A short-term wind power prediction method and system based on dynamic weighted combination
By using a dynamic weighted combination method and particle swarm optimization algorithm to optimize the weights of the wind power sub-model, and by integrating multiple prediction models, the problem of unstable prediction performance under different meteorological conditions is solved, and higher accuracy short-term wind power prediction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DEZHOU POWER SUPPLY COMPANY OF STATE GRID SHANDONG ELECTRIC POWER
- Filing Date
- 2022-12-12
- Publication Date
- 2026-05-29
AI Technical Summary
Existing short-term wind power forecasting methods fail to effectively consider the performance variations of various forecasting sub-models under different meteorological conditions, resulting in poor forecasting performance.
A dynamic weighted combination method is adopted, which optimizes the weights of the wind power sub-prediction model through particle swarm optimization algorithm, integrates the prediction results of gated cyclic unit, lightweight gradient booster and least squares support vector machine models, and generates the optimal dynamic weighted combination by taking into account meteorological similarity and historical data characteristics.
It improves the accuracy and precision of short-term wind power forecasting, reduces forecasting errors, and is particularly flexible and accurate under different meteorological conditions.
Smart Images
Figure CN116090608B_ABST
Abstract
Description
Technical Field
[0001] This disclosure belongs to the field of wind power prediction technology, specifically relating to a short-term wind power prediction method and system based on dynamic weighted combination. Background Technology
[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.
[0003] Wind energy has been vigorously developed due to its green, environmentally friendly, and sustainable characteristics, and wind power generation plays a vital role in its utilization. The unpredictability of wind speed is the most significant factor affecting wind power generation, resulting in dramatic fluctuations in power output, sometimes rising rapidly and sometimes falling sharply. While wind power generation benefits society, large-scale grid connection presents significant challenges. For example, if wind power generation cannot be accurately predicted, dispatch operators may make unreasonable power balance decisions. Therefore, wind power forecasting has become an indispensable component of centralized wind farms.
[0004] Short-term wind power forecasting refers to predicting wind power generation from 00:00 the following day to the next 72 hours. Short-term wind power forecasting typically employs statistical methods, namely the ARMIA model, vector autoregressive exogenous model, and Gaussian process. With the rapid development of machine learning and deep learning, artificial intelligence methods are gradually showing significant advantages in short-term wind power forecasting.
[0005] Most current forecasting methods design only one forecasting model for short-term wind power forecasting. However, each forecasting model has its own advantages and disadvantages, and no single model can perform optimally in every situation. Therefore, combined methods have been proposed and widely used in short-term wind power forecasting. However, while existing combined methods can improve forecasting performance, they rarely consider the time-varying characteristics of the weights of the sub-models. Summary of the Invention
[0006] To address the aforementioned issues, this disclosure proposes a short-term wind power prediction method and system based on dynamic weighted combination. This method considers the performance changes of each prediction sub-model under different meteorological conditions and achieves short-term wind power prediction through dynamic weighted combination, thereby improving the prediction performance of short-term wind power.
[0007] According to some embodiments, the first solution of this disclosure provides a short-term wind power prediction method based on dynamic weighted combination, which adopts the following technical solution:
[0008] A short-term wind power forecasting method based on dynamic weighted combination includes:
[0009] Obtain meteorological data from wind farms;
[0010] Based on the acquired meteorological data and the preset wind power sub-prediction model, wind power is predicted, and the prediction results of the wind power sub-prediction model are obtained.
[0011] Based on the particle swarm optimization algorithm, the weights of the wind power sub-prediction model are optimized to obtain a dynamic weighted combination;
[0012] Based on the obtained dynamic weighted combination, the prediction results of the obtained wind power sub-prediction model are integrated to obtain the short-term wind power prediction results;
[0013] The preset wind power prediction sub-model includes a gated cyclic unit sub-prediction model, a lightweight gradient booster sub-prediction model, and a least squares support vector machine prediction model.
