Wind power climbing event prediction method and system based on feature extraction and deep learning, medium and equipment
Through feature extraction and deep learning methods, using parameter-adaptive revolving door algorithm and neural network technology, wind power ramping features are extracted from historical data, which solves the local optimal problem of the existing model and achieves high-precision prediction of wind power ramping events.
Patent Information
- Application Number
- CN202510001589.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-02
- Publication Date
- 2025-10-10
- Estimated Expiration
- 2045-01-02
AI Technical Summary
Existing wind power ramping event prediction models lack feature engineering, are prone to falling into local optimality, and ignore historical training status, resulting in insufficient prediction accuracy.
A method based on feature extraction and deep learning is adopted. The parameter-adaptive revolving door algorithm is used to extract ramp features from historical meteorological and wind power data. Convolutional neural networks and long short-term memory neural networks are combined for feature extraction and prediction. The gating parameters are adaptively adjusted to improve the prediction accuracy.
Accurately capturing the spatiotemporal dependencies and uncertainties in wind power and meteorological data provides accurate and information-rich wind power ramp-up event prediction results, providing a reliable forecast basis for the development of new energy.
Smart Images
Figure CN119940615B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of digital power grid and artificial intelligence technology, and in particular to a method, system, medium and equipment for predicting wind power ramping events based on feature extraction and deep learning. Background Art
[0002] Wind power ramping events refer to the phenomenon of rapid, short-term fluctuations in wind speed caused by meteorological factors such as severe low voltage, thunderstorms, or gusty winds during the actual grid connection process. Wind power ramping events can easily disrupt the system's power balance, threatening the reliability, security, and economic viability of the power system. Compared to wind power forecasting, ramping prediction differs significantly. Ramping is a low-probability event that occurs in the wind power sequence, and the prediction targets these events themselves. Prediction methods are mainly divided into direct and indirect prediction methods. Direct prediction typically relies on training models with a large amount of historical ramping data, while current research on wind power ramping event prediction mostly uses indirect prediction methods, identifying ramping events based on wind power forecast results to ultimately obtain prediction results. Because ramping events increase the difficulty of analyzing wind power output characteristics, existing research often uses wind power-related characteristic data as model input to improve prediction accuracy. However, these methods often lack feature engineering support, which can cause the model to become trapped in local optima and ignore the influence of historical training states. Summary of the Invention
[0003] In response to the above problems, the purpose of the present invention is to provide a wind power ramp event prediction method, system, medium and equipment based on feature extraction and deep learning, so as to solve the technical problems that the existing wind power ramp prediction model lacks feature engineering, is prone to falling into local optimality, and ignores historical training status, thereby improving the accuracy of wind power ramp event prediction.
[0004] To achieve the above-mentioned objectives, in the first aspect, the technical solution adopted by the present invention is: a method for predicting wind power ramping events based on feature extraction and deep learning, which includes: selecting historical meteorological data related to wind power prediction, and corresponding historical wind power data, using the historical meteorological data as wind power prediction features, and forming a wind power prediction feature set with the historical wind power data; using a parameter-adaptive revolving door algorithm to extract wind power ramping events from the feature set, and processing the historical data into a variety of ramping features; using meteorological data, historical wind power data and the obtained ramping features as inputs to the prediction model, and extracting features twice through a convolutional neural network to obtain a feature map; inputting the feature map into a long short-term memory neural network to obtain predicted wind power; using the forecast results of wind power to extract ramping events through a parameter-adaptive revolving door algorithm to obtain prediction results of wind power ramping events.
[0005] Furthermore, the parameter-adaptive revolving door algorithm includes:
[0006] Calculate the upper door opening slope and the lower door opening slope between the detection point and the previous storage point;
[0007] A discrimination rule is used to save the upper door opening slope greater than the slope threshold and the lower door opening slope less than the slope threshold. Based on the newly saved door opening slope, it is determined whether to continue traversing the data or temporarily store the wind power point.
[0008] Calculate the compression error between the original data and the temporarily stored data, compare the compression error with the set threshold, if the compression error is greater than the set threshold, reduce the gating parameter and return to the previous storage point, and re-perform the revolving door algorithm to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, add the gating parameter and then perform the same operation as the one greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored;
[0009] Determine the ramp start time for wind power ramp-up and ramp-down events based on the clear definition and conditions of wind power ramp events , climbing rate , climbing duration Three characteristic parameters.
