Wind power prediction method and system based on physical feature enhanced SBLS

Abstract

This invention relates to the field of wind power prediction technology, and discloses a wind power prediction method and system based on Physical Feature Enhancement (SBLS). The method includes: acquiring operational data from the wind turbine SCADA system; treating the constructed wind speed-power dynamic characteristic curve and decision interval as power dynamic curve knowledge; mapping this knowledge to physical nodes alongside feature nodes and enhancement nodes; constructing an SBLS network topology containing physical nodes; training the SBLS prediction model containing physical nodes using a training set; merging the outputs of feature nodes, enhancement nodes, and physical nodes to construct an augmented matrix; analytically solving the output weights using a ridge regression pseudo-inverse algorithm; and applying the trained SBLS prediction model containing physical nodes to a test set for wind power prediction. This invention effectively improves wind power prediction accuracy by explicitly embedding physical knowledge through changing the network topology, and its training speed is superior to that of deep neural networks.

Description

Technical Field

[0001] This invention belongs to the field of wind power prediction technology, and specifically relates to a wind power prediction method and system based on Physical Feature Enhancement (SBLS). Background Technology

[0002] With the continuous growth of global demand for clean energy, the development of renewable energy is showing a rapid trend. In the context of digital operation of wind farms, massive amounts of Supervisory Control and Data Acquisition (SCADA) data provide the foundation for power prediction. As the scale of wind farms expands, higher demands are placed on the training speed and deployment efficiency of prediction models. Traditional prediction methods often face problems such as high computational resource consumption and long model update cycles, making it difficult to meet the actual needs of wind farms for rapid response and online learning. With the continuous development of machine learning technology, more and more nonlinear models are being introduced into the field of time series prediction. In wind power prediction tasks, many traditional shallow models such as random forests, linear regression, and gradient boosting decision trees are still widely used. In recent years, however, deep learning technology, especially long short-term memory networks and gated recurrent units, has become an indispensable tool in wind power prediction due to its excellent nonlinear modeling capabilities. Although machine learning has achieved significant predictive results in wind power prediction tasks, its models usually contain a large number of parameters, which not only makes the training process more complex but also greatly extends training time and computational costs. To alleviate this problem, Stacked Broad Learning System (SBLS), as an emerging learning architecture, has been used in wind power prediction tasks. Unlike traditional deep learning methods, SBLS constructs a wide network structure, rather than a deep hierarchical structure, which can effectively reduce the training time and complexity of the model while maintaining prediction accuracy. However, in practical applications, existing SBLS prediction models often rely on data-driven feature mapping and augmentation mapping, neglecting the dynamic relationship between wind power and wind speed. Under complex operating conditions, this may lead to prediction results inconsistent with the dynamic physical operation of the wind turbine. There is still room for improvement in prediction accuracy and physical interpretability.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address or at least alleviate one or more of the above problems, a wind power prediction method and system based on physical feature-enhanced SBLS is provided.

[0005] To achieve the above objectives, according to a first aspect of the present invention, a wind power prediction method based on physically enhanced SBLS is provided, comprising the following steps: Acquire operational data from the wind turbine data acquisition and monitoring control SCADA system, including wind speed and power; divide the preprocessed operational data into training and testing sets; Based on the operational data in the training set, a wind speed-power scatter plot was established. The wind speed interval was discretized using the binning method, that is, the wind speed was divided into sections with an interval of 1 m / s, and the centroid of each interval was calculated. The centroid was used as the characteristic trend point. The trend point was fitted using the B-spline interpolation method to construct the wind speed-power dynamic characteristic curve. Based on the aforementioned wind speed-power dynamic characteristic curve, a wind speed-power decision interval is constructed. The constructed wind speed-power dynamic characteristic curve and decision interval are regarded as power dynamic curve knowledge, and mapped to physical nodes in parallel with feature nodes and enhancement nodes, thus obtaining the SBLS prediction model containing physical nodes, that is, the SBLS prediction model with physical feature enhancement. The embedded SBLS prediction model with physical nodes is trained using the training set. The outputs of feature nodes, augmentation nodes, and physical nodes are merged to construct an augmented matrix. The output weights are then analytically solved using the ridge regression pseudo-inverse algorithm. The trained SBLS prediction model with physical nodes was applied to the test set for wind power prediction.

[0006] Furthermore, the preprocessing includes using the sliding window quartile method to detect outliers for wind speed and power data; and using ridge regression extrapolation to correct missing and outlier values ​​to ensure the continuity of the data and the integrity of the time series.