[0014] As a further technical limitation, the acquired wind farm meteorological data includes wind speed, wind direction, temperature, pressure, and relative humidity.
[0015] As a further technical limitation, based on the acquired meteorological data from the wind farm, meteorological correlation analysis was performed using the Pearson correlation coefficient, i.e., the Pearson correlation coefficient. for
[0016] ;in, Represents meteorological variables; Indicates wind force; Represents the expected value of a meteorological variable; This represents the expected value of wind power generation. The sample size is indicated by the Pearson correlation coefficient. The correlation between different meteorological factors and wind power is calculated, and the determined wind speed, wind direction, temperature and relative humidity are used as input data for the wind power sub-prediction model.
[0017] As a further technical limitation, the gated recurrent unit sub-prediction model improves the network structure by setting update and reset gates, thereby simplifying the structure and reducing the training time. The lightweight gradient boosting machine sub-prediction model includes a tree algorithm based on decision histograms, a gradient-based one-sided sampling strategy, exclusive feature bundling operations, and a leaf growth strategy with depth limitations, improving prediction accuracy and computational efficiency. The least squares support vector machine sub-prediction model transforms the quadratic programming problem into a solution to a system of linear equations, improving the model's solution efficiency.
[0018] As a further technical limitation, the feature is that, in the process of optimizing the weights of the wind power sub-prediction model, a particle swarm optimization algorithm is used to construct a fitness function based on the normalized root mean square error, update the weights and iterative adjustment values of each wind power sub-prediction model, and obtain the optimal dynamic weighted combination.
[0019] Furthermore, the fitness function for
[0020] ;in, This indicates the total number of historical days; Indicates a historical date; Indicates the index for the target forecast day; Indicates actual wind energy; Represents the weights of the gated recurrent unit sub-prediction model; This represents the prediction result of the gated recurrent unit sub-prediction model; This represents the weights of the lightweight gradient booster predictive model; This represents the prediction results of the lightweight gradient booster prediction model; Represents the weights of the least squares support vector machine sub-prediction model; This represents the prediction result of the least squares support vector machine sub-prediction model.
[0021] As a further technical constraint, considering meteorological similarity, the weights of the gated cyclic unit sub-prediction model, the lightweight gradient booster sub-prediction model, and the least squares support vector machine sub-prediction model are adjusted; based on the obtained optimal dynamic weighted combination, the prediction results of the obtained wind power sub-prediction model are fused to obtain the short-term wind power prediction results.
[0022] According to some embodiments, the second aspect of this disclosure provides a short-term wind power prediction system based on dynamic weighted combination, employing the following technical solution:
[0023] A short-term wind power forecasting system based on dynamic weighted combination includes:
[0024] The acquisition module is configured to acquire meteorological data from the wind farm.
[0025] The preliminary prediction module is configured to predict wind power based on the acquired meteorological data and the preset wind power sub-prediction model, and obtain the prediction results of the wind power sub-prediction model.
[0026] The optimization module is configured to optimize the weights of the wind power sub-prediction model based on the particle swarm optimization algorithm to obtain a dynamic weighted combination;
[0027] The prediction module is configured to combine the prediction results of the obtained wind power sub-prediction model with the obtained dynamic weighted combination to obtain the short-term wind power prediction result.
[0028] The preset wind power prediction sub-model includes a gated cyclic unit sub-prediction model, a lightweight gradient booster sub-prediction model, and a least squares support vector machine prediction model.
[0029] According to some embodiments, a third aspect of this disclosure provides a computer-readable storage medium, employing the following technical solution:
[0030] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the short-term wind power prediction method based on dynamic weighted combination as described in the first aspect of this disclosure.
[0031] According to some embodiments, the fourth solution of this disclosure provides an electronic device that adopts the following technical solution:
[0032] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the short-term wind power prediction method based on dynamic weighted combination as described in the first aspect of this disclosure.