[0010] Furthermore, the judgment rules are:
[0011]
[0012] in:
[0013]
[0014] Where, The upward door opening slope stored for point n; The slope of the upper door opening calculated for point n; The upward door opening slope stored at point n-1; The downward door opening slope stored for point n; The slope of the lower door opening calculated for point n; The lower door opening slope stored at point n-1; represents the wind power at point a; represents the wind power at point b; represents the moment of point a; represents the moment of point b; Represents the gating parameters of the revolving door algorithm.
[0015] Furthermore, the compression error between the original data and the temporarily stored data is calculated, including:
[0016]
[0017] Where, represents the compression error of the algorithm; is the original number of data points; is the original power of the kth point, is the power of the kth point after the compressed signal is restored.
[0018] Furthermore, the compression error is compared with the set threshold. If the compression error is greater than the set threshold, the gating parameter is reduced and the previous storage point is returned to, and the revolving door algorithm is re-performed to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, the gating parameter is added, and then the same operation is performed as if it is greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored as:
[0019]
[0020] Where, Indicates the setting of the initial gating parameter value; For a single parameter growth; represents the gating parameter value after adaptive adjustment; Represents the gating parameter value before adaptive adjustment; is the process coefficient, the initial value is 1; represents the compression error of the algorithm; To set the threshold; is the maximum allowable deviation.
[0021] Furthermore, the convolutional neural network model is:
[0022]
[0023] Where, Indicates the Layer feature map output; Represents the convolution operation; To connect Tier The feature map and Tier The convolution kernel weight matrix between feature maps; is the bias matrix; is the set of input feature maps; For the Layer feature map output; is the activation function.
[0024] Furthermore, the long short-term memory neural network model is:
[0025]
[0026]
[0027]
[0028] in, as well as Represent the weights and biases of the forget gate, input gate, and output gate respectively; Represent the outputs of the forget gate, input gate, and output gate respectively; Represents the sigmoid function; Represents vector concatenation; represents the hidden state of the long short-term memory neural network model at time t-1, Represents the feature map at time t.
[0029] In the second aspect, the technical solution adopted by the present invention is: a wind power climbing event prediction system based on feature extraction and deep learning, which includes: a data acquisition unit, which selects historical meteorological data related to wind power prediction, and corresponding historical wind power data, uses the historical meteorological data as wind power prediction features, and forms a wind power prediction feature set with the historical wind power data; a climbing event extraction unit, which uses a parameter-adaptive revolving door algorithm to extract wind power climbing events from the feature set, and processes the historical data into a variety of climbing features; a feature extraction unit, which uses meteorological data, historical wind power data and the obtained climbing features as inputs of the prediction model, and extracts features twice through a convolutional neural network to obtain a feature map; a wind power prediction unit, which inputs the feature map into a long short-term memory neural network to obtain predicted wind power; the forecast result of wind power is used to extract climbing events through a parameter-adaptive revolving door algorithm to obtain a prediction result of wind power climbing events.
[0030] Furthermore, the parameter-adaptive revolving door algorithm includes:
[0031] Calculate the upper door opening slope and the lower door opening slope between the detection point and the previous storage point;
[0032] A discrimination rule is used to save the upper door opening slope greater than the slope threshold and the lower door opening slope less than the slope threshold. Based on the newly saved door opening slope, it is determined whether to continue traversing the data or temporarily store the wind power point.
[0033] Calculate the compression error between the original data and the temporarily stored data, compare the compression error with the set threshold, if the compression error is greater than the set threshold, reduce the gating parameter and return to the previous storage point, and re-perform the revolving door algorithm to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, add the gating parameter and then perform the same operation as the one greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored;
[0034] Determine the ramp start time for wind power ramp-up and ramp-down events based on the clear definition and conditions of wind power ramp events , climbing rate , climbing duration Three characteristic parameters.
[0035] Furthermore, the judgment rules are:
[0036]
[0037] in:
[0038]
[0039] Where, The upward door opening slope stored for point n; The slope of the upper door opening calculated for point n; The upward door opening slope stored at point n-1; The downward door opening slope stored for point n; The slope of the lower door opening calculated for point n; The lower door opening slope stored at point n-1; represents the wind power at point a; represents the wind power at point b; represents the moment of point a; represents the moment of point b; Represents the gating parameters of the revolving door algorithm.
