Short-term wind power prediction method, device and equipment based on ensemble learning, medium and product

Abstract

The invention discloses a short-term wind power prediction method and device based on integrated learning, equipment, a medium and a product, and relates to the field of wind power generation prediction, and the method comprises the steps: obtaining wind power data of a wind power plant and corresponding associated feature data; the associated characteristic data comprises wind speed, air temperature, air pressure, humidity, time and air density; preprocessing the associated feature data, and determining the preprocessed associated feature data; constructing a Stacking ensemble learning framework model according to the pre-processed associated feature data and the corresponding wind power data; initializing a zebra population of each hyper-parameter of the Stacking ensemble learning framework model, and determining basic parameters of an improved zebra optimization algorithm; based on the basic parameters, training the Stacking ensemble learning framework model by using the improved zebra optimization algorithm; and utilizing the trained Stacking integrated learning framework model to predict the wind power in the future time period. The wind power prediction method and the wind power prediction device can accurately predict the wind power.

Description

Technical Field

[0001] The present application relates to the field of wind power generation prediction, and particularly to a short-term wind power prediction method, device, equipment, medium and product based on ensemble learning. Background Art

[0002] The development of renewable energy is an important means to combat future climate change and environmental deterioration. The world has a very rich variety of renewable energy sources. As a kind of renewable energy, wind energy has attracted more and more attention due to its clean and abundant characteristics.

[0003] In view of the volatility and intermittency of wind energy, ensuring the stability of wind energy grid connection operation faces a series of problems. Therefore, implementing accurate wind power prediction is extremely important for optimizing the energy management of the power system, improving power supply reliability, and reducing the cost of system reserve capacity. Summary of the Invention

[0004] The purpose of the present application is to provide a short-term wind power prediction method, device, equipment, medium and product based on ensemble learning, which can accurately predict wind power.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In the first aspect, the present application provides a short-term wind power prediction method based on ensemble learning, including:

[0007] Obtain the wind power data of the wind farm and the corresponding associated feature data; the associated feature data includes wind speed, air temperature, air pressure, humidity, time, and air density;

[0008] Preprocess the associated feature data to determine the preprocessed associated feature data;

[0009] Construct a Stacking ensemble learning framework model according to the preprocessed associated feature data and the corresponding wind power data;

[0010] Initialize the zebra population of each hyperparameter of the Stacking ensemble learning framework model to determine the basic parameters of the improved zebra optimization algorithm; the hyperparameters include ridge regression regularization hyperparameter, the number of base learners of XGBoost, the LSTM hidden layer size, the learning rate of LSTM, and the learning rate of ResNet; the basic parameters include the maximum number of iterations and the number of zebra individuals;

[0011] Based on the basic parameters, train the Stacking ensemble learning framework model using the improved zebra optimization algorithm;

[0012] Predict the wind power in the future period by using the trained Stacking integrated learning framework model.

[0013] In a second aspect, the present application provides a short-term wind power prediction device based on integrated learning, including:

[0014] A data acquisition module for acquiring the wind power data of a wind farm and the corresponding various associated feature data; the associated feature data includes wind speed, temperature, air pressure, humidity, time, and air density;

[0015] A data preprocessing module for preprocessing the associated feature data to determine the preprocessed associated feature data;

[0016] A construction module for constructing a Stacking integrated learning framework model according to the preprocessed associated feature data and the corresponding wind power data;

[0017] A basic parameter determination module for initializing the zebra population of each hyperparameter of the Stacking integrated learning framework model and determining the basic parameters of the improved zebra optimization algorithm; the hyperparameters include ridge regression regularization hyperparameters, the number of base learners of XGBoost, the LSTM hidden layer size, the learning rate of LSTM, and the learning rate of ResNet; the basic parameters include the maximum number of iterations and the number of zebra individuals;

[0018] A training module for training the Stacking integrated learning framework model by using the improved zebra optimization algorithm based on the basic parameters;

[0019] A prediction module for predicting the wind power in the future period by using the trained Stacking integrated learning framework model.

[0020] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the short-term wind power prediction method based on integrated learning described in any one of the above.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the short-term wind power prediction method based on integrated learning described in any one of the above.

[0022] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the short-term wind power prediction method based on integrated learning described in any one of the above.