[0033] Compared with the prior art, the beneficial effects of this disclosure are as follows:
[0034] This disclosure integrates various meteorological data and searches for similar days, fully analyzes the characteristics of historical power data, and integrates richer information. Considering the performance changes of sub-models under different meteorological conditions, it proposes a dynamic weighted combination prediction method that considers meteorological similarity to improve short-term wind power prediction. By adopting a combination method and using particle swarm optimization algorithm to optimize the weights of sub-models, the final prediction results are generated, which helps to improve prediction performance. Attached Figure Description
[0035] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.
[0036] Figure 1 This is a flowchart of the short-term wind power prediction method based on dynamic weighted combination in Embodiment 1 of this disclosure;
[0037] Figure 2(a) is a schematic diagram comparing the prediction performance of different models in a strong wind scenario in Embodiment 1 of this disclosure;
[0038] Figure 2(b) is a schematic diagram comparing the prediction performance of different models in a light wind scenario in Embodiment 1 of this disclosure;
[0039] Figure 3 This is a structural diagram of the gated loop unit in Embodiment 1 of this disclosure;
[0040] Figure 4 This is a diagram illustrating the training process of the lightweight gradient booster in Embodiment 1 of this disclosure;
[0041] Figure 5 This is a flowchart of the sub-model weight optimization process in Embodiment 1 of this disclosure;
[0042] Figure 6 This is a flowchart of the dynamic weighted combination prediction model in Embodiment 1 of this disclosure;
[0043] Figure 7 This is a histogram of normalized root mean square error (NRMSE) under different similar number of days in Embodiment 1 of this disclosure;
[0044] Figure 8 This is a comparison chart of the prediction error results of different models in Embodiment 1 of this disclosure;
[0045] Figure 9 This is a structural block diagram of the short-term wind power prediction system based on dynamic weighted combination in Embodiment 2 of this disclosure. Detailed Implementation
[0046] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.
[0047] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.
[0048] Where there is no conflict, the embodiments and features described herein can be combined with each other.
[0049] Example 1
[0050] Embodiment 1 of this disclosure introduces a short-term wind power prediction method based on dynamic weighted combination.
[0051] like Figure 1 The method shown is a short-term wind power prediction method based on dynamic weighted combination, including:
[0052] Obtain meteorological data from wind farms;
[0053] Based on the acquired meteorological data and the preset wind power sub-prediction model, wind power is predicted, and the prediction results of the wind power sub-prediction model are obtained.
[0054] Based on the particle swarm optimization algorithm, the weights of the wind power sub-prediction model are optimized to obtain a dynamic weighted combination;
[0055] Based on the obtained dynamic weighted combination, the prediction results of the obtained wind power sub-prediction model are integrated to obtain the short-term wind power prediction results;
[0056] The preset wind power prediction sub-model includes a gated cyclic unit sub-prediction model, a lightweight gradient booster sub-prediction model, and a least squares support vector machine prediction model.
[0057] This embodiment considers the performance variations of sub-models under different meteorological conditions and improves wind power forecasting through a dynamic weighted combination forecasting method. To improve forecasting accuracy, a dynamic combination forecasting model for short-term wind power forecasting is constructed. To overcome the shortcomings of traditional combination methods, this embodiment considers the impact of meteorological conditions on sub-model performance and adjusts the weights of sub-models based on the prediction results of historical similar days using the particle swarm optimization algorithm. In this way, the proposed combination method is more flexible and improves accuracy compared to the fixed weighted combination method.
[0058] Specifically, the detailed implementation process of this embodiment is as follows:
[0059] The relevant influencing variables for short-term wind power prediction in the sub-model were identified, and meteorological correlation analysis was performed using the Pearson correlation coefficient (PCC). Pearson correlation coefficient for:
[0060] 1)
[0061] in, Represents meteorological variables; Indicates wind force; Represents the expected value of a meteorological variable; This represents the expected value of wind power generation. Indicates the number of samples; , The larger the absolute value, the stronger the correlation.