[0007] Furthermore, the step of using B-spline interpolation to fit trend points and construct the wind speed-power dynamic characteristic curve includes: Feature trend points identified on the training set The wind speed-power dynamic characteristic curve is fitted using the B-spline interpolation method; the B-spline interpolation method is defined as follows: Given B-spline basis functions: Basis functions of order 0: ; Basis functions of order 1: ; Basis functions of order 2: ; Basis functions of order 3: ; Wind speed-power dynamic characteristic curve fitted by B-spline interpolation: ; in, It is the first Wind speed values ​​at each characteristic trend point , , , They represent the first , , , Wind speed values ​​at each characteristic trend point , , These represent the wind speed values ​​at the 1st, 2nd, and nth characteristic trend points, respectively. It is the power value of the i-th characteristic trend point. , , These represent the power values ​​of the 1st, 2nd, and nth characteristic trend points, respectively. , , and They are respectively in the wind speed range [ , B-spline basis functions of orders 0, 1, 2, and 3 on [ ] , , They are respectively in the wind speed range [ , B-spline basis functions of orders 0, 1, and 2 on [ ] It is the fitted wind speed-power dynamic characteristic curve, representing the theoretical power at a given wind speed v.

[0008] Furthermore, the wind speed-power decision range includes: Based on the constructed wind speed-power dynamic characteristic curve, the deviation distance from the data points near each characteristic trend point to the dynamic characteristic curve is calculated. The MAD method is used to perform robustness analysis on the deviation to find the upper and lower boundary points. Then, the B-spline interpolation method is used to fit the upper and lower boundary points to obtain the upper and lower boundaries. The area within the upper and lower boundaries is the wind speed-power decision interval.

[0009] Furthermore, the MAD method includes: Calculate the residuals: ; Median residual: ; Calculate the absolute deviation: ; Calculate MAD: ; Calculate the upper and lower boundaries: ; ; in, Indicates the first The actual power and the power deviation on the power dynamic characteristic curve corresponding to each characteristic trend point. It is the first Power values ​​near a characteristic trend point yes The power value on the corresponding power dynamic characteristic curve, It is the first The power deviation on the actual power and power dynamic characteristic curve corresponding to the nth characteristic trend point. (·) is the median function. It is the median of the residuals. It is the absolute deviation of the median. It is the first The upper boundary point corresponding to each characteristic trend point It is the first The lower boundary point corresponding to each characteristic trend point.

[0010] Furthermore, the constructed wind speed-power dynamic characteristic curve and decision interval are considered as power dynamic curve knowledge; wherein, Wind speed-power dynamic characteristic curve: ; The wind speed-power decision range includes: Upper boundary: ; Lower boundary: ; in, and These are the dynamic operating curves corresponding to wind speeds, and the power values ​​corresponding to the upper and lower decision boundaries v. It's the wind speed value. It's the cut-in wind speed. That is the rated wind speed. , , , These are the turning wind speeds corresponding to the upper and lower boundary curves; It is the rated power of the wind speed-power dynamic characteristic curve. It is the fitted wind speed-power dynamic characteristic curve, representing the theoretical power at a given wind speed v.

[0011] Furthermore, the SBLS prediction model containing physical nodes is composed of K stacked BLS models containing physical nodes: The kth BLS input is: when k=1: , ; When k>1: , k=2,…,K; The kth BLS feature node: ; ; The k-th BLS enhancement node: ; ; Physical nodes: ; Augmented matrix: ; Output of the kth BLS: ; in: ; Final output: ; Among them, SBLS contains K BLS, It is the first The goal of BLS It is the input matrix of each BLS. This represents the input of the first BLS. Indicates the goal of the first BLS. This represents the feature matrix constructed from the training set data. Represents the power matrix of the training set. , and It is the first , and the The output of each BLS, It is a vector composed of wind speed data from the training set. It is the feature mapping node of the p-th group in the k-th BLS module. It is the feature mapping node of the 1st, 2nd, ..., Pth group of the kth BLS module. The feature mapping node of the Pth group is defined as... , , and These are the activation function, weight matrix, and bias term of the feature node, respectively. It is the enhancement mapping node of the qth group of the kth BLS module. These represent the augmentation mapping nodes of the 1st, 2nd, ..., Qth groups of the kth BLS module, respectively. The Qth group of augmentation mapping nodes is defined as follows: , , These are the activation function, weight matrix, and bias term of the augmentation node, respectively. It is a physical node. It is the weight of the power dynamic characteristic curve. It is the weight of the center value of the decision interval. and These are the dynamic operating curves, upper decision boundary, and lower decision boundary corresponding to wind speeds, respectively. The corresponding power value; It is the augmented matrix of the k-th BLS module. yes transpose, This is the output of the k-th BLS module. It is the regression parameter matrix of the k-th BLS module. It is the regularization term of the k-th BLS module. This is the final output.