[0023] According to the specific embodiments provided in this application, the following technical effects are disclosed in this application:

[0024] This application considers the wind power data of a wind farm and various associated feature data associated with different wind powers. Starting from multiple data sources, a Stacking integrated learning framework model is constructed. The Stacking integrated learning framework model is stacking generalization, including multiple models, belonging to a hierarchical model fusion strategy, which has better performance than a single model at different time points. It can effectively improve the accuracy of wind power prediction, reduce errors, provide technical support and data support for the stable operation of the power grid, and train the Stacking integrated learning framework model based on the improved zebra optimization algorithm. Using the trained Stacking integrated learning framework model to predict the wind power in the future period, based on the characteristics of the improved zebra optimization algorithm, the prediction accuracy of wind power is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0026] Figure 1 Flowchart of the short-term wind power prediction method based on integrated learning provided by this application;

[0027] Figure 2 Grid analysis diagram of different associated feature data provided by this application;

[0028] Figure 3 Radar analysis diagram of different associated feature data provided by this application;

[0029] Figure 4 Wind power prediction effect diagrams of Model 1 - Model 5 provided by this application;

[0030] Figure 5 Relative error prediction effect diagrams of Model 1 - Model 5 provided by this application;

[0031] Figure 6 Wind power prediction effect diagrams between different models provided by this application;

[0032] Figure 7 Relative error prediction effect diagrams of Model a - Model f provided by this application;

[0033] Figure 8 Schematic diagram of the computer device provided by this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] Next, in combination with the accompanying drawings in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0035] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in combination with the accompanying drawings and specific embodiments.

[0036] The embodiment of the present application provides a short-term wind power prediction method based on ensemble learning. This method is executed by a computer device, specifically, it can be executed alone by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiment of the present application, as Figure 1 shown, this method includes the following steps.

[0037] S1: Obtain the wind power data of the wind farm and the corresponding associated feature data; the associated feature data includes wind speed, temperature, air pressure, humidity, time, and air density.

[0038] S2: Preprocess the associated feature data to determine the preprocessed associated feature data.

[0039] S3: Construct a Stacking ensemble learning framework model according to the preprocessed associated feature data and the corresponding wind power data.

[0040] S4: Initialize the zebra population of each hyperparameter of the Stacking ensemble learning framework model to determine the basic parameters of the improved zebra optimization algorithm; the hyperparameters include the ridge regression regularization hyperparameter, the number of base learners of XGBoost, the LSTM hidden layer size, the learning rate of LSTM, and the learning rate of ResNet; the basic parameters include the maximum number of iterations and the number of zebra individuals;

[0041] S5: Based on the basic parameters, use the improved zebra optimization algorithm to train the Stacking ensemble learning framework model.

[0042] S6: Use the trained Stacking ensemble learning framework model to predict the wind power in the future period.

[0043] In an exemplary embodiment, S1 specifically includes:

[0044] Obtain wind power data of the wind farm with a time scale of 15 minutes for 60 days and various associated feature data. The wind power of the wind farm is P t =[P t-5760 ,P t-5759 ,…,P t-1 T , and the wind speed and other features (temperature, air pressure, humidity, time, air density) are X δ ={x δ (t - 5760),x δ (t - 5759),…x δ (t - 1)}, δ ∈ {1, 2, ..., D}, where D is the number of features.

[0045] In an exemplary embodiment, S2 can be replaced by the following steps.

[0046] S21: Use the wind power data as the reference sequence and the associated feature data as the comparison sequence.

[0047] S22: Perform dimensionless processing on the reference sequence and the comparison sequence respectively to determine the dimensionless processed wind power data and the dimensionless processed associated feature data.

[0048] S23: Based on the dimensionless processed wind power data and the dimensionless processed associated feature data, perform grey relational degree analysis and Pearson correlation coefficient analysis on the various associated feature data, and screen out the associated feature data with an influence degree on the wind power data lower than the set influence degree threshold.

[0049] S24: Determine whether the wind speed is retained in the screened associated feature data. If so, execute S25; if not, execute S26.

[0050] S25: Perform abnormal data processing on the wind speed, and use the associated feature data after abnormal data processing as the preprocessed associated feature data.

[0051] S26: Determine the screened associated feature data as the preprocessed associated feature data.

[0052] In practical applications, the wind power P of the wind farm t =[P t-5760 ,P t-5759 ,…,P t-1 T is used as the reference sequence, and the wind speed and other features (temperature, air pressure, humidity, time, air density) that affect the system behavior are used as the comparison sequence X δ ={x δ (t - 5760),x δ (t - 5759),…x​​δ (t - 1)}, δ ∈ {1, 2, …, D}.

[0053] The Z - score normalization method is respectively used for dimensionless processing of the reference sequence and the comparison sequence, and its mathematical formula is:

[0054]

[0055] In the formula, N is the number of data points; x δ (k), p t (k) are respectively the values of the δ - th feature and the k - th data point of wind power; N is the number of data points; μ δ , μ0 are respectively the averages of all data points of the δ - th feature and wind power; σ δ , σ0 are respectively the standard deviations of the δ - th feature and wind power; r δ (k) and r0(k) are respectively the normalized values of the k - th data point on the δ - th feature and wind power.

