Wind power climbing event prediction method and system based on depth state space model
Through the wind power climbing event prediction method based on the deep state space model, combined with convolutional neural network and Kalman filtering, the accuracy and robustness of wind power climbing event prediction at the wind farm station are solved, and more efficient Capture and forecast wind power changes in wind farms.
Patent Information
- Application Number
- CN202411270131.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prediction of wind power climbing events, the accuracy rate is difficult to improve and lacks robustness. Especially when dealing with long-term correlation and extreme weather effects, the prediction effect is poor.
The wind power climbing event prediction method based on the deep state space model is adopted. By constructing a deep state space model with attention mechanism, combining convolutional neural network and Kalman filtering, end-to-end iterative training is carried out to extract the historical wind of the wind farm. Power and meteorological information are learned from the perspective of time and space characteristics to predict wind power climbing data.
It improves the prediction accuracy and robustness of wind power climbing events, can more accurately capture the complex changes in wind farms, reduce the impact of noise, and enhances the robustness and prediction performance of the model.
Smart Images

Figure CN119026751A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of wind power generation, and in particular to a method and system for predicting wind power ramping events based on a deep state space model. Background Art
[0002] Wind power ramping events refer to large changes in wind power in a short period of time. When the power change amplitude and change rate exceed the adjustment capacity of the control strategy in the power system, it will cause power imbalance. This process is called a ramping event. Wind power ramping events are often caused by extreme weather such as thunderstorms, gusts, and strong low-pressure systems; photovoltaic ramping events are mainly caused by factors such as day and night changes and sudden changes in meteorological conditions in a short period of time. After the ramping event occurs, it is easy to cause a short-term imbalance in the power of the power system, threatening the power quality of the power grid and the safe and stable operation of the system. Specifically, when there is a strong low-pressure system, low-altitude jet stream, thunderstorm, gusts or similar long-term extreme meteorological events, it will cause a sudden increase in power in the wind farm for a period of time, and an up-ramp event will occur. When a relative meteorological event occurs or the wind speed of the wind turbine is higher than the cut-out wind speed, some wind turbines will be withdrawn from operation one after another, and the power of the wind farm will drop sharply for a period of time, and a down-ramp event will occur.
[0003] Methods for predicting wind power ramping events include statistical models, physical models, and comprehensive models. Traditional prediction of ramping events consists of two parts: wind power prediction and ramping detection. Wind power prediction uses data mining algorithms to train models that represent the relationship between wind power generation and other variables. Usually, the Markov autoregression model and the historical residuals of wind power can be used to perform residual prediction to obtain the predicted value of the residual. Ramp detection uses the revolving door algorithm to extract linear trend segments from the wind power residual prediction value and predict the wind power ramping time. The revolving door algorithm is a trend data compression algorithm that compresses the detailed information of the event sequence and retains only the key information to identify the time-varying process with certain characteristics.
[0004] However, the Markov assumption may oversimplify the complexity of wind power changes. For wind power time series with long-term correlation or significant impacts of seasonal factors and extreme weather events, it cannot accurately capture the trend of changes, affecting the actual prediction effect. The revolving door algorithm has the advantages of small computational complexity and the ability to track process trend changes. However, when there is noise in wind power data, the compression performance of the revolving door algorithm may not be ideal. Noise may cause the algorithm to misjudge reality, thereby reducing the compression ratio. In addition, the revolving door algorithm relies on preset values. If the preset values do not match the actual wind power data characteristics, the training effect may be affected. Summary of the invention
[0005] In view of the deficiencies in the prior art, the present application provides a method and system for predicting wind power ramp events based on a deep state space model, which solves the problem that the accuracy of wind power ramp event prediction is difficult to further improve and lacks robustness.
[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:
[0007] In a first aspect, an embodiment of the present application provides a method for predicting wind power ramping events based on a deep state space model, the method comprising: dividing the determined target data information into a training set and a test set according to a preset ratio; wherein the target data information is used to characterize the historical wind power information and meteorological information of the wind farm; constructing a deep state space model based on an attention mechanism, wherein the deep state space model includes a state transfer equation and an observation model; inputting the training set into the deep state space model, and performing iterative training in an end-to-end manner based on the pytorch framework, and when the deep state space model converges, determining the deep state space model as the first target model; testing the first target model based on the test set to determine the second target model; predicting wind power ramping data through the second target model, analyzing the predicted wind power ramping data according to a preset extreme value method, and determining whether a ramping event occurs.