[0040] Furthermore, the compression error between the original data and the temporarily stored data is calculated, including:
[0041]
[0042] Where, represents the compression error of the algorithm; is the original number of data points; is the original power of the kth point, is the power of the kth point after the compressed signal is restored.
[0043] Furthermore, the compression error is compared with the set threshold. If the compression error is greater than the set threshold, the gating parameter is reduced and the previous storage point is returned to, and the revolving door algorithm is re-performed to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, the gating parameter is added, and then the same operation is performed as if it is greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored as:
[0044]
[0045] Where, Indicates the setting of the initial gating parameter value; For a single parameter growth; represents the gating parameter value after adaptive adjustment; Represents the gating parameter value before adaptive adjustment; is the process coefficient, the initial value is 1; represents the compression error of the algorithm; To set the threshold; is the maximum allowable deviation.
[0046] Furthermore, the convolutional neural network model is:
[0047]
[0048] Where, Indicates the Layer feature map output; Represents the convolution operation; To connect Tier The feature map and Tier The convolution kernel weight matrix between feature maps; is the bias matrix; is the set of input feature maps; For the Layer feature map output; is the activation function.
[0049] Furthermore, the long short-term memory neural network model is:
[0050]
[0051]
[0052]
[0053] in, as well as Represent the weights and biases of the forget gate, input gate, and output gate respectively; Represent the outputs of the forget gate, input gate, and output gate respectively; Represents the sigmoid function; Represents vector concatenation; represents the hidden state of the long short-term memory neural network model at time t-1, Represents the feature map at time t.
[0054] In a third aspect, the technical solution adopted by the present invention is: a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes any one of the above methods.
[0055] In a fourth aspect, the technical solution adopted by the present invention is: a computing device, comprising: one or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the above methods.
[0056] The present invention has the following advantages due to the adoption of the above technical solution:
[0057] The wind power ramp event prediction method, system, device and medium based on feature extraction and deep learning of the present invention use historical meteorological data as input data, detect wind power ramp events through parameter adaptive revolving door algorithm, and obtain ramp features. The historical wind power data and ramp features are used as input of the prediction model, and then the features are extracted twice through convolutional neural network, and then the relationship in the time series is learned through long short-term memory neural network. The predicted wind power is used as the output of the model, and after ramp detection, the prediction result of the ramp event is obtained. The present invention can accurately capture the complex spatiotemporal dependencies and uncertainties in wind power and meteorological data, and provide accurate and information-rich forecast results of wind power ramp events, providing a reference basis for the development of new energy. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] Figure 1 This is a schematic diagram of the steps of a wind power ramp event prediction method based on feature extraction and deep learning provided by a preferred embodiment of the present application;
[0059] Figure 2 This is a schematic diagram of the steps of a parameter adaptive revolving door algorithm provided by a preferred embodiment of the present application;
[0060] Figure 3 This is a schematic diagram of a wind power ramp event prediction system based on feature extraction and deep learning provided by a preferred embodiment of the present application;
[0061] Figure 4 This is a schematic diagram of a computer device provided by a preferred embodiment of the present application. DETAILED DESCRIPTION
[0062] In order to solve the technical problems that the existing wind power prediction model needs to rely on a large amount of feature engineering, is prone to falling into local optimality, and ignores the historical training status, the present invention provides a wind power ramp event prediction method, system, equipment and medium based on feature extraction and deep learning. Historical meteorological data is used as input data, and the wind power ramp events are detected through the parameter adaptive revolving door algorithm to obtain the ramp features. The historical wind power data and the ramp features are used as the input of the prediction model, and then the features are extracted twice through the convolutional neural network, and then the relationship in the time series is learned through the long short-term memory neural network. The predicted wind power is used as the output of the model, and after the ramp detection is performed, the prediction result of the ramp event is obtained. The present invention can accurately capture the complex spatiotemporal dependencies and uncertainties in wind power and meteorological data, and provide accurate and information-rich forecast results of wind power ramp events, providing a reference basis for the development of new energy.
[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.
[0064] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.