[0012] To achieve the above objectives, according to a second aspect of the present invention, a wind power prediction system based on Physical Feature Enhancement (SBLS) is provided, the wind power prediction system comprising: The acquisition module is used to acquire the operating data of the wind turbine data acquisition and monitoring control SCADA system, including wind speed and power; and divides the preprocessed operating data into training set and test set; The module is used to build a wind speed-power scatter plot based on the running data in the training set. The wind speed interval is discretized by binning, that is, the wind speed is divided into intervals of 1 m / s, and the centroid of each interval is calculated and used as the feature trend point. The trend point is fitted by B-spline interpolation method to construct the wind speed-power dynamic characteristic curve. Based on the aforementioned wind speed-power dynamic characteristic curve, a wind speed-power decision interval is constructed. The training module is used to treat the constructed wind speed-power dynamic characteristic curve and decision interval as power dynamic curve knowledge, and map them into physical nodes that are parallel to feature nodes and enhancement nodes, so as to obtain an SBLS prediction model containing physical nodes. The embedded SBLS prediction model with physical nodes is trained using the training set. The outputs of feature nodes, augmentation nodes, and physical nodes are merged to construct an augmented matrix. The output weights are then analytically solved using the ridge regression pseudo-inverse algorithm. The application module is used to apply the trained SBLS prediction model with physical nodes to the test set for wind power prediction.

[0013] To achieve the above objectives, according to a third aspect of the present invention, a computer-readable storage medium is provided storing a computer program, which, when executed by a processor, is used to implement the wind power prediction method based on physical feature enhancement SBLS as described above.

[0014] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: This invention introduces a method for constructing wind speed-power dynamic characteristic curves and wind speed-power decision intervals. First, a wind speed-power scatter plot is established with wind speed as the horizontal axis and power as the vertical axis. Then, the continuous wind speed domain is discretized using a binning method, and the centroid of each interval is extracted as a representative trend point. The wind speed-power dynamic characteristic curve is obtained by fitting these trend points using B-spline interpolation. Then, all characteristic trend points are traversed, and the deviation of data points near the trend points from the power curve is calculated. Based on these deviations, MAD is used to identify upper and lower boundary points, and then B-spline interpolation is used to fit the upper and lower boundaries. Points within the upper and lower boundaries are considered as decision intervals. This invention treats the wind speed-power dynamic characteristic curve and wind speed-power decision interval as knowledge of the power dynamic operation curve. This provides a reliable prior foundation for subsequent embedding of SBLS prediction models and constructing SBLS with physical nodes.

[0015] In this invention, by constructing dedicated physical nodes, the extracted dynamic operating curve knowledge is explicitly embedded into the network topology of the stacked width learning system. Compared to traditional SBLS prediction models without physical nodes, this method significantly enhances the model's fitting ability and generalization performance when dealing with the complex nonlinear and fluctuating conditions of wind power. Furthermore, by effectively suppressing the irrationality of prediction results through physical nodes, it improves the accuracy and physical interpretability of the prediction results.