[0062] This embodiment considers wind speed, wind direction, temperature, pressure, and relative humidity, and their Pearson correlation coefficients with wind energy are shown in Table 1. Except for pressure, the other meteorological variables are strongly correlated with wind power. Therefore, wind speed, wind direction, temperature, and relative humidity were determined as the input features of the sub-model.
[0063] Table 1. Pearson correlation coefficients between meteorological variables and wind force
[0064] meteorological variables Pearson correlation coefficient wind speed 0.78 wind direction 0.23 temperature 0.28 pressure ﹣0.053 relative humidity 0.36
[0065] No single prediction model can outperform another in all scenarios. Regarding actual wind energy, Figure 2(a) shows strong winds, and Figure 2(b) represents light winds. The performance of the three prediction models is drastically different. Under strong wind conditions, Model 2 performs best, with its predicted wind power curve closest to the actual wind power curve. Under light wind conditions, Model 3 shows the smallest prediction error. Although only two test days are shown here, it still illustrates that the performance of the sub-models changes accordingly when meteorological conditions change. Therefore, fixed weights for the sub-models are inappropriate and limit the combined performance. Dynamic weights that reflect changes in meteorological conditions can further improve the prediction accuracy of the combined model. This is the motivation behind the proposed dynamic combination method for short-term wind power forecasting.
[0066] Before building the combined model, the first step is to select sub-models. First, the prediction accuracy of the sub-models should not differ significantly. Second, the prediction results of the sub-models should cover real wind power as comprehensively as possible. Combining models in this way will be more effective. Therefore, the sub-models need to be based on different prediction principles. Commonly used learning models include tree models, neural network models, and support vector machines, whose prediction principles are quite different. This embodiment ultimately determined to use gated recurrent units (GRU), lightweight gradient boosters (light-GBM), and least squares support vector machines (LSSVM) to construct the combined model. Detailed explanations of the principles are as follows.
[0067] 1) Gated Cyclic Unit (GRU)
[0068] Traditional artificial neural networks excel at mapping the nonlinear relationship between input features and network output. However, time dependencies have proven to be a crucial component of short-term wind power forecasting, and artificial neural networks lack the ability to learn these temporal relationships. Recurrent neural networks (RNNs) were developed in this context. A drawback of RNNs is that the model may fail as the time series lengthens, a phenomenon known as short-term dependency. Long Short-Term Memory (LSTM) networks and gated recurrent units (GRUs) overcome this limitation. Furthermore, GRUs offer similar predictive performance to LSTM networks with higher computational efficiency.
[0069] This embodiment selected a gated recurrent unit (GRU) from among numerous artificial intelligence methods. Compared to Long Short-Term Memory (LSTM) networks, GRUs improve the network structure. LSM networks have three gates: an input gate, a forget gate, and an output gate. GRUs have only two gates: an update gate and a reset gate. This simplified structure helps to reduce the time consumption of the training process. The unit structure is as follows: Figure 3 As shown. Update the last hidden state of the door control. Update to current hidden state To what extent, the gate control resets the information currently input that has been remembered.
[0070] The working process of the gated loop unit is as follows:
[0071] (2)
[0072] (3)
[0073] (4)
[0074] (5)
[0075] In the formula: The input features represent the current time t. Represents the hidden state at the current time t. and These represent updating the door and resetting the door, respectively. and It is a sigmoid function and a hyperbolic tangent function. , ,and It is a weight matrix. , ,and It is the corresponding bias vector.