[0008] According to the first aspect of the embodiment of the present application, before the determined target data information is divided into a training set and a test set according to a preset ratio, the wind power ramp event prediction method based on the deep state space model may also include the following steps: obtaining initial data information of the wind farm station, the initial data information including historical wind power data and historical meteorological data; performing outlier processing and normalization processing on the historical wind power data in the initial data information to obtain first intermediate data information; performing One-Hot processing on the historical meteorological data to obtain second intermediate data information; and determining the first intermediate data information and the second intermediate data information as target data information.
[0009] According to a first aspect of the embodiment of the present application, the deep state space model satisfies the expression:
[0010] y′ t+h =G(X t-L+1 , X t-L+2 , X t-L+3 , ...X t-1 , e t-L+1 , e t-L+2 , e t-L+3 ,...e t-1 )
[0011] In the formula, y′ t+his the wind power forecast value at time t+h, h represents the forecast interval, G is the formula expression of the deep state space model, X t-L+i represents the wind power at the historical moment, e t-L+i It represents the external eigenvector at a certain moment, 1≤i≤L-1, and L represents the length of recent historical data used for wind power prediction.
[0012] According to a first aspect of the embodiment of the present application, in a deep state space model, the observation model O cnn Determined based on convolutional neural network and satisfies the expression:
[0013] O cnn =g cnn (X1, X2, ..., X T )
[0014] In the formula, g cnn (·) uses layer normalization and ReLU activation function, and is used to represent two combinations of convolution, regularization, and activation functions. d c express The dimension of X is 1≤t≤T. t Represents the wind power matrix at a historical moment.
[0015] According to a first aspect of an embodiment of the present application, the state transfer equation is determined based on Kalman filtering, the Kalman filtering includes two fully connected layers, and the process of determining the observation value and the observation vector through Kalman filtering satisfies the expression:
[0016]
[0017] In the formula, z t is the observed value, l t is the observation vector, is O at time t cnn , e t is the external eigenvector corresponding to time t, d k Indicates z t and l t The dimension of and Represents the learning parameters of the fully connected layer.
[0018] According to a first aspect of the embodiment of the present application, the output value of the Kalman filter is Satisfies the expression:
[0019]
[0020] In the formula, is the output value of the Kalman filter at time t, w t and bt They represent learnable parameters respectively;
[0021] Observe the output of the model Satisfies the expression:
[0022]
[0023] In the formula, w ar and b ar Represent the weight and bias of the observation model respectively;
[0024] The wind power ramp data y′ predicted by the second target model t+h Satisfies the expression:
[0025] y′ t+h =Relu(w T O t +b)
[0026] In the formula,
[0027] Indicates that the dimension at time t is d e The external feature vector of , w and b are both learnable parameters.
[0028] According to a first aspect of an embodiment of the present application, wind power climbing data is used to characterize a climbing amplitude and a climbing rate; the aforementioned analysis of the predicted wind power climbing data according to a preset extreme value method to determine whether a climbing event occurs includes the following steps:
[0029] Calculate the predicted wind power ramp data and determine the target change range of the wind power ramp data within 4 hours;
[0030] When the target change amplitude reaches at least 50% of the installed capacity, it is determined that a ramp event occurs at the wind farm.
[0031] In the second aspect, an embodiment of the present application provides a wind power climbing event prediction system based on a deep state space model, the system comprising: a division module, a construction module, a training module, a testing module and a prediction module; the division module is used to divide the determined target data information into a training set and a test set according to a preset ratio; wherein the target data information is used to characterize the historical wind power information and meteorological information of the wind farm; the construction module is used to construct a deep state space model based on an attention mechanism, wherein the deep state space model includes a state transfer equation and an observation model; the training module is used to input the training set into the deep state space model, based on the pytorch framework, and adopts an end-to-end iterative training method, and when the deep state space model converges, determines that the deep state space model is the first target model; the testing module is used to test the first target model based on the test set to determine the second target model; the prediction module is used to predict the wind power climbing data through the second target model, and analyzes the predicted wind power climbing data according to the preset extreme value method to determine whether a climbing event occurs.
[0032] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for predicting wind power ramping events based on a deep state space model in the first aspect is implemented.
[0033] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a program or instruction, which, when executed by a processor, implements the wind power ramp event prediction method based on a deep state space model in the aforementioned first aspect.
