Copula function: Where I represents the empirical function, F(x i ) and G(y i ) represents the marginal distribution function of meteorological factors; Calculate the Euclidean distance between the CDF value of each Copula function and the CDF value of the empirical Copula function: Among them, d Eu represents the Euclidean distance, C n Represents the CDF value of the Copula function, C p Represents the CDF value of the empirical Copula function; When the Euclidean distance is minimum, the optimal Copula function is obtained.

5. The scenario-based short-term wind speed prediction method according to claim 3, characterized in that: The Spearman coefficient and the Kendall coefficient are: ρ s =12∫0 1 ∫0 1 C(u,v)dudv-3 τ ij =4∫0 1 ∫0 1 C(u,v)dC(u,v)-1 Among them, ρ s is the Spearman coefficient, τ ij is the Kendall coefficient.

6. The scenario-based short-term wind speed prediction method according to claim 3, characterized in that: The method for determining the optimal input variable includes: drawing a graph with the number of variables as the horizontal axis and the Spearman coefficient and the Kendall coefficient as the vertical axis, and judging whether there is a clear inflection point in the graph. If so, selecting data with a large comprehensive correlation coefficient value as the optimal input variable.

7. The scenario-based short-term wind speed prediction method according to claim 1, characterized in that: The method for obtaining the wind speed prediction model includes: Constructing an LSTM neural network model, dividing the historical weather data into a training set and a test set; Determine the optimal input variables under each weather scenario, set 24-hour weather parameters as features of the training set, and set the wind speed in the last 15 minutes as a label of the training set; Based on the training set and the optimal input variables, training the LSTM neural network model in respective scenarios; The LSTM neural network model is tested using the test set, and the root mean square error RMSE and the coefficient of determination R of the test set prediction results are calculated. 2 , if the root mean square error RMS and the determination coefficient R 2 If all meet the preset values, the model training is completed and the wind speed prediction model is obtained.

8. A scenario-based short-term wind speed prediction system, the prediction system applying the prediction method according to any one of claims 1 to 7, characterized in that: include: Scenario clustering module, variable confirmation module, model training module and prediction module; The scene clustering module is used to collect historical weather data, cluster the historical weather data, and obtain several weather scenes; The variable confirmation module is used to construct the Copula function of wind speed and other meteorological factors in each scene under different weather scenes, and determine the optimal input variables of wind speed and other meteorological factors based on the Copula function; The model training module is used to construct an LSTM neural network model, and train the LSTM neural network model based on the historical weather data and the optimal input variables to obtain a wind speed prediction model; The prediction module is used to obtain weather factor data of the scene to be predicted, and use the wind speed prediction model to predict the wind speed data of the next period.

Images