[0076] 2) Lightweight Gradient Bumper Machine (light-GBM)
[0077] Lightweight Gradient Boosting Machine (LGBGM) is an ensemble learning method based on Gradient Boosting Decision Tree (GBDT) models. The drawback of GBDT is that traversing the training dataset requires significant computational resources and may encounter cache misses. To ensure prediction accuracy and accelerate model training, this embodiment introduces a lightweight gradient boosting machine through several improvements, including a tree algorithm based on decision histograms, a gradient-based one-sided sampling strategy, exclusive feature bundling operations, and a leaf growth strategy with depth limitations. This significantly improves prediction accuracy and computational efficiency, enabling the lightweight gradient boosting machine to handle the massive amounts of data required for short-term wind power forecasting. The training process of the lightweight gradient boosting machine is as follows: Figure 4 As shown. TD represents the training dataset, and TDn represents a subset of TD. Weak learners are trained based on the training dataset and the initialized weights. The weights are adjusted based on the prediction error of the weak learners. In this training cycle, there are a total of n trained weak learners. The final prediction result is composed of the weak learners, as shown in (6):
[0078] (6)
[0079] In the formula, X represents the model input features, and y represents the predicted wind energy. This represents the output of the weak learner in the nth iteration.
[0080] 3) Least Squares Support Vector Machine (LSSVM)
[0081] Least squares support vector machine (LSVM) is an improvement on the traditional support vector machine model. It transforms the quadratic programming problem into a solution to a system of linear equations, significantly improving the efficiency of the model's solution. Given a training dataset... The least squares support vector machine model can be represented by (7):
[0082] (7)
[0083] in, 'b' refers to the weight vector, and 'b' is the bias vector. The model solution process can be described as an optimization problem:
[0084] (8)
[0085] As can be seen, the above optimization problem has equality constraints, and the Lagrange multiplier method is used here. Finally, a system of linear equations can be obtained:
[0086] (9)
[0087] in, Let represent the kernel function, be a symmetric matrix satisfying the Mercer condition, and α be the Lagrange multipliers. The linear regression expression for the least squares support vector machine can be expressed as:
[0088] (10)
[0089] This embodiment considers dynamic weighting of meteorological similarity. First, a search for meteorologically similar days is performed. Wind speed, wind direction, temperature, and relative humidity are identified as features of the short-term wind power forecasting model and are also considered as features of historically similar days for the target forecast date. A Gaussian kernel function is used to measure the similarity between two days for certain meteorological aspects. For multiple meteorological similarities, a weighted average is calculated using the Pearson correlation coefficient. The Gaussian kernel function is defined as follows:
[0090] (11)
[0091] in, Meteorological variables representing the target forecast date, The weather variables represent historical days. The final similarity between the target day and historical days can be calculated using (12):
[0092] (12)
[0093] In the formula, Let j be the index for the target forecast day, and j be the historical daily index. The Pearson correlation coefficient between wind speed and wind power. The Pearson correlation coefficient for wind direction and wind power. It is the Pearson correlation coefficient between temperature and wind power. It is the Pearson correlation coefficient between relative humidity and wind power. , It is the variance that controls the shape of the Gaussian kernel.
[0094] Suppose we need to find M similar days in N days of historical data. Based on numerical weather prediction (NWP) results, we calculate and sort the similarity between the target forecast day and the N historical days, and use the number of days corresponding to the top M similarity scores to determine the weights of the target forecast day model. The dataset is described as follows: ,in This represents the prediction result of the gated loop unit. This represents the prediction results of a lightweight gradient booster. This represents the prediction result of the least squares support vector machine. This refers to actual wind energy.
[0095] When considering the dynamic weights of the sub-models, based on the dataset A particle swarm optimization (PSO) algorithm is constructed to optimize the weights of the target prediction day model. PSO is an evolutionary computational technique derived from the study of bird flocks' foraging behavior. Its core idea is to utilize information sharing among individuals within a group to induce an evolutionary process from disorder to order in the problem space, thereby obtaining a feasible solution. One advantage of PSO is its simplicity of implementation, requiring minimal parameter tuning. Each particle has two main attributes: position and velocity. The position of all particles is evaluated using a fitness function constructed from the normalized root mean square error (NRMSE) of the combined prediction results. The velocity and position of each particle are updated as follows:
[0096] (13)
[0097] (14)
[0098] in, and They represent the first The velocity of a particle in the d-th iteration It is inertial weight. and These are individual and social learning rates, and Represents a random number between 0 and 1. Representing the The optimal position of each particle in the d-th iteration. It is the globally optimal position at the d-th iteration.