[0016] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0017] The accompanying drawings, as part of this invention, are provided to further illustrate the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation thereof. Clearly, the drawings described below are merely some embodiments; those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0018] In the attached diagram: Figure 1 This is a flowchart illustrating the wind power prediction method based on Physical Feature Enhancement (SBLS) in this specific embodiment. Figure 2 This is a scatter plot of wind speed-power on the training set in this specific embodiment; Figure 3 This is a schematic diagram of the power dynamic characteristic curve identified based on B-spline interpolation in this specific embodiment; Figure 4 This is a schematic diagram of the wind speed-power decision interval constructed based on the MAD method in this specific embodiment; Figure 5 This is a prediction fitting curve of the SBLS model with physical feature enhancement in this specific embodiment; Figure 6 This is a schematic diagram of the wind power prediction system based on physical feature enhancement SBLS in this specific embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments will be clearly and completely described below with reference to the accompanying drawings. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0020] Please see Figure 1 This invention provides a wind power prediction method based on Physical Feature Enhancement (SBLS), comprising the following steps: Acquire operational data from the wind turbine data acquisition and monitoring control SCADA system, including wind speed and power; divide the preprocessed operational data into training and testing sets; Based on the operational data in the training set, a wind speed-power scatter plot was established. The wind speed interval was discretized using the binning method, that is, the wind speed was divided into sections with an interval of 1 m / s, and the centroid of each interval was calculated. The centroid was used as the characteristic trend point. The trend point was fitted using the B-spline interpolation method to construct the wind speed-power dynamic characteristic curve. Based on the aforementioned wind speed-power dynamic characteristic curve, a wind speed-power decision interval is constructed. The constructed wind speed-power dynamic characteristic curve and decision interval are regarded as power dynamic curve knowledge, and mapped to physical nodes in parallel with feature nodes and enhancement nodes, thus obtaining the SBLS prediction model containing physical nodes, that is, the SBLS prediction model with physical feature enhancement. The embedded SBLS prediction model with physical nodes is trained using the training set. The outputs of feature nodes, augmentation nodes, and physical nodes are merged to construct an augmented matrix. The output weights are then analytically solved using the ridge regression pseudo-inverse algorithm. The trained SBLS prediction model with physical nodes was applied to the test set for wind power prediction.

[0021] In this embodiment, the power generation of wind turbines is used as the prediction target, and a physical feature enhanced SBLS prediction model is established. First, a wind speed-power scatter plot is established with wind speed as the independent variable and output power as the dependent variable. Based on the changes in wind speed, the wind speed intervals are discretized using a binning method. Then, the centroid of each wind speed interval is calculated, serving as the characteristic trend point for that interval. By determining the wind speed and power values ​​at 95% of the maximum power, the rated power and rated wind speed are determined, and a constant power range is set after the rated wind speed to ensure that these points reflect the actual variation between wind speed and power. Next, all trend points are arranged in ascending order according to wind speed, and B-spline interpolation is used to fit these points, generating a smooth power dynamic characteristic curve. Based on this curve, a wind speed-power decision interval is further constructed to define the upper and lower limits of power prediction. By analyzing the power residuals near the characteristic trend points, the MAD method is used to evaluate the distribution of the residuals, thereby determining the upper and lower limits of power prediction. These upper and lower limit boundary points are also fitted using B-spline interpolation to construct decision intervals that conform to physical laws. This invention treats the wind speed-power dynamic characteristic curve and decision interval as dynamic power curve knowledge. Finally, this dynamic power curve knowledge is embedded into the SBLS model, forming physical nodes. During the training process of the SBLS model, external physical laws are introduced, thereby adding physical constraints during data learning. In this way, the SBLS model can effectively combine data-driven learning with prior physical knowledge, improving the model's predictive ability while ensuring that the prediction results conform to the physical characteristics of the wind turbine in actual operation. This invention provides a more accurate solution for wind power prediction that conforms to actual operating laws by integrating physical knowledge and data learning. By introducing the dynamic characteristics of wind speed-power and physically meaningful decision intervals, prediction accuracy can be effectively improved, and the model's prediction results can better match the actual operating characteristics of wind turbines.

[0022] This embodiment employs a wind power prediction method based on Physical Feature Enhancement (SBLS), as shown in the flowchart below. Figure 1 As shown, the specific steps include: Step 1: Acquire operational data from a real-world SCADA (Supervisory Control and Data Acquisition) system for a generating unit, including key parameters such as wind speed and power. Raw data often contains missing or outlier values. To ensure high-quality input data, this embodiment employs a rapid cleaning strategy based on global statistical distribution. Specific steps include: using a sliding window quartile method to detect outliers for key wind speed and power data to avoid the impact of extreme data on the model; and uniformly correcting missing and outlier values ​​using ridge regression extrapolation to ensure data continuity and the integrity of the time series.

[0023] Step two involves dividing the processed data into a training set and a test set. Specifically, the first 80% of the samples are used as the training set to build knowledge of the power dynamic curve and train the prediction model; while the remaining 20% ​​of the samples are used as the test set to evaluate the generalization ability and actual prediction performance of the proposed model.

[0024] Step 3: Based on the training data obtained in Step 2, construct a scatter plot with wind speed as the independent variable and power as the dependent variable, as shown below. Figure 2 As shown, the wind speed intervals are discretized using a binning method, that is, the wind speed is divided into sections with an interval of 1 m / s, and the centroid of each interval is calculated and used as the characteristic trend point. Then, the B-spline interpolation method is used to fit these trend points to obtain a continuous and smooth wind speed-power dynamic characteristic curve, as shown. Figure 3 As shown, this characterizes the nonlinear mapping relationship between wind speed and power.