[0056] Grey - degree correlation analysis is carried out on each feature affecting wind power, and its mathematical formula is:

[0057]

[0058] In the formula, is the correlation coefficient of the grey - degree correlation of the k - th data point of the δ - th feature; φ is the resolution coefficient, taking values between [0, 1]; is the correlation coefficient of the grey - degree correlation of the δ - th feature.

[0059] Pearson correlation coefficient analysis is carried out on each feature affecting wind power and the absolute value is taken, and its mathematical formula is:

[0060]

[0061] In the formula, is the Pearson correlation coefficient of the δ - th feature; are respectively the means of the δ - th feature and wind power.

[0062] After determining the degree of association of each feature, first, normalization of the degree of association is carried out on it, then the mean value of the correlation coefficients of each feature of the two correlation analysis methods is processed, and then correlation analysis is carried out on each feature, and its mathematical formula is:

[0063]

[0064] In the formula, is the normalized value of the grey - degree correlation analysis and Pearson correlation coefficient of the δ - th feature; It is the correlation coefficient of the δ-th feature after mean processing. Since too many features will increase the complexity of the model, leading to an increase in computational cost and the risk of overfitting, and too few features will cause the model to lose key information, thereby affecting the prediction performance, two features with relatively low correlation with wind power are removed to obtain X δ′ To achieve the prediction effect of the model, where δ′ ∈ {1, 2, 3, 4}.

[0065] If the wind speed feature is successfully retained after feature screening, wind speed anomaly data processing is performed. Anomaly data usually refers to those values that are significantly different from other data points, and these data may be generated due to equipment errors, input errors, or very rare weather conditions. If these outliers are not processed during model training, they may distort the model's learning process, thereby reducing the prediction accuracy of the model under normal conditions. OneSVM is an algorithm that uses support vector machine technology for anomaly detection and is specifically designed for training in the case of only normal sample data. The process of screening wind speed outliers using the OneSVM algorithm is as follows:

[0066] (1) Obtain the 60-day wind speed historical data V with a time scale of 15 minutes from the Xihe Energy Meteorological Big Data Platform of the wind farm t = [v t-5760 , v t-5759 ,..., v t-1 T , and calculate the similarity between every two data points through the RBF kernel function to map the data into a high-dimensional space so that OneSVM can handle nonlinear problems. Its mathematical formula is:

[0067] K(v i , v j ) = exp(-β||v i - v j || 2 )

[0068] In the formula, v i , v j are the wind speeds at the i-th and j-th time points respectively, where i, j ∈ [t - 5760, t - 5759,..., t - 1] represents β as the width parameter of the RBF kernel; K(v i , v j ) represents the similarity between v i and v j .

[0069] (2) Use the objective function of OneSVM to calculate and find the Lagrange multiplier α value corresponding to each time point. Its calculation formula is

[0070] ​

[0071] 0 ≤ α i ≤ C

[0072] Where α is an optimal combined column vector containing Lagrange multipliers for each time node, and α i , α j are Lagrange multipliers at any data point in the 60-day wind speed historical data; C is a positive constant used to control the relaxation degree of the model.

[0073] (3) Use the decision function for decision screening. If the decision function at a certain time point is greater than 0, it is a normal data point; if it is less than 0, it is an abnormal data point. Its calculation formula is

[0074]

[0075] Where f(z i ) is the decision function, which is used to evaluate whether the wind speed value of the i-th data point is abnormal; sign(·) is the sign function, which returns 1 or -1 according to the positive or negative of the input value; χ is the offset of the decision boundary.

[0076] After removing the abnormal wind speed data points, use their adjacent normal values for replacement and update. The update formula is:

[0077]

[0078] Finally, where v i ′ is the updated and replaced wind speed data at the i-th time point. If v i is abnormal wind speed data and v i-1 ∈ V problem , v i+1 ∈ V problem then the wind speed value v i ′ of the i-th time node after update selects a normal data point farther away; v i-1 , v i , v i+1 are the wind speed data before update at the (i - 1)-th, i-th, and (i + 1)-th time points; V problem is the set of abnormal wind speed data. When the updated wind speed model data set is V t ′ = [v t ′ -5760 , v t ′ -5759 ,..., v t ′ -1 T .

[0079] In an exemplary embodiment, S3 can be replaced by the following steps.

[0080] ​S31: Construct a sample data set; the sample data set includes the preprocessed associated feature data and the corresponding wind power data.

[0081] S32: Divide the sample data set into a training set and a test set.

[0082] S33: Divide the training set into multiple subsets. For each primary learner, sequentially select one subset as the data for the validation subset, and use the remaining subsets as the data for the training subset.

[0083] S34: For each primary learner, record the prediction results of the validation subset and the prediction results of the test set.

[0084] S35: Use the prediction results of the validation subset as training data, use the average value of the prediction results of the test set as new test data, and train a second-layer learner using the training data and the test data to construct a Stacking ensemble learning framework model; the second-layer learner is a meta-learner.