[0099] The entire work process is as follows:
[0100] In the weight optimization application of particle swarm optimization, the particle's position refers to the weight of the sub-model, the velocity refers to the adjustment value in each iteration, and the fitness function is calculated using the following formula:
[0101] (15)
[0102] in, Represents fitness. , , and These are the weights of the selected sub-model gated recurrent unit, lightweight gradient booster, and least squares support vector machine, respectively. The workflow for optimizing sub-model weights is as follows: Figure 5 As shown.
[0103] Figure 6 This document demonstrates the entire working process of the model proposed in this embodiment. The entire dataset can be divided into three parts: a training dataset, a validation dataset, and a test dataset. During the training phase, the three sub-models—gated recurrent unit, lightweight gradient booster, and least squares support vector machine—are trained using the training dataset, which includes numerical weather forecast data and observed wind power data. The numerical weather forecast data consists of wind speed, wind direction, temperature, and relative humidity. During the validation phase, the numerical weather forecast data is input into the trained sub-models to generate prediction results. The numerical weather forecast data, observed wind power data, and predicted wind power data for the three prediction models constitute the entire validation dataset. During the testing phase, the numerical weather forecast data is first input into the sub-models to generate three prediction results. Then, the Gaussian similarity index is used to find N similar days based on meteorological characteristics. Next, using the three predicted wind power values and the observed wind power values, the weights of the three sub-models are optimized using a particle swarm optimization algorithm. Finally, the optimized weights are used to obtain the combined prediction result. In this process, the weights of the sub-models are adjusted considering meteorological similarity, as different models perform differently under different meteorological conditions.
[0104] This embodiment uses power generation data from a 100 MW wind farm as an example to verify the effectiveness of the method. The 15-minute resolution wind power data comes from the SCADA system, while the 3km and 15-minute resolution numerical weather prediction data comes from the WRF model. The entire dataset spans from January 2017 to November 2018, and the data missing rate is within an acceptable range. The detailed breakdown of the dataset is shown in Table 2.
[0105] The training dataset is used to train the sub-model, the validation dataset is used to find historical similar days and generate dynamic weights, and the test dataset is used to test the predictive performance of the proposed dynamic combined prediction model.
[0106] Table 2 Dataset Partitioning
[0107] Dataset time Training dataset January 2017 - August 2017 Validation dataset September 2017 - April 2018 Test dataset May 2018 - December 2018
[0108] To verify the effectiveness of the proposed method, this embodiment compares the proposed dynamic combinatorial prediction model with a selected benchmark. First, the three sub-models are compared to demonstrate the effectiveness of the combinatorial method. Then, the traditional fixed-weight combinatorial prediction model using the particle swarm optimization (PSO) algorithm is compared with the proposed dynamic combinatorial prediction method; the former is labeled PSO-FCF, and the latter is labeled PSO-DCF.
[0109] The number of similar days is crucial for constructing a subset of data to generate weights for combining the predictions of the target forecast day model. If the number is too small, the generated weights may be unrepresentative and unstable. Conversely, if the number is too large, dissimilar meteorological conditions may be found, and the resulting weights may not reflect the sub-model's performance on the meteorological conditions of the target forecast day. Therefore, a balance in the number of similar days should be found, and in this embodiment, pre-testing was conducted to find the optimal setting.
[0110] like Figure 7 As shown, the normalized root mean square error (NRMSE) first decreases and then increases with the increase in the number of similar days, reaching its minimum at a number of 5. Therefore, in this embodiment, the number of similar days is set to 5.