[0025] In this embodiment, the method used to identify feature trend points refers to: A wind speed-power scatter plot was created with wind speed as the independent variable and power as the dependent variable. The wind speed was then divided into zones with intervals of 1 m / s. The centroid of each zone was calculated. Specifically, the x-axis of the characteristic trend point represents the average wind speed of the current zone, and the y-axis represents the average power value of the zone. The calculated centroids were used as characteristic trend points, and B-spline interpolation was used to fit these trend points to obtain the wind speed-power dynamic characteristic curve.

[0026] In this embodiment, the B-spline interpolation method is as follows: Feature trend points identified on the training set The wind speed-power dynamic characteristic curve is fitted using the B-spline interpolation method. The B-spline interpolation method is defined as follows: Given B-spline basis functions: Basis functions of order 0: ; Basis functions of order 1: ; Basis functions of order 2: ; Basis functions of order 3: ; Wind speed-power dynamic characteristic curve fitted by B-spline interpolation: ; in, It is the first Wind speed values ​​at each characteristic trend point , , , They represent the first , , , Wind speed values ​​at each characteristic trend point , , These represent the wind speed values ​​at the 1st, 2nd, and nth characteristic trend points, respectively. It is the power value of the i-th characteristic trend point. , , These represent the power values ​​of the 1st, 2nd, and nth characteristic trend points, respectively. , , and They are respectively in the wind speed range [ , B-spline basis functions of orders 0, 1, 2, and 3 on [ ] , , They are respectively in the wind speed range [ , B-spline basis functions of orders 0, 1, and 2 on [ ] It is the fitted wind speed-power dynamic characteristic curve, representing the theoretical power at a given wind speed v.

[0027] Step four: Based on the wind speed-power dynamic characteristic curve constructed in step (3), this embodiment further constructs the wind speed-power decision interval as follows: Figure 4 As shown in the diagram. The specific steps are as follows: First, calculate the deviation distance between the nearby data points and the dynamic characteristic curve for each characteristic trend point; then, use the MAD method to perform robustness analysis on these deviations, determine the upper and lower bounds of the residuals, and obtain the decision points; finally, use the B-spline interpolation method to fit the upper and lower bounds to form continuous upper and lower boundaries. The region between the upper and lower boundaries is the decision interval, used to limit the reasonable range of power prediction.

[0028] In this embodiment, the wind speed-power decision interval construction method refers to: Based on the previously constructed wind speed-power dynamic characteristic curve, the deviation distance between the data points near each characteristic trend point and the dynamic characteristic curve is calculated. The MAD method is used to perform robustness analysis on these deviations to find the upper and lower boundary points. Then, the B-spline interpolation method is used again to fit the upper and lower boundaries to obtain the upper and lower boundaries. The points within the boundaries are regarded as the decision interval.

[0029] MAD stands for: Calculate the residuals: ; Median residual: ; Calculate the absolute deviation: ; Calculate MAD: ; Calculate the upper and lower boundary points: ; ; in, It is the first Power values ​​near a characteristic trend point yes The power value on the corresponding power dynamic characteristic curve, It is the power deviation between the actual power and the power dynamic characteristic curve. (·) is the median function. It is the median of the residuals. It is the absolute deviation of the median. It is the first The upper boundary point corresponding to each characteristic trend point It is the first The lower boundary point corresponding to each characteristic trend point.

[0030] Step five: Building upon step four, this embodiment treats the identified wind speed-power dynamic characteristic curve and decision interval as power dynamic curve knowledge and embeds them into the SBLS prediction model to obtain SBLS with physical nodes, i.e., SBLS with enhanced physical features. To enhance the physical interpretability of the model, nodes are introduced into the SBLS, making the model's prediction results consistent with the actual operating characteristics of wind turbines.

[0031] In this embodiment, the knowledge of the power dynamic curve refers to: Wind speed-power dynamic characteristic curve: ; Wind speed-power decision range: Upper boundary: ; Lower boundary: ; in, and These are the dynamic operating curves corresponding to wind speeds, and the power values ​​corresponding to the upper and lower decision boundaries v. It's the wind speed value. It's the cut-in wind speed. That is the rated wind speed. These are the turning wind speeds corresponding to the upper and lower boundary curves; It is the rated power of the wind speed-power dynamic characteristic curve.