[0065] In one embodiment of the present invention, a method for predicting wind power ramping events based on feature extraction and deep learning is provided to solve the technical problems that the existing wind power prediction model needs to rely on a large amount of feature engineering, is prone to falling into local optimality, and ignores historical training status, thereby effectively improving the accuracy of wind power ramping event prediction. In this embodiment, if Figure 1 As shown, the method includes the following steps:
[0066] S1. Select historical meteorological data related to wind power prediction and corresponding historical wind power data, use the historical meteorological data as wind power prediction features, and form a wind power prediction feature set with the historical wind power data;
[0067] S2. Adopt parameter-adaptive revolving door algorithm to extract wind power ramping events from feature set and process historical data into multiple ramping features;
[0068] S3, meteorological data, historical wind power data and the obtained climbing characteristics will be as the input of the prediction model, the features are extracted twice through the convolutional neural network to obtain the feature map;
[0069] S4, the feature map is input into the long short-term memory neural network to obtain the predicted wind power;
[0070] S5, the prediction result of the wind power is extracted through the parameter adaptive rotating door algorithm to obtain the prediction result of the wind power climbing event.
[0071] In the embodiment, the numerical weather prediction related to the wind power prediction is selected from the global scale numerical weather prediction model output, including 100-meter vertical direction wind speed, 100-meter horizontal direction wind speed, temperature, humidity, air pressure, and corresponding historical wind power data, and the original meteorological parameters and historical wind power data are cleaned to construct the wind power prediction feature set.
[0072] In the above step S2, the parameter adaptive rotating door algorithm includes the following steps:
[0073] S21, data processing based on the original SDA algorithm. The upper opening door slope and the lower opening door slope between the detection point and the previous storage point are calculated.
[0074] S22, the upper opening door slope greater than the slope threshold and the lower opening door slope less than the slope threshold are saved by using the discrimination rule, and whether to continue to traverse the data or to temporarily store the wind power point is determined according to the newly saved opening door slope;
[0075] S23, data processing based on parameter adaptive selection. The compression error between the original data and the temporarily stored data is calculated, and the compression error and the set threshold are compared:
[0076] If the compression error is greater than the set threshold, the gating parameter is reduced and the previous storage point is returned, and the rotating door algorithm is performed again to determine whether the power point is in the opening door range; if the compression error is less than the set threshold, the gating parameter is added, and then the same operation as greater than the set threshold is performed; if the compression error is within the threshold range, the point will be finally stored;
[0077] S24, identification based on the definition of wind power ramp event. According to the definition and conditions of the explicit wind power ramp event, the ramp start time , the ramp rate , and the ramp duration of the wind power generation up-ramp event and the down-ramp event are determined.
[0078] In the above step S22, the upper opening door slope greater than the slope threshold and the lower opening door slope less than the slope threshold are saved by using the discrimination rule, wherein the discrimination rule is:
[0079]
[0080] in:
[0081]
[0082] Where, The upward door opening slope stored for point n; The upper door opening slope calculated for point n; The upward door opening slope stored at point n-1; The downward door opening slope stored for point n; The slope of the lower door opening calculated for point n; The lower door opening slope stored at point n-1; represents the wind power at point a; represents the wind power at point b; represents the moment of point a; represents the moment of point b; Represents the gating parameters of the revolving door algorithm.
[0083] In the above step S23, the compression error between the original data and the temporarily stored data is calculated as:
[0084]
[0085] Where, represents the compression error of the algorithm; is the original number of data points; is the original power of the kth point, is the power of the kth point after the compressed signal is restored.
[0086] In this embodiment, the compression error is compared with the set threshold. If the compression error is greater than the set threshold, the gating parameter is reduced and the previous stored point is returned to, and the revolving door algorithm is re-performed to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, the gating parameter is added, and then the same operation as when the power point is greater than the set threshold is performed; if the compression error is within the threshold range, the point will be finally stored as:
[0087]
[0088] Where, Indicates the setting of the initial gating parameter value; For a single parameter growth; represents the gating parameter value after adaptive adjustment; Represents the gating parameter value before adaptive adjustment; is the process coefficient, the initial value is 1; represents the compression error of the algorithm; To set the threshold; is the maximum allowable deviation.
[0089] In the above step S3, meteorological data, historical wind power data and the obtained ramp characteristics are used as inputs of the prediction model, and then the features are extracted twice by the convolutional neural network to obtain a feature map. The convolutional neural network model is:
[0090]
[0091] Where, Indicates the Layer feature map output; Represents the convolution operation; To connect Tier The feature map and Tier The convolution kernel weight matrix between feature maps; is the bias matrix; is the set of input feature maps; For the Layer feature map output; is the activation function.