[0034] The present application provides a method and system for predicting wind power ramp events based on a deep state space model. Compared with the prior art, it has the following beneficial effects:
[0035] In order to predict wind power ramping, the present application obtains historical target data information of wind farms and divides the data into training sets and test sets. The target data information can characterize the historical wind power information and meteorological information of wind farms. Learning and training from the perspective of time and space characteristics can effectively extract feature information and improve the prediction accuracy. By constructing a deep state space model, an end-to-end model training is performed based on the training set until the model converges. Then, the first target model obtained through training is tested based on the test set to determine a mature second target model. The present application predicts wind power ramping data through the second target model. Based on the wind power ramping data, the extreme value method can be used to determine whether it corresponds to a ramping event, thereby realizing the prediction of wind power ramping events. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0037] Figure 1 It is a flow chart of a method for predicting wind power ramping events based on a deep state space model provided in an embodiment of the present application;
[0038] Figure 2 It is a structural schematic diagram of a wind power ramp event prediction system based on a deep state space model provided in an embodiment of the present application;
[0039] Figure 3 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0042] The embodiments of the present application provide a method and system for predicting wind power ramping events based on a deep state space model, thereby solving the problem that the accuracy of wind power ramping event prediction is difficult to further improve and lacks robustness.
[0043] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:
[0044] A wind power ramping event refers to a large change in wind power in a short period of time. When the power change amplitude and change rate exceed the regulation capacity of the control strategy in the power system, it will cause a power imbalance. This process is called a ramping event. Wind power ramping events are often caused by extreme weather factors such as thunderstorms, gusty winds, and strong low-voltage systems; photovoltaic ramping events are mainly caused by factors such as day and night changes and sudden changes in meteorological conditions in a short period of time. After a ramping event occurs, it is easy to cause a short-term imbalance in the power system power, threatening the power quality of the power grid and the safe and stable operation of the system.
[0045] Specifically, when there is a strong low-pressure system, low-level jet stream, thunderstorm, gusty wind or similar long-term extreme weather events, the wind farm power will increase suddenly for a period of time, and an up-climbing event will occur. When a relative weather event occurs or the wind speed of the wind turbine is higher than the cut-out wind speed, some wind turbines will be out of operation one after another, and the wind farm power will drop sharply for a period of time, and a down-climbing event will occur. The climbing event is a complex nonlinear problem. Under the premise of considering the climbing characteristics, control requirements, operating conditions and equipment constraints, how to achieve the optimal control effect still needs in-depth research.
[0046] In terms of ramp prediction, there are currently a variety of prediction methods. The existing ramp event prediction methods can be divided into direct methods and indirect methods based on the prediction steps. The direct method uses meteorological data to directly predict ramp events, while the indirect method requires the use of power prediction results on the basis of power prediction to achieve the final ramp prediction.
[0047] At present, the methods for predicting wind power ramping events include statistical models, physical models and comprehensive models. Traditional prediction of ramping events consists of two parts: wind power prediction and ramping detection. Wind power prediction uses data mining algorithms to train models that represent the relationship between wind power generation and other variables. These models include autoregressive moving average models, neural networks, support vector machines, Markov autoregressive models, etc. For example, Markov autoregression is a statistical model that uses the Markov autoregressive model and the historical residuals of wind power to perform residual prediction and obtain the predicted value of the residual. Ramp detection uses the revolving door algorithm to extract linear trend segments from the wind power residual prediction value and predict the wind power ramping time. The revolving door algorithm is a trend data compression algorithm that compresses the detailed information of the event sequence and retains only the key information to identify the time-varying process with certain characteristics.
[0048] However, although the Markov autoregressive model has the advantages of simplifying the model complexity and being easy to implement when predicting wind power, the Markov assumption may oversimplify the complexity of wind power changes. For wind power time series with long-term correlation or significant impacts of seasonal factors and extreme weather events, it cannot accurately capture the trend of changes, affecting the actual prediction effect. The revolving door algorithm has the advantages of small computational complexity and the ability to track process trend changes, but when there is noise in wind power data, the compression performance of the revolving door algorithm may not be ideal. Noise may cause the algorithm to misjudge reality, thereby reducing the compression ratio. In addition, the revolving door algorithm relies on preset values. If the preset values do not match the characteristics of the actual wind power data, the training effect may be affected.
[0049] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0050] The following first introduces a wind power ramp event prediction method based on a deep state space model provided in an embodiment of the present application.