[0085] In practical applications, the Stacking ensemble learning framework, also known as stacking generalization, belongs to a hierarchical model fusion strategy, and its workflow is outlined as follows:

[0086] Define the data set D Stacking , as the input of the Stacking ensemble learning strategy. The time scale of 15 minutes is used as the sample of the collected data points, including 5,760 samples. Each sample consists of the feature vector X δ′ processed by the data preprocessing module and its corresponding wind power. The data set where the execution steps of the data preprocessing module are S2.

[0087] This data preprocessing module can effectively screen out several power features that have a greater impact on wind power, which helps to solve the problem of unequal input feature dimensions in model training and prediction, and capture the causal relationship between power and meteorological features. The prediction accuracy of the model after screening and processing features is significantly higher than that of the model before screening features.

[0088] After the data preprocessing module processes and corrects the retained wind speed features, the prediction is more in line with the actual prediction scenario, achieving good results in the typical wind farms described in the article and having higher applicability.

[0089] Perform segmentation on the data set to obtain training and test data sets.

[0090] Construct Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), Residual Network (Resnet), and Ridge Regression (RR) as the first layer of primary learners, and apply 5-fold cross-validation to the training data. This step includes dividing the training data into 5 equal subsets. For each primary learner, one subset is sequentially selected as the data for the validation subset, and the remaining 4 subsets are used as the data for the training subset. For each primary learner, record its prediction results on the validation subset and the test set, so as to generate a corresponding set of validation subset and test set prediction results for each type of primary learner;

[0091] Build the second-layer learner, and use XGBoost as the meta-learner. This layer of learner uses all the prediction results of the validation subsets of each primary learner in the first layer as the training data, and the average of the prediction results of the test sets is used as the new test data. The meta-learner is trained on this basis to form the ultimate prediction model.

[0092] In an exemplary embodiment, S4 can be replaced by the following steps.

[0093] S41: Randomly initialize any hyperparameter of the Stacking ensemble learning framework model to generate an initial zebra population with multiple zebra individuals, and each zebra individual has d-dimensional position information; d is the maximum position dimension of each zebra individual.

[0094] S42: Based on the position of the current zebra individual in the initial zebra population, set the first maximum number of iterations.

[0095] S43: Based on the first maximum number of iterations, use the elite opposition-based learning method to learn the improved zebra optimization algorithm and introduce a dynamic opposition point.

[0096] S44: Obtain an opposition zebra population with multiple zebra individuals, and each zebra individual has d-dimensional position information according to the dynamic opposition point.

[0097] S45: Compare the fitness values of the initial zebra population and the opposition zebra population to determine the improved zebra population.

[0098] S46: Determine the basic parameters of the improved zebra optimization algorithm according to the improved zebra population.

[0099] In practical applications, use the improved zebra optimization algorithm to train the hyperparameters of each model in the Stacking ensemble learning framework.

[0100] The initialization of the zebra population is divided into three steps:

[0101] The first step: Randomly initialize a zebra population of a zebra individuals, each with d-dimensional position information.

[0102] The second step: Through the zebra optimization algorithm improved by elite opposition-based learning, introduce a dynamic opposition point:

[0103] Among them, the position of the current zebra individual is denoted as Set the maximum number of iterations iter max , then the dynamic opposition point introduced by the elite opposition-based learning method is:

[0104]

[0105] In the formula, ν1 and ν2 are random numbers between (0, 0.5), following a uniform distribution; are the minimum and maximum values of the j-th dimensional position information of the μ-th zebra individual respectively; is the j-th dimensional position information of the μ-th zebra individual after introducing the dynamic opposition point; is the j-th dimensional position information of the μ-th zebra individual before introducing the dynamic opposition point, μ ∈ [1, a], j ∈ [1, d].

[0106] Obtain an opposition zebra population of a zebra individuals, each with d-dimensional position information, through the dynamic opposition point of the elite opposition-based learning method.

[0107] The third step: Compare the fitness values of the position information of each zebra individual in the initial zebra population and its dynamic opposition zebra population, and select a zebra individuals with better fitness from high to low to form the final initialized population. Then, the zebra population W improved by the elite opposition-based learning method is expressed as:

[0108]

[0109] In the formula, d represents the maximum position dimension of each individual zebra, referring to the various parameters of the Stacking model to be optimized, namely the various parameters of XGBoost, LSTM, Resnet, and RR.

[0110] The calculation formula for fitness is:

[0111]

[0112] In the formula, is the fitness of the μ-th zebra; is the number of wind power prediction power samples; is the true power value at time t.

[0113] In an exemplary embodiment, S5 can be replaced by the following steps.

[0114] S51: Initialize the improved zebra population and set the second maximum number of iterations.