[0092] The pooling layer performs secondary feature extraction and information filtering on the output of the convolutional layer, thereby retaining the most significant features. By compressing the parameters and data through the pooling layer, it can effectively reduce overfitting and reduce the complexity of the network. The calculation process of the pooling layer is shown in the formula:
[0093]
[0094] Where, For the Layer feature map output; Indicates the Layer feature map output; represents the downsampling function.
[0095] In the above step S4, the long short-term memory neural network model is:
[0096]
[0097]
[0098]
[0099] in, as well as Represent the weights and biases of the forget gate, input gate, and output gate respectively; Represent the outputs of the forget gate, input gate, and output gate respectively; Represents the sigmoid function; Represents vector concatenation; represents the hidden state of the long short-term memory neural network model at time t-1, Represents the feature map at time t.
[0100] In this embodiment, the wind power forecast results are used to extract ramp events through a parameter-adaptive revolving door algorithm to obtain prediction results of wind power ramp events.
[0101] The wind power ramp event prediction method based on feature extraction and deep learning proposed in the present invention takes historical meteorological data as input data, detects wind power ramp events through a parameter adaptive revolving door algorithm, and obtains ramp features. The historical wind power data and ramp features are used as inputs of the prediction model, and then the features are extracted secondary through a convolutional neural network, and then the relationship in the time series is learned through a long short-term memory neural network. The predicted wind power is used as the output of the model, and after ramp detection, the prediction result of the ramp event is obtained. The present invention can accurately capture the complex spatiotemporal dependencies and uncertainties in wind power and meteorological data, and provide accurate and information-rich forecast results of wind power ramp events, providing a reference basis for the development of new energy.
[0102] In one embodiment of the present invention, a wind power ramp event prediction system based on feature extraction and deep learning is provided. Figure 3 As shown, it includes:
[0103] The data acquisition unit 1 selects historical meteorological data related to wind power prediction and corresponding historical wind power data, uses the historical meteorological data as wind power prediction features, and forms a wind power prediction feature set with the historical wind power data;
[0104] The ramp event extraction unit 2 uses a parameter-adaptive revolving door algorithm to extract wind power ramp events from the feature set and processes historical data into multiple ramp features;
[0105] Feature extraction unit 3 uses meteorological data, historical wind power data, and the obtained ramp characteristics as inputs to the prediction model, and extracts features again through a convolutional neural network to obtain a feature map;
[0106] The wind power prediction unit 4 inputs the feature map into the long short-term memory neural network to obtain the predicted wind power; the predicted result of the wind power is used to extract the climbing event through the parameter adaptive revolving door algorithm to obtain the prediction result of the wind power climbing event.
[0107] In the above embodiment, the parameter-adaptive revolving door algorithm includes:
[0108] Calculate the upper door opening slope and the lower door opening slope between the detection point and the previous storage point;
[0109] A discrimination rule is used to save the upper door opening slope greater than the slope threshold and the lower door opening slope less than the slope threshold. Based on the newly saved door opening slope, it is determined whether to continue traversing the data or temporarily store the wind power point.
[0110] Calculate the compression error between the original data and the temporarily stored data, compare the compression error with the set threshold, if the compression error is greater than the set threshold, reduce the gating parameter and return to the previous storage point, and re-perform the revolving door algorithm to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, add the gating parameter and then perform the same operation as the one greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored;
[0111] Determine the ramp start time for wind power ramp-up and ramp-down events based on the clear definition and conditions of wind power ramp events , climbing rate , climbing duration Three characteristic parameters.
[0112] In the above embodiment, the discrimination rule is:
[0113]
[0114] in:
[0115]
[0116] Where, The upward door opening slope stored for point n; The upper door opening slope calculated for point n; The upward door opening slope stored at point n-1; The downward door opening slope stored for point n; The slope of the lower door opening calculated for point n; The lower door opening slope stored at point n-1; represents the wind power at point a; represents the wind power at point b; represents the moment of point a; represents the moment of point b; Represents the gating parameters of the revolving door algorithm.