[0032] Step six: Train the SBLS model containing physical nodes using the training set. Merge the outputs of feature nodes, augmentation nodes, and physical nodes to construct an augmented matrix. Utilize the ridge regression pseudo-inverse algorithm to analytically solve for the output weights, avoiding the tedious iterative training process.

[0033] In this embodiment, the SBLS prediction model containing physical nodes refers to: The kth BLS input is: when k=1: , ; When k>1: , k=2,…,K; The kth BLS feature node: ; ; The k-th BLS enhancement node: ; ; Physical nodes: ; Augmented matrix: ; Output of the kth BLS: ; in: ; Final output: ; Among them, SBLS contains K BLS, It is the first The goal of BLS It is the input matrix of each BLS. This represents the input of the first BLS. Indicates the goal of the first BLS. This represents the feature matrix constructed from the training set data. Represents the power matrix of the training set. , and It is the first , and the The output of each BLS, It is a vector composed of wind speed data from the training set. It is the feature mapping node of the p-th group in the k-th BLS module. It is the feature mapping node of the 1st, 2nd, ..., Pth group of the kth BLS module. The feature mapping node of the Pth group is defined as... , , and These are the activation function, weight matrix, and bias term of the feature node, respectively. It is the enhancement mapping node of the qth group of the kth BLS module. These represent the augmentation mapping nodes of the 1st, 2nd, ..., Qth groups of the kth BLS module, respectively. The Qth group of augmentation mapping nodes is defined as follows: , , These are the activation function, weight matrix, and bias term of the augmentation node, respectively. It is a physical node. It is the weight of the power dynamic characteristic curve. It is the weight of the center value of the decision interval. and These are the dynamic operating curves, upper decision boundary, and lower decision boundary corresponding to wind speeds, respectively. The corresponding power value; It is the augmented matrix of the k-th BLS module. yes transpose, This is the output of the k-th BLS module. It is the regression parameter matrix of the k-th BLS module. It is the regularization term of the k-th BLS module. This is the final output.

[0034] Step seven, finally, evaluate the performance of the prediction results. RMSE, MAE, and R² metrics are used to quantitatively assess the model's prediction accuracy.

[0035] In this embodiment, RMSE, MAE, R 2 This refers to: RMSE definition: ; MAE definition: ; R 2 definition: ; Where m is the amount of data in the test set. Is the model in the first Power prediction values ​​at each time step. It is the first The actual power value at each time step It is the average power on the test set.

[0036] Step 8: The physical feature-enhanced SBLS prediction model is compared and analyzed with the standard SBLS prediction model, as well as with the deep learning BiGRU. The evaluation metrics and training time of the three are shown in Table 1.

[0037] Table 1: Quantitative Evaluation Table of Performance Indicators for Standard SBLS and Physically Enhanced SBLS

[0038] As shown in Table 1, this method demonstrates a significant performance improvement compared to BiGRU and standard SBLS. Regarding accuracy, thanks to the constraints of physical principles, this method achieves the lowest RMSE and the highest R-value. 2 It even outperforms deep neural networks. In terms of efficiency, this method basically maintains the fast computation characteristics of SBLS, taking only 12.2 seconds, with only a small increase in computational overhead compared to standard SBLS. This proves that this method is a lightweight solution that combines high accuracy and high efficiency.

[0039] The SBLS prediction results proposed in this method based on physical feature enhancement are as follows: Figure 5 As shown in the figure, it can be seen that this method can fit the trend of wind power change well, and the prediction results have a high degree of consistency with the actual values.

[0040] After verifying the embedding of dynamic operating mechanism knowledge into the model, this method improves prediction accuracy, generalization ability, and physical interpretability while ensuring the computational efficiency and training speed of the model, thus proving the feasibility and superiority of the method in this embodiment.