[0115] S52: Determine the behaviors of zebra individuals within the improved zebra population according to the second maximum number of iterations; the behaviors of zebra individuals include the foraging behavior of zebras and the defensive behavior of zebras against predators.

[0116] S53: Update the dimensional position information of zebra individuals in real time according to the behaviors of zebra individuals until the second maximum number of iterations is reached, and output the dimensional position information of all dimensions of the globally optimal zebra individual to complete the training of the Stacking ensemble learning framework model.

[0117] In an exemplary embodiment, when the behavior of zebra individuals is the defensive behavior of zebras against predators, different defensive strategies are adopted according to the types of predators; the defensive strategies include a zigzag escape route and an evasion strategy of random lateral transfer, and a counterattack strategy in which all zebra individuals form a defensive formation.

[0118] In practical applications, iterative search of the zebra population is performed to train the Stacking ensemble learning framework model.

[0119] The first step: Initialize the zebra population W and set the maximum number of iterations iter max .

[0120] The second step: If the number of iterations τ ≤ 0.75iter max When, then perform the foraging behavior of zebras, that is, the first stage, and refresh the population composition by imitating the behavior of zebras searching for food. The main foods of zebras include grass and Cyperaceae plants. However, when the preferred food sources are insufficient, they may also switch to eating tender shoots, fruits, bark, roots or leaves. The foraging time of zebras can account for 60 to 80% of their daily time, depending on the quality and availability of plants. Especially for plains zebras, such herbivores tend to eat the taller and less nutritious grasslands first, thus creating living conditions for other species seeking shorter and more nutritious forage. In the improved Zebra Optimization Algorithm (IZOA), the best-performing population members are regarded as leading zebras, and they guide other members towards the target position in the search space. Therefore, the update of the position during the simulation of the zebra foraging process can be achieved through the following mathematical model:

[0121]

[0122] In the formula, is the tentative new position of the μ-th zebra in the j-th dimension at the τ-th iteration; r is a random number between 0 and 1; is the original position of the μ-th zebra in the j-th dimension at the τ-th iteration; is the position of the zebra with the best fitness (i.e., the pioneer zebra) in the j-th dimension at the τ-th iteration; is the new position of the μ-th zebra in the j-th dimension at the (τ + 1)-th iteration. At this time, if the fitness of the new position of the μ-th zebra is better than the fitness of the current position then the zebra will move to the new position; otherwise, it remains at the current position. The root mean square error is used to evaluate the size of the fitness.

[0123] Step 3: When the number of iterations τ > 0.75iter max perform the defensive behavior of zebras against predators, i.e., the second stage. In the second stage, simulate the escape strategy of zebras when being attacked by predators, which is used to update the positions of zebra members in the IZOA algorithm within the search area. Facing the attack of lions, zebras adopt a zigzag escape path and a random lateral transfer escape method. While facing smaller predators such as hyenas and wild dogs, zebras will show more aggressive behaviors, using the collective strength to confuse and intimidate predators. In the setting of the IZOA algorithm, the following two scenario assumptions occur with equal probability:

[0124] Scenario (1): Facing the attack of lions, zebras adopt an escape strategy.

[0125] Scenario (2): Facing the attack of other predators, zebras adopt a counterattack strategy.

[0126] In Scenario (1), the zebras attacked by lions will quickly escape near their current positions. Mathematically, this escape strategy can be expressed by the following formula S1'. In Scenario (2), when facing the attack of other predators, the zebra group will move closer to the attacked zebra and try to make the predators retreat and become confused by forming a defensive formation. This collective defense strategy can be mathematically expressed by the following formula S2'. During the process of updating the zebra positions, if the new position can bring a better value to the objective function, then this new position is adopted. This update mechanism is modeled by the following formula:

[0127]

[0128] In the formula, R, P s are random numbers between 0 and 1; is the position of the attacked zebra in the j-th dimension at the τ-th iteration; is to simulate the increasing fatigue degree of zebras over time.

[0129] Step 4: Update the zebra position according to the second and third steps. If the iteration exceeds the maximum number of iterations, the training ends, and the position information of all dimensions of the globally optimal zebra is output, that is, the parameters of XGBoost, LSTM, Resnet, and RR. Otherwise, jump to the second step to continue parameter optimization.

[0130] In an exemplary embodiment, S6 can be replaced by the following steps.

[0131] Input each associated feature data corresponding to the wind power data to be measured and the wind power data to be measured after being processed by the data preprocessing module into the trained Stacking integrated learning framework model, and the predicted wind power values at each moment in the next 10 days can be obtained.