[0117] In the above embodiment, calculating the compression error between the original data and the temporarily stored data includes:
[0118]
[0119] Where, represents the compression error of the algorithm; is the original number of data points; is the original power of the kth point, is the power of the kth point after the compressed signal is restored.
[0120] Among them, the compression error is compared with the set threshold. If the compression error is greater than the set threshold, the gating parameter is reduced and the previous storage point is returned to, and the revolving door algorithm is re-performed to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, the gating parameter is added, and then the same operation is performed as if it is greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored as:
[0121]
[0122] Where, Indicates the setting of the initial gating parameter value; For a single parameter growth; represents the gating parameter value after adaptive adjustment; Represents the gating parameter value before adaptive adjustment; is the process coefficient, the initial value is 1; represents the compression error of the algorithm; To set the threshold; is the maximum allowable deviation.
[0123] In the above embodiment, the convolutional neural network model is:
[0124]
[0125] Where, Indicates the Layer feature map output; Represents the convolution operation; To connect Tier The feature map and Tier The convolution kernel weight matrix between feature maps; is the bias matrix; is the set of input feature maps; For the Layer feature map output; is the activation function.
[0126] In the above embodiment, the long short-term memory neural network model is:
[0127]
[0128]
[0129]
[0130] in, as well as Represent the weights and biases of the forget gate, input gate, and output gate respectively; Represent the outputs of the forget gate, input gate, and output gate respectively; Represents the sigmoid function; Represents vector concatenation; represents the hidden state of the long short-term memory neural network model at time t-1, Represents the feature map at time t.
[0131] The system provided in this embodiment is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for specific processes and detailed contents, which will not be repeated here.
[0132] In summary, the present invention addresses the technical problems that the existing wind power ramp-up event prediction model lacks feature engineering, is prone to falling into local optimality, and ignores historical training status. The present invention uses historical meteorological data as input data, detects wind power ramp-up events through a parameter-adaptive revolving door algorithm, and obtains ramp-up features. The historical wind power data and ramp-up features are used as inputs to the prediction model, and then the features are extracted secondary through a convolutional neural network, and then the relationship in the time series is learned through a long short-term memory neural network. The predicted wind power is used as the output of the model, and after performing ramp-up detection, the prediction result of the ramp-up event is obtained. The present invention can accurately capture the complex spatiotemporal dependencies and uncertainties in wind power and meteorological data, and provide accurate and information-rich forecast results of wind power ramp-up events, providing a reference basis for the development of new energy.
[0133] like Figure 4As shown, one embodiment of the present invention provides a computing device structure. The computing device may be a terminal and may include: a processor, a communications interface, a memory, a display, and an input device. The processor, communications interface, and memory communicate with each other via a communications bus. The processor is configured to provide computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and a computer program. When executed by the processor, the computer program implements the methods described in the above embodiments. The internal memory provides an environment for the operating system and computer program in the non-volatile storage medium to run. The communications interface is configured to communicate with an external terminal via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a network management system, NFC (near field communication), or other technologies. The display may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen layer covering the display, or may be buttons, a trackball, or a touchpad provided on the computing device housing, or may be an external keyboard, touchpad, or mouse. The processor may invoke logic instructions in the memory.
[0134] In addition, the logical instructions in the above-mentioned memory can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0135] In one embodiment of the present invention, a computer program product is provided, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments.
[0136] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium stores server instructions. The computer instructions enable a computer to execute the methods provided in the above embodiments.
[0137] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effects are similar to those of the above method embodiment, and will not be repeated here.
[0138] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0139] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A wind power ramp event prediction method based on feature extraction and deep learning, characterized in that: include: Selecting historical meteorological data related to wind power prediction and corresponding historical wind power data, using the historical meteorological data as wind power prediction features, and forming a wind power prediction feature set with the historical wind power data; Adopting parameter-adaptive revolving door algorithm, wind power ramping events are extracted from feature sets, and historical data are processed into multiple ramping features; The meteorological data, historical wind power data, and the obtained ramp characteristics are used as inputs to the prediction model. The convolutional neural network is used to extract the features and obtain the feature map. The feature map is input into the long short-term memory neural network to obtain the predicted wind power; The wind power forecast results are used to extract the ramp events through the parameter adaptive revolving door algorithm to obtain the prediction results of wind power ramp events.