[0051] A flow chart of a method for predicting wind power ramp events based on a deep state space model provided in an embodiment of the present application is as follows: Figure 1 As shown, the method may include the following steps S110-S150.
[0052] S110, dividing the determined target data information into a training set and a test set according to a preset ratio; wherein the target data information is used to characterize the historical wind power information and meteorological information of the wind farm.
[0053] S120. Construct a deep state space model based on the attention mechanism, wherein the deep state space model includes a state transfer equation and an observation model.
[0054] S130, inputting the training set into the deep state space model, performing iterative training in an end-to-end manner based on the pytorch framework, and determining the deep state space model as the first target model when the deep state space model converges.
[0055] S140: Test the first target model based on the test set to determine the second target model.
[0056] S150, predicting wind power ramp data through the second target model, analyzing the predicted wind power ramp data according to a preset extreme value method, and determining whether a ramp event occurs.
[0057] The above is a specific implementation method of the wind power ramp event prediction method based on the deep state space model provided in this application. In order to perform wind power ramp prediction, this application obtains the historical target data information of the wind farm station and divides the training set and the test set. The target data information can characterize the historical wind power information and meteorological information of the wind farm station. It can be understood that the wind power information corresponds to spatial characteristics, and the meteorological information corresponds to time series characteristics. This application learns and trains from the perspective of time and space characteristics, which can effectively refine the feature information and improve the prediction accuracy.
[0058] Specifically, the present application constructs a deep state space model and performs model training in an end-to-end manner based on a training set until the model converges; then the trained first target model is tested based on a test set to determine a mature second target model; the present application predicts wind power climbing data through the second target model, and based on the wind power climbing data, it can determine whether it corresponds to a climbing event through the extreme value method, thereby realizing the prediction of wind power climbing events.
[0059] In one example, meteorological data is used to reflect information such as temperature, humidity, pressure, 10m wind speed, 80m wind speed, whether it is snowing, whether it is raining, whether it is hail, whether it is freezing rain, etc. of a wind farm station.
[0060] In some embodiments, before the determined target data information is divided into a training set and a test set according to a preset ratio, that is, before the aforementioned S110, the wind power ramp event prediction method based on the deep state space model may further include the following steps:
[0061] S210, obtaining initial data information of the wind farm station, the initial data information including historical wind power data and historical meteorological data;
[0062] S220, performing outlier processing and normalization processing on the historical wind power data in the initial data information to obtain first intermediate data information;
[0063] S230, performing One-Hot processing on the historical meteorological data to obtain second intermediate data information;
[0064] S240: Determine that the first intermediate data information and the second intermediate data information are target data information.
[0065] In the embodiments of the present application, it is understandable that since abnormal data may exist in the acquired initial data information, in the process of model training based on the acquired initial data information, the initial data information needs to be preprocessed to ensure the reliability of the data involved in the model training prediction.
[0066] In one example, the deep state-space model satisfies the expression:
[0067] y′ t+h =G(X t-L+1 , X t-L+2 , X t-L+3 , ...X t-1 , e t-L+1 , e t-L+2 , e t-L+3 ,...e t-1 )
[0068] In the formula, y′ t+h is the wind power forecast value at time t+h, h represents the forecast interval, G is the formula expression of the deep state space model, X t-L+i represents the wind power at the historical moment, e t-L+i It represents the external eigenvector at a certain moment, 1≤i≤L-1, and L represents the length of recent historical data used for wind power prediction.
[0069] In one example, in a deep state-space model, the observation model O cnn Determined based on convolutional neural network and satisfies the expression:
[0070] O cnn =g cnn (X1, X2, ..., X T )
[0071] In the formula, g cnn (·) uses layer normalization and ReLU activation function, and is used to represent two combinations of convolution, regularization, and activation functions. d c express The dimension of X is 1≤t≤T. t Represents the wind power matrix at a historical moment.
[0072] In the embodiments of the present application, it can be understood that the convolutional neural network performs well in feature extraction of time series data, and can effectively capture the spatial structure of the input data and the event structure of the time series data, and can better train features.
[0073] In some embodiments, the state transfer equation is determined based on Kalman filtering, which includes two fully connected layers. The process of determining the observation value and the observation vector through Kalman filtering satisfies the expression:
[0074]
[0075] In the formula, z t is the observed value, l t is the observation vector, is O at time t cnn , et is the external eigenvector corresponding to time t, d k Indicates z t and l t The dimension of and Represents the learning parameters of the fully connected layer.