[0132] For the situation where the influence of different wind power characteristics on wind power varies greatly, use the established model data preprocessing module to analyze and screen each different wind power characteristic, and conduct grey correlation analysis and Pearson correlation coefficient analysis on each wind power characteristic. The grey correlation graph and Pearson radar graph are shown by Figure 2 、 Figure 3 As shown, Table 1 is a schematic table of the associated analysis data values of each wind power characteristic, and the results are shown in Table 1. It can be seen from Table 1 that the wind speed characteristic has the greatest influence on wind power, followed by air pressure, humidity, and temperature.

[0133] Table 1

[0134]

[0135]

[0136] After the wind speed characteristic is retained, abnormal data processing is performed on the wind speed characteristic. In order to verify the influence of the processing of abnormal wind speed characteristic data on the wind power correlation, correlation analysis is performed on different degrees of wind speed characteristic data processing. Table 2 is a schematic table of the associated analysis data values of different wind speed processing degrees. Table 2 includes the data on the influence of randomly processing 10% of the wind speed data, randomly processing 50% of the wind speed data, and processing 100% of the wind speed data on the wind power correlation. It can be seen from Table 2 that when the wind speed characteristic is retained, the correlation coefficients are improved compared with not processing the wind speed for randomly processing 10% of the wind speed data, randomly processing 50% of the wind speed data, and randomly processing 100% of the wind speed data. Among them, the correlation between randomly processing 100% of the wind speed data and wind power is the largest.

[0137] Table 2

[0138]

[0139] To verify that the preprocessing module of the proposed IZOA-Stacking wind power prediction model improves the prediction accuracy and robustness of the prediction model, this application respectively performs power prediction on the IZOA-Stacking prediction models of Model 1 - wind power characteristics without screening, Model 2 - wind power characteristics after screening but without wind speed abnormal data processing, Model 3 - wind power characteristics after screening and wind speed abnormal data processing (randomly processing 10% of wind speed data), Model 4 - wind power characteristics after screening and wind speed abnormal data processing (randomly processing 50% of wind speed data), and Model 5 - wind power characteristics after screening and wind speed abnormal data processing (randomly processing 100% of wind speed data). Table 3 is a schematic table of wind power prediction results before partial wind speed threshold division, and the prediction results are shown in Table 3, Figure 4 and Figure 5 as shown. From Figure 4 and Figure 5 the results in, it can be seen that the IZOA-Stacking wind power prediction model is closer to the actual value compared with other prediction models. Table 4 is a schematic table of evaluation indicators for Model 1 to Model 5. The root mean square error (RMSE), coefficient of determination (Coefficient of Determination, R 2 ), and standard deviation (Standard Deviation, SD) prediction evaluation indicators of the prediction results of each model are shown in Table 4. It can be seen from Table 4 that after screening different characteristics of wind power, the RMSE, R 2 , and SD of the Stacking ensemble learning framework of the improved zebra optimization algorithm (Improved Zebra Optimization Algorithm-Stacked Generalization, IZOA-Stacking) wind power prediction model are reduced by 0.307 MW, increased by 1.27%, and reduced by 0.036 MW respectively; the RMSE gradually decreases by an average of 0.161 MW, and R 2 gradually increases by an average of 0.57%, and the SD gradually increases by an average of 0.019 MW for randomly processing 10% of wind speed data, randomly processing 50% of wind speed data, and randomly processing 100% of wind speed data. The above results show that the preprocessing module of the IZOA-Stacking wind power prediction model does reduce the wind power prediction error and improves a certain prediction accuracy and goodness of fit.

[0140] Table 3

[0141]

[0142] Table 4

[0143] Evaluation Index Model 1 Model 2 Model 3 Model 4 Model 5 RMSE / MW 2.917 2.610 2.345 2.208 2.128 <![CDATA[R 2 / %]]> 93.64 94.91 95.89 96.35 96.62 SD / MW 2.095 2.059 2.076 2.099 2.117

[0144] Based on the preprocessing module of the IZOA-Stacking wind power prediction model, the wind farm data is analyzed and processed. The multi-dimensional data composed of the associated features of the wind power after screening (the processed wind speed features, air pressure, humidity, and temperature) and the wind power is used as the input. The IZOA-Stacking wind power prediction model is used to predict the wind power. To evaluate the effectiveness and accuracy of the proposed wind power prediction method, multiple models such as XGBoost, LSTM, Resnet, RR, Stacking, and IZOA-Stacking are compared. Table 5 is a schematic table of the wind power prediction results before the division of some wind speed thresholds. The prediction results are shown in Table 5, Figure 6 and Figure 7 as shown Figure 7 Figure is the box plot of each prediction model. Among them, Table 5 and Figure 7 Models a-f are XGBoost, LSTM, Resnet, RR, Stacking, and IZOA-Stacking models respectively.

[0145] Table 5

[0146]

[0147] From Figure 6 and Figure 7 the results, it can be seen that the IZOA-Stacking wind power prediction model is closer to the actual value than other prediction models, indicating that the prediction accuracy of this method is higher.