2. The wind power ramp event prediction method based on feature extraction and deep learning according to claim 1, characterized in that: Parameter-adaptive revolving door algorithm, including: Calculate the upper door opening slope and the lower door opening slope between the detection point and the previous storage point; A discrimination rule is used to save the upper door opening slope greater than the slope threshold and the lower door opening slope less than the slope threshold. Based on the newly saved door opening slope, it is determined whether to continue traversing the data or temporarily store the wind power point. Calculate the compression error between the original data and the temporarily stored data, compare the compression error with the set threshold, if the compression error is greater than the set threshold, reduce the gating parameter and return to the previous storage point, and re-perform the revolving door algorithm to determine whether the power point is within the open gate range; if the compression error is less than the set threshold, add the gating parameter and then perform the same operation as the one greater than the set threshold; if the compression error is within the threshold range, the point will be finally stored; Determine the ramp start time for wind power ramp-up and ramp-down events based on the clear definition and conditions of wind power ramp events , climbing rate , climbing duration Three characteristic parameters.
3. The wind power ramp event prediction method based on feature extraction and deep learning as claimed in claim 2, characterized in that: The judgment rules are: in: Where, The upward door opening slope stored at point n; The upper door opening slope calculated for point n; The upward door opening slope stored at point n-1; The downward door opening slope stored for point n; The slope of the lower door opening calculated for point n; The lower door opening slope stored at point n-1; represents the wind power at point a; represents the wind power at point b; represents the moment of point a; represents the moment of point b; Represents the gating parameters of the revolving door algorithm.
4. The wind power ramp event prediction method based on feature extraction and deep learning as claimed in claim 2, characterized in that: Calculate the compression error between the original data and the temporarily stored data, including: Where, represents the compression error of the algorithm; is the original number of data points; is the original power of the kth point, is the power of the kth point after the compressed signal is restored.
5. The wind power ramp event prediction method based on feature extraction and deep learning as claimed in claim 4, characterized in that: Compare the compression error with the set threshold. If the compression error is greater than the set threshold, reduce the gating parameter and return to the previous storage point, and re-perform the revolving door algorithm to determine whether the power point is within the open gate range. If the compression error is less than the set threshold, add the gating parameter and then perform the same operation as if it is greater than the set threshold. If the compression error is within the threshold range, the point will be finally stored as follows: Where, Indicates the setting of the initial gating parameter value; For a single parameter growth; represents the gating parameter value after adaptive adjustment; Represents the gating parameter value before adaptive adjustment; is the process coefficient, the initial value is 1; represents the compression error of the algorithm; To set the threshold; is the maximum allowable deviation.
6. The wind power ramp event prediction method based on feature extraction and deep learning according to claim 1, characterized in that: The convolutional neural network model is: Where, Indicates the Layer feature map output; Represents the convolution operation; To connect Tier The feature map and Tier The convolution kernel weight matrix between feature maps; is the bias matrix; is the set of input feature maps; For the Layer feature map output; is the activation function.
7. The method for predicting wind power ramping events based on feature extraction and deep learning according to claim 1, characterized in that: The long short-term memory neural network model is: in, as well as Represent the weights and biases of the forget gate, input gate, and output gate respectively; Represent the outputs of the forget gate, input gate, and output gate respectively; Represents the sigmoid function; Represents vector concatenation; represents the hidden state of the long short-term memory neural network model at time t-1, Represents the feature map at time t.
8. A wind power ramp event prediction system based on feature extraction and deep learning, characterized in that: include: The data acquisition unit selects historical meteorological data related to wind power prediction and corresponding historical wind power data, uses the historical meteorological data as wind power prediction features, and forms a wind power prediction feature set with the historical wind power data; The ramp event extraction unit uses a parameter-adaptive revolving door algorithm to extract wind power ramp events from the feature set and processes historical data into multiple ramp features; The feature extraction unit uses meteorological data, historical wind power data, and the obtained ramp characteristics as inputs to the prediction model, and extracts features again through a convolutional neural network to obtain a feature map. The wind power prediction unit inputs the feature map into the long short-term memory neural network to obtain the predicted wind power; the forecast result of the wind power is used to extract the climbing event through the parameter adaptive revolving door algorithm to obtain the prediction result of the wind power climbing event.
9. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any one of the methods of claims 1 to 7 .
10. A computing device, characterized in that include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any one of the methods according to claims 1 to 7.