[0076] In the embodiment of the present application, it can be understood that Kalman filtering is an algorithm that uses a linear system state equation to optimally estimate the system state through system input and output observation data. Since the observation data includes the influence of noise and interference in the system, the optimal estimation can also be regarded as a filtering process.
[0077] Data filtering is a data processing technology that removes noise and restores real data. Kalman filtering can estimate the state of a dynamic system from a series of data with measurement noise when the measurement variance is known. Since Kalman filtering is easy to implement through computer programming and can update and process data collected on-site in real time, Kalman filtering is currently the most widely used filtering method and has been well applied in many fields such as communication, navigation, guidance and control.
[0078] It should be noted that the present application uses Kalman filtering to estimate the state at future moments, which is inherited into the aforementioned end-to-end iterative training and trained together with the convolutional neural network. During the iterative process, the loss function uses root mean square error and L1 regularization, which can better handle model uncertainty and external disturbances, thereby improving the robustness of the model.
[0079] In one example, the output of the Kalman filter is Satisfies the expression:
[0080]
[0081] In the formula, is the output value of the Kalman filter at time t, w t and b t They represent learnable parameters respectively;
[0082] Observe the output of the model Satisfies the expression:
[0083]
[0084] In the formula, w ar and b ar Represent the weight and bias of the observation model respectively;
[0085] The wind power ramp data y′ predicted by the second target model t+h Satisfies the expression:
[0086] y′ t+h =Relu(w T O t +b)
[0087] In the formula,
[0088] Indicates that the dimension at time t is d e The external feature vector of , w and b are both learnable parameters.
[0089] In some embodiments, the aforementioned wind power ramp data is used to characterize the ramp amplitude and ramp rate; the aforementioned analysis of the predicted wind power ramp data according to the preset extreme value method to determine whether a ramp event occurs includes the following steps:
[0090] S310, statistically predicting wind power ramp data, and determining a target change range of the wind power ramp data within 4 hours.
[0091] S320: When the target change amplitude reaches at least 50% of the installed capacity, it is determined that a ramp event occurs at the wind farm.
[0092] In the embodiments of the present application, it can be understood that the present application converts the prediction of wind power climbing events into a prediction of wind power, that is, converts the prediction of a discrete data into a regression model, thereby reducing the error caused by judging the climbing state of the original data.
[0093] In some embodiments, in Kalman filtering, it is assumed that the system noise and the observation noise both have mean 0 and variances Q and R respectively. t Gaussian random variables. Kalman filtering includes linear Kalman filtering and nonlinear Kalman filtering.
[0094] The linear Kalman filter calculates the initial estimated state to satisfy the expression:
[0095] X t|t-1 =FX t-1|t-1
[0096] P t|t-1 =FP t-1|t-1 F T +Q
[0097] Where F is the transfer matrix obtained by learning the convolutional neural network, (X t-1 , P t-1 ) represents the optimal estimation state at time t-1, Q is the process noise covariance matrix of the Kalman filter, and the λ vector is the diagonal element, Q=diag(λ), diag represents the diagonal element of the matrix.
[0098] In one example, the optimal estimate state satisfies the expression:
[0099] K t =P t|t-1 Z T [ZP t|t-1 Z T +R t ] -1
[0100] X t =X t|t-1 +K t ×v t
[0101] P t =(I-Kt Z ) t|t-1
[0102] In the formula, K t represents the Kalman gain at time t, I represents the identity matrix, Z represents the measurement matrix, P t is the estimated state covariance matrix, v t is the observation noise.
[0103] In some embodiments, for nonlinear Kalman filtering, the present application locally linearizes the nonlinear model and extends the initial estimation state and the optimal estimation state of the Kalman filter; wherein the initial estimation state of the extended Kalman filter satisfies the expression:
[0104] X t|t-1 =f(X t-1|t-1 )
[0105]
[0106] Where f(·) is the nonlinear state transfer function, A t|t-1 It is in X t-1|t-1 The Jacobian matrix of the state at .