[0111] Normalized root mean square error (NRMSE) and normalized mean absolute error (NMAE) were used to evaluate the accuracy of the benchmark and the proposed model. Table 3 shows the performance on the test dataset. For NRMSE, compared with gated recurrent unit (GRU), lightweight gradient booster (light-GBM), least squares support vector machine (LSSVM), and particle swarm optimization algorithm-based conventional fixed-weighted combination prediction model (PSO-FCF), the prediction error of the proposed particle swarm optimization algorithm-based dynamic combination prediction model (PSO-DCF) in this embodiment is reduced by 8.19%, 9.04%, 6.68%, and 2.42%, respectively. For NMAE, the proposed PSO-DCF has the lowest NMAE of 11.34%. Figure 8 A prediction case is provided, demonstrating that the proposed PSO-DCF has a predicted wind energy curve that most closely resembles the actual wind energy curve. Overall, the PSO-DCF proposed in this embodiment exhibits the best performance among all selected benchmarks.
[0112] Table 3 Comparison of prediction errors for each model
[0113] Model NRMSE (%) NMAE (%) Gated Cyclic Unit (GRU) 14.89 12.43 Lightweight Gradient Booster (Light-GBM) 15.03 12.51 Least Squares Support Vector Machine (LSSVM) 14.65 12.17 Particle Swarm Optimization Algorithm: Traditional Fixed Weight Combination Prediction Model (PSO-FCF) 14.01 11.95 Dynamic Combination Prediction Model of Particle Swarm Optimization Algorithm (PSO-DCF) 13.67 11.34
[0114] This embodiment integrates various meteorological data and searches for similar days, fully analyzes the characteristics of historical power data, and fuses richer information. Considering the performance changes of sub-models under different meteorological conditions, it proposes a dynamic weighted combination prediction method that considers meteorological similarity to improve short-term wind power prediction. By adopting a combination method and using particle swarm optimization algorithm to optimize the weights of sub-models, the final prediction results are generated, which helps to improve prediction performance.
[0115] Example 2
[0116] Embodiment 2 of this disclosure introduces a short-term wind power prediction system based on dynamic weighted combination.
[0117] like Figure 9 The short-term wind power forecasting system shown includes:
[0118] The acquisition module is configured to acquire meteorological data from the wind farm.
[0119] The preliminary prediction module is configured to predict wind power based on the acquired meteorological data and the preset wind power sub-prediction model, and obtain the prediction results of the wind power sub-prediction model.
[0120] The optimization module is configured to optimize the weights of the wind power sub-prediction model based on the particle swarm optimization algorithm to obtain a dynamic weighted combination;
[0121] The prediction module is configured to fuse the prediction results of the obtained wind power sub-prediction model with the obtained dynamic weighted combination to obtain the short-term wind power prediction result.
[0122] The preset wind power prediction sub-model includes a gated cyclic unit sub-prediction model, a lightweight gradient booster sub-prediction model, and a least squares support vector machine prediction model.
[0123] The detailed steps are the same as those of the short-term wind power prediction method based on dynamic weighted combination provided in Example 1, and will not be repeated here.
[0124] Example 3
[0125] Embodiment 3 of this disclosure provides a computer-readable storage medium.
[0126] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps in the short-term wind power prediction method based on dynamic weighted combination as described in Embodiment 1 of this disclosure.
[0127] The detailed steps are the same as those of the short-term wind power prediction method based on dynamic weighted combination provided in Example 1, and will not be repeated here.
[0128] Example 4
[0129] Embodiment 4 of this disclosure provides an electronic device.
[0130] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the short-term wind power prediction method based on dynamic weighted combination as described in Embodiment 1 of this disclosure.
[0131] The detailed steps are the same as those of the short-term wind power prediction method based on dynamic weighted combination provided in Example 1, and will not be repeated here.
[0132] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A short-term wind power prediction method based on dynamic weighted combination, characterized in that, include: Obtain meteorological data from wind farms; Based on the acquired meteorological data and the preset wind power sub-prediction model, wind power is predicted, and the prediction results of the wind power sub-prediction model are obtained. Based on the particle swarm optimization algorithm, the weights of the wind power sub-prediction model are optimized to obtain a dynamic weighted combination; Based on the obtained dynamic weighted combination, the prediction results of the obtained wind power sub-prediction model are integrated to obtain the short-term wind power prediction results; The preset wind power prediction sub-model includes a gated cyclic unit sub-prediction model, a lightweight gradient booster sub-prediction model, and a least squares support vector machine prediction model.