[0148] Table 6 is a schematic table of the evaluation indicators of each model. The RMSE, R 2 and SD prediction evaluation indicators of the prediction results of each model are shown in Table 6. Under the RMSE and SD indicators, the model of this application is reduced by 2.749MW and 2.12MW, 3.58MW and 1.762MW, 1.905MW and 1.859MW, 2.23MW and 1.641MW, 2.211MW and 2.202MW respectively compared with the XGBoost, LSTM, Resnet, RR, Stacking prediction models. Under the R 2 indicator, the model of this application is increased by 14.39%, 20.96%, 8.77%, 10.81%, 10.69% respectively compared with the XGBoost, LSTM, Resnet, RR, Stacking prediction models. This shows that the model of this application has higher prediction accuracy and more stable prediction results.

[0149] Table 6

[0150]

[0151]

[0152] The Stacking ensemble learning framework model based on the improved zebra optimization algorithm constructed in this application has the advantages of high prediction accuracy, high prediction reliability, high efficiency, etc., meeting the actual application requirements.

[0153] Based on the same inventive concept, an embodiment of this application also provides a short-term wind power prediction device based on ensemble learning for implementing the short-term wind power prediction method based on ensemble learning involved above. The implementation solutions provided by this device to solve problems are similar to those recorded in the above method. Therefore, the specific limitations in one or more embodiments of the short-term wind power prediction device based on ensemble learning provided below can refer to the limitations on the short-term wind power prediction method based on ensemble learning in the above text, and will not be repeated here.

[0154] In an exemplary embodiment, a short-term wind power prediction device based on ensemble learning is provided, including:

[0155] A data acquisition module, configured to acquire the wind power data of a wind farm and the corresponding various associated feature data; the associated feature data includes wind speed, temperature, air pressure, humidity, time, and air density.

[0156] A data preprocessing module, configured to preprocess the associated feature data to determine the preprocessed associated feature data.

[0157] A construction module, configured to construct a Stacking ensemble learning framework model according to the preprocessed associated feature data and the corresponding wind power data.

[0158] A basic parameter determination module, configured to initialize the zebra population of each hyperparameter of the Stacking ensemble learning framework model and determine the basic parameters of the improved zebra optimization algorithm; the hyperparameters include the ridge regression regularization hyperparameter, the number of base learners of XGBoost, the LSTM hidden layer size, the learning rate of LSTM, and the learning rate of ResNet; the basic parameters include the maximum number of iterations and the number of zebra individuals;

[0159] A training module, configured to train the Stacking ensemble learning framework model based on the basic parameters by using the improved zebra optimization algorithm.

[0160] A prediction module, configured to predict the wind power in a future period by using the trained Stacking ensemble learning framework model.

[0161] In an exemplary embodiment, a computer device is provided, such as Figure 8As shown, the computer device can be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store short-term wind power prediction data based on integrated learning. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. The computer program, when executed by the processor, implements a short-term wind power prediction method based on integrated learning.

[0162] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the above method is implemented.

[0163] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the above method is implemented.

[0164] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the above method is implemented.

[0165] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random-access memories (ReRAM), magnetoresistive random-access memories (MRAM), ferroelectric random-access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0166] In this application, all actions of obtaining signals, information, or data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the location is located and obtaining authorization from the owner of the corresponding device.

[0167] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0168] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0169] In this application, specific examples are used to elaborate on the principles and implementation manners of the application. The description of the above embodiments is only used to help understand the method and its core idea of the application; at the same time, for those of ordinary skill in the art, according to the idea of the application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation on the application.

Claims

1. A short-term wind power prediction method based on ensemble learning, characterized in that The short-term wind power prediction method based on ensemble learning includes: Obtain the wind power data of the wind farm and the corresponding associated feature data; the associated feature data includes wind speed, temperature, air pressure, humidity, time, and air density; Preprocess the associated feature data to determine the preprocessed associated feature data; Construct a Stacking ensemble learning framework model based on the preprocessed associated feature data and the corresponding wind power data; Initialize the zebra population of each hyperparameter of the Stacking ensemble learning framework model to determine the basic parameters of the improved zebra optimization algorithm; the hyperparameters include the ridge regression regularization hyperparameter, the number of base learners of XGBoost, the LSTM hidden layer size, the learning rate of LSTM, and the learning rate of ResNet; the basic parameters include the maximum number of iterations and the number of zebra individuals; Based on the basic parameters, use the improved zebra optimization algorithm to train the Stacking ensemble learning framework model; Use the trained Stacking ensemble learning framework model to predict the wind power in the future period.