[0041] Please see Figure 6 The present invention also provides a wind power prediction system based on Physical Feature Enhancement (SBLS), the wind power prediction system comprising: The acquisition module is used to acquire the operating data of the wind turbine data acquisition and monitoring control SCADA system, including wind speed and power; and divides the preprocessed operating data into training set and test set; The module is used to build a wind speed-power scatter plot based on the running data in the training set. The wind speed interval is discretized by binning, that is, the wind speed is divided into intervals of 1 m / s, and the centroid of each interval is calculated and used as the feature trend point. The trend point is fitted by B-spline interpolation method to construct the wind speed-power dynamic characteristic curve. Based on the aforementioned wind speed-power dynamic characteristic curve, a wind speed-power decision interval is constructed. The training module is used to treat the constructed wind speed-power dynamic characteristic curve and decision interval as power dynamic curve knowledge, and map them into physical nodes that are parallel to feature nodes and enhancement nodes, so as to obtain an SBLS prediction model containing physical nodes. The SBLS prediction model containing physical nodes is trained using the training set. The outputs of feature nodes, augmentation nodes, and physical nodes are merged to construct an augmented matrix. The output weights are then analytically solved using the ridge regression pseudo-inverse algorithm. The application module is used to apply the trained SBLS prediction model with physical nodes to the test set for wind power prediction.

[0042] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the wind power prediction method based on physical feature enhancement SBLS as described above.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-described technical content to create equivalent embodiments without departing from the scope of the present invention. The implementation schemes in the above embodiments can also be further combined or replaced. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A wind power prediction method based on Physical Feature Enhancement (SBLS), characterized in that, Includes the following steps: Acquire operational data from the wind turbine data acquisition and monitoring control SCADA system, including wind speed and power; divide the preprocessed operational data into training and testing sets; Based on the operational data in the training set, a wind speed-power scatter plot was established. The wind speed interval was discretized using the binning method, that is, the wind speed was divided into intervals of 1 m / s, and the centroid of each interval was calculated. The centroid was used as the characteristic trend point. The trend points were fitted using the B-spline interpolation method to construct the wind speed-power dynamic characteristic curve; Based on the constructed wind speed-power dynamic characteristic curve, a wind speed-power decision interval is constructed. The constructed wind speed-power dynamic characteristic curve and decision interval are regarded as power dynamic curve knowledge. The power dynamic curve knowledge is mapped to physical nodes that are parallel to feature nodes and enhancement nodes, resulting in an SBLS prediction model containing physical nodes, i.e., an SBLS prediction model with physical feature enhancement. The SBLS prediction model containing physical nodes is trained using the training set. The outputs of feature nodes, augmentation nodes, and physical nodes are merged to construct an augmented matrix. The output weights are then analytically solved using the ridge regression pseudo-inverse algorithm. The trained SBLS prediction model with physical nodes was applied to the test set for wind power prediction.

2. The method according to claim 1, characterized in that, The preprocessing includes using the sliding window quartile method to detect outliers for wind speed and power data; and using ridge regression extrapolation to correct missing and outlier values ​​to ensure data continuity and the integrity of the time series.

3. The method according to claim 2, characterized in that, The method of fitting trend points using B-spline interpolation to construct the wind speed-power dynamic characteristic curve includes: Feature trend points identified on the training set The wind speed-power dynamic characteristic curve is fitted using the B-spline interpolation method; the B-spline interpolation method is defined as follows: Given B-spline basis functions: Basis functions of order 0: ； Basis functions of order 1: ； Basis functions of order 2: ； Basis functions of order 3: ； Wind speed-power dynamic characteristic curve fitted by B-spline interpolation: ； in, It is the first Wind speed values ​​at each characteristic trend point , , , They represent the first , , , Wind speed values ​​at each characteristic trend point , , These represent the wind speed values ​​at the 1st, 2nd, and nth characteristic trend points, respectively. It is the power value of the i-th characteristic trend point. , , These represent the power values ​​of the 1st, 2nd, and nth characteristic trend points, respectively. , , and They are respectively in the wind speed range [ , B-spline basis functions of orders 0, 1, 2, and 3 on [ ] , , They are respectively in the wind speed range [ , B-spline basis functions of orders 0, 1, and 2 on [ ] It is the fitted wind speed-power dynamic characteristic curve, representing the theoretical power at a given wind speed v.

4. The method according to claim 3, characterized in that, The wind speed-power decision range includes: Based on the constructed wind speed-power dynamic characteristic curve, the deviation distance from the data points near each characteristic trend point to the dynamic characteristic curve is calculated. The MAD method is used to perform robustness analysis on the deviation to find the upper and lower boundary points. Then, the B-spline interpolation method is used to fit the upper and lower boundary points to obtain the upper and lower boundaries. The area within the upper and lower boundaries is the wind speed-power decision interval.