2. The short-term wind power prediction method based on dynamic weighted combination as described in claim 1, characterized in that, The acquired meteorological data for wind farms includes wind speed, wind direction, temperature, pressure, and relative humidity.
3. The short-term wind power prediction method based on dynamic weighted combination as described in claim 1, characterized in that, Based on the acquired meteorological data from the wind farm, meteorological correlation analysis was conducted using the Pearson correlation coefficient, i.e., the Pearson correlation coefficient. for ;in, Represents meteorological variables; Indicates wind force; Represents the expected value of a meteorological variable; This represents the expected value of wind power generation. The sample size is indicated by the Pearson correlation coefficient. The correlation between different meteorological factors and wind power is calculated, and the determined wind speed, wind direction, temperature and relative humidity are used as input data for the wind power sub-prediction model.
4. The short-term wind power prediction method based on dynamic weighted combination as described in claim 1, characterized in that, The gated recurrent unit sub-prediction model improves the network structure by setting update and reset gates, simplifying the structure to reduce training time. The lightweight gradient boosting sub-prediction model includes a tree algorithm based on decision histograms, a gradient-based one-sided sampling strategy, exclusive feature bundling operations, and a leaf growth strategy with depth limitations, improving prediction accuracy and computational efficiency. The least squares support vector machine sub-prediction model transforms the quadratic programming problem into a solution to a system of linear equations, improving model solution efficiency.
5. The short-term wind power prediction method based on dynamic weighted combination as described in claim 1, characterized in that, In optimizing the weights of the wind power sub-prediction model, a particle swarm optimization algorithm is used to construct a fitness function based on the normalized root mean square error, update the weights and iterative adjustment values of each wind power sub-prediction model, and obtain the optimal dynamic weighted combination.
6. The short-term wind power prediction method based on dynamic weighted combination as described in claim 5, characterized in that, The fitness function for ;in, This indicates the total number of historical days; Indicates a historical date; Indicates the index for the target forecast day; Indicates actual wind energy; Represents the weights of the gated recurrent unit sub-prediction model; This represents the prediction result of the gated recurrent unit sub-prediction model; This represents the weights of the lightweight gradient booster prediction model; This represents the prediction results of the lightweight gradient booster prediction model; Represents the weights of the least squares support vector machine sub-prediction model; This represents the prediction result of the least squares support vector machine sub-prediction model.
7. The short-term wind power prediction method based on dynamic weighted combination as described in claim 1, characterized in that, Considering meteorological similarity, the weights of the gated cyclic unit sub-prediction model, the lightweight gradient booster sub-prediction model, and the least squares support vector machine sub-prediction model are adjusted; based on the obtained optimal dynamic weighted combination, the prediction results of the obtained wind power sub-prediction model are fused to obtain the short-term wind power prediction results.
8. A short-term wind power forecasting system based on dynamic weighted combination, characterized in that, include: The acquisition module is configured to acquire meteorological data from the wind farm. The preliminary prediction module is configured to predict wind power based on the acquired meteorological data and the preset wind power sub-prediction model, and obtain the prediction results of the wind power sub-prediction model. The optimization module is configured to optimize the weights of the wind power sub-prediction model based on the particle swarm optimization algorithm to obtain a dynamic weighted combination; The prediction module is configured to fuse the prediction results of the obtained wind power sub-prediction model with the obtained dynamic weighted combination to obtain the short-term wind power prediction result. The preset wind power prediction sub-model includes a gated cyclic unit sub-prediction model, a lightweight gradient booster sub-prediction model, and a least squares support vector machine prediction model.
9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the short-term wind power prediction method based on dynamic weighted combination as described in any one of claims 1-7.
10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the short-term wind power prediction method based on dynamic weighted combination as described in any one of claims 1-7.