2. The short-term wind power prediction method based on ensemble learning according to claim 1, wherein Preprocess the associated feature data to determine the preprocessed associated feature data, specifically including: Use the wind power data as the reference sequence and the associated feature data as the comparison sequence; Perform dimensionless processing on the reference sequence and the comparison sequence respectively to determine the dimensionless processed wind power data and the dimensionless processed associated feature data; According to the dimensionless processed wind power data and the dimensionless processed associated feature data, perform gray relational degree analysis and Pearson correlation coefficient analysis on each associated feature data, and screen out the associated feature data with an influence degree on the wind power data lower than the set influence degree threshold; Judge whether the wind speed is retained in the screened associated feature data; If so, perform abnormal data processing on the wind speed, and use the associated feature data after abnormal data processing as the preprocessed associated feature data; If not, determine the screened associated feature data as the preprocessed associated feature data.

3. The short-term wind power prediction method based on ensemble learning according to claim 1, wherein Construct a Stacking ensemble learning framework model based on the preprocessed associated feature data and the corresponding wind power data, specifically including: Construct a sample data set; the sample data set includes the preprocessed associated feature data and the corresponding wind power data; Divide the sample data set into a training set and a test set; Divide the training set into multiple subsets. For each primary learner, sequentially select one subset as the data of the validation subset, and use the remaining subsets as the data of the training subset; For each primary learner, record the prediction results of the validation subset and the prediction results of the test set; Use the prediction results of the validation subset as the training data, use the average value of the prediction results of the test set as the new test data, and use the training data and the test data to train the second-layer learner to construct a Stacking ensemble learning framework model; the second-layer learner is the meta-learner.

4. The short-term wind power prediction method based on ensemble learning according to claim 1, characterized in that Initialize the zebra population of each hyperparameter of the Stacking ensemble learning framework model, and determine the basic parameters of the improved zebra optimization algorithm, specifically including: Randomly initialize any hyperparameter of the Stacking ensemble learning framework model to generate an initial zebra population with multiple zebra individuals, and each zebra individual has d-dimensional position information; d is the maximum position dimension of each zebra individual; Based on the position of the current zebra individual in the initial zebra population, set the first maximum number of iterations; Based on the first maximum number of iterations, use the elite opposition-based learning method to learn the improved zebra optimization algorithm and introduce a dynamic opposition point; Obtain an opposition zebra population with multiple zebra individuals, and each zebra individual has d-dimensional position information according to the dynamic opposition point; Compare the fitness values of the initial zebra population and the opposition zebra population to determine the improved zebra population; Determine the basic parameters of the improved zebra optimization algorithm according to the improved zebra population.

5. The short-term wind power prediction method based on ensemble learning according to claim 4, wherein Based on the basic parameters, use the improved zebra optimization algorithm to train the Stacking ensemble learning framework model, specifically including: Initialize the improved zebra population and set the second maximum number of iterations; According to the second maximum number of iterations, determine the zebra individual behaviors within the improved zebra population; the zebra individual behaviors include zebra foraging behavior and the defense behavior of zebras against predators; Update the dimensional position information of zebra individuals in real time according to the zebra individual behaviors until the second maximum number of iterations is reached, and output the all-dimensional position information of the globally optimal zebra individual to complete the training of the Stacking ensemble learning framework model.

6. The short-term wind power prediction method based on ensemble learning according to claim 5, wherein When the zebra individual behavior is the defense behavior of zebras against predators, different defense strategies are adopted according to the type of predators; the defense strategies include the zigzag escape route and the evasion strategy of random lateral transfer, and the counterattack strategy in which all zebra individuals form a defense formation.

7. A short-term wind power prediction device based on ensemble learning, characterized in that, The short-term wind power prediction device based on ensemble learning includes: A data acquisition module for acquiring the wind power data of a wind farm and the corresponding various associated feature data; the associated feature data includes wind speed, temperature, air pressure, humidity, time, and air density; A data preprocessing module for preprocessing the associated feature data to determine the preprocessed associated feature data; A construction module for constructing a Stacking ensemble learning framework model according to the preprocessed associated feature data and the corresponding wind power data; A basic parameter determination module for initializing the zebra population of each hyperparameter of the Stacking ensemble learning framework model and determining the basic parameters of the improved zebra optimization algorithm; the hyperparameters include the ridge regression regularization hyperparameter, the number of base learners of XGBoost, the LSTM hidden layer size, the learning rate of LSTM, and the learning rate of ResNet; the basic parameters include the maximum number of iterations and the number of zebra individuals; A training module, configured to train the Stacking ensemble learning framework model based on the basic parameters by using the improved zebra optimization algorithm; A prediction module, configured to predict the wind power in a future period by using the trained Stacking ensemble learning framework model.

8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the short-term wind power prediction method based on ensemble learning according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the short-term wind power prediction method based on ensemble learning according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the short-term wind power prediction method based on ensemble learning according to any one of claims 1-6.

Images