5. The method according to claim 4, characterized in that, The MAD method includes: Calculate the residuals: ； Median residual: ； Calculate the absolute deviation: ； Calculate MAD: ； Calculate the upper and lower boundaries: ； ； in, Indicates the first The actual power and the power deviation on the power dynamic characteristic curve corresponding to each characteristic trend point. It is the first Power values ​​near a characteristic trend point yes The power value on the corresponding power dynamic characteristic curve, It is the first The power deviation on the actual power and power dynamic characteristic curve corresponding to the nth characteristic trend point. (·) is the median function. It is the median of the residuals. It is the absolute deviation of the median. It is the first The upper boundary point corresponding to each characteristic trend point It is the first The lower boundary point corresponding to each characteristic trend point.

6. The method according to claim 1, characterized in that, The constructed wind speed-power dynamic characteristic curve and decision interval are considered as power dynamic curve knowledge; where: Wind speed-power dynamic characteristic curve: ； The wind speed-power decision range includes: Upper boundary: ； Lower boundary: ； in, and These are the dynamic operating curves corresponding to wind speeds, and the power values ​​corresponding to the upper and lower decision boundaries v. It's the wind speed value. It's the cut-in wind speed. That is the rated wind speed. , , , These are the turning wind speeds corresponding to the upper and lower boundary curves; It is the rated power of the wind speed-power dynamic characteristic curve. It is the fitted wind speed-power dynamic characteristic curve, representing the theoretical power at a given wind speed v.

7. The method according to claim 1, characterized in that, The SBLS prediction model with physical nodes is composed of K stacked BLS models with physical nodes: The kth BLS input is: when k=1: , ; When k>1: , k=2,…,K; The kth BLS feature node: ； ； The k-th BLS enhancement node: ； ； Physical nodes: ； Augmented matrix: ； Output of the kth BLS: ； in: ； Final output: ； Among them, SBLS contains K BLS, It is the first The goal of BLS It is the input matrix of each BLS. This represents the input of the first BLS. Indicates the goal of the first BLS. This represents the feature matrix constructed from the training set data. Represents the power matrix of the training set. , and It is the first , and the The output of each BLS, It is a vector composed of wind speed data from the training set. It is the feature mapping node of the p-th group in the k-th BLS module. It is the feature mapping node of the 1st, 2nd, ..., Pth group of the kth BLS module. The feature mapping node of the Pth group is defined as... , , and These are the activation function, weight matrix, and bias term of the feature node, respectively. It is the enhancement mapping node of the qth group of the kth BLS module. These represent the augmentation mapping nodes of the 1st, 2nd, ..., Qth groups of the kth BLS module, respectively. The Qth group of augmentation mapping nodes is defined as follows: , , These are the activation function, weight matrix, and bias term of the augmentation node, respectively. It is a physical node. It is the weight of the power dynamic characteristic curve. It is the weight of the center value of the decision interval. and These are the dynamic operating curves, upper decision boundary, and lower decision boundary corresponding to wind speeds, respectively. The corresponding power value; It is the augmented matrix of the k-th BLS module. yes transpose, This is the output of the k-th BLS module. It is the regression parameter matrix of the k-th BLS module. It is the regularization term of the k-th BLS module. This is the final output.

8. A wind power prediction system based on Physical Feature Enhancement-Simplified SBLS, characterized in that, The wind power prediction system includes: The acquisition module is used to acquire the operating data of the wind turbine data acquisition and monitoring control SCADA system, including wind speed and power; and divides the preprocessed operating data into training set and test set; The module is used to build a wind speed-power scatter plot based on the running data in the training set. The wind speed interval is discretized by binning, that is, the wind speed is divided into intervals of 1 m / s, and the centroid of each interval is calculated and used as the feature trend point. The trend point is fitted by B-spline interpolation method to construct the wind speed-power dynamic characteristic curve. Based on the aforementioned wind speed-power dynamic characteristic curve, a wind speed-power decision interval is constructed. The training module is used to treat the constructed wind speed-power dynamic characteristic curve and decision interval as power dynamic curve knowledge, and to map the power dynamic curve knowledge as physical nodes alongside feature nodes and enhancement nodes, so as to obtain the SBLS prediction model containing physical nodes, i.e., the SBLS prediction model with physical feature enhancement. The SBLS prediction model containing physical nodes is trained using the training set. The outputs of feature nodes, augmentation nodes, and physical nodes are merged to construct an augmented matrix. The output weights are then analytically solved using the ridge regression pseudo-inverse algorithm. The application module is used to apply the trained SBLS prediction model with physical nodes to the test set for wind power prediction.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it is used to implement the wind power prediction method based on physical feature enhancement SBLS as described in any one of claims 1-7.

Images