[0107] In some embodiments, Figure 2 As shown, the present application provides a wind power ramp event prediction system 400 based on a deep state space model. The system 400 may specifically include the following modules:
[0108] A division module 410 is used to divide the determined target data information into a training set and a test set according to a preset ratio; wherein the target data information is used to characterize the historical wind power information and meteorological information of the wind farm station;
[0109] A construction module 420 is used to construct a deep state space model based on the attention mechanism, wherein the deep state space model includes a state transfer equation and an observation model;
[0110] A training module 430 is used to input the training set into the deep state space model, perform iterative training in an end-to-end manner based on the pytorch framework, and determine the deep state space model as the first target model when the deep state space model converges;
[0111] A testing module 440, configured to test the first target model based on the test set to determine the second target model;
[0112] The prediction module 450 is used to predict the wind power ramp data through the second target model, analyze the predicted wind power ramp data according to a preset extreme value method, and determine whether a ramp event occurs.
[0113] According to an embodiment of the present application, any multiple modules among the partition module 410, the construction module 420, the training module 430, the test module 440 and the prediction module 450 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module.
[0114] In some embodiments, the system 400 may further include a preprocessing model 460, which may be specifically used to:
[0115] Obtaining initial data information of the wind farm station, the initial data information includes historical wind power data and historical meteorological data;
[0116] Performing outlier processing and normalization processing on the historical wind power data in the initial data information to obtain first intermediate data information;
[0117] Perform One-Hot processing on historical meteorological data to obtain second intermediate data information;
[0118] The first intermediate data information and the second intermediate data information are determined to be target data information.
[0119] In some embodiments, the prediction module 450 may be specifically used to:
[0120] Calculate the predicted wind power ramp data and determine the target change range of the wind power ramp data within 4 hours;
[0121] When the target change amplitude reaches at least 50% of the installed capacity, it is determined that a ramp event occurs at the wind farm.
[0122] Figure 2Each module in the system shown has the function of implementing each step in the aforementioned wind power ramp event prediction method based on the deep state space model and can achieve its corresponding technical effect. For the sake of concise description, it will not be repeated here.
[0123] In some embodiments, the present application provides an electronic device, the structure diagram of the electronic device is as follows Figure 3 shown.
[0124] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.
[0125] Specifically, the processor 510 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.
[0126] The memory 520 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 520 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 520 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 520 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 520 is a non-volatile solid-state memory.
[0127] The memory 520 may include a read-only memory (ROM), a random access memory (RAM), a disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory 520 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer executable instructions, and when the software is executed (e.g., by one or more processors), it can perform the operations described in any one of the wind power ramp event prediction methods based on the deep state space model in the above-mentioned embodiments.
[0128] The processor 510 implements any one of the wind power ramp event prediction methods based on the deep state space model in the above embodiments by reading and executing the computer program instructions stored in the memory 520 .
[0129] In one example, the electronic device may further include a communication interface 530 and a bus 500. Figure 3As shown, the processor 510, the memory 520, and the communication interface 530 are connected via a bus 500 and communicate with each other.
[0130] The communication interface 530 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.
[0131] Bus 500 includes hardware, software or both, and the parts of online data flow billing equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-end bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 500 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the present application considers any suitable bus or interconnection.
[0132] In addition, in combination with the wind power ramp event prediction method based on the deep state space model in the above embodiment, the embodiment of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the wind power ramp event prediction methods based on the deep state space model in the above embodiment is implemented.
[0133] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.
[0134] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.
[0135] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.
[0136] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram 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, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.
[0137] In summary, compared with the prior art, this application has the following beneficial effects:
[0138] 1. This application constructs a deep state space model and adopts an end-to-end approach to perform model training until the model converges; the wind power climbing data is predicted by a determined second target model, and based on the wind power climbing data, it can be determined whether it corresponds to a climbing event through the extreme value method, thereby realizing the prediction of wind power climbing events, and converting wind power climbing events into wind power predictions, that is, converting a discrete data prediction into a regression model, thereby reducing the error caused by judging the climbing state of the original data.
[0139] 2. This application obtains target data information of wind farms. The target data information can characterize the historical wind power information and meteorological information of wind farms. Learning and training from the perspective of time and space characteristics can effectively refine feature information and improve prediction accuracy.
[0140] 3. The deep state space model constructed in this application utilizes the powerful feature extraction and multi-scale information processing capabilities of convolutional neural networks to effectively capture the characteristics of meteorological data, and adopts Kalman filtering to extract the long-term time domain characteristics of wind power, effectively integrating the above information and improving the prediction performance of the model.
[0141] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. 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 embodiments of the present invention.
Claims
1. A method for predicting wind power ramp events based on a deep state space model, characterized in that: include: Dividing the determined target data information into a training set and a test set according to a preset ratio; wherein the target data information is used to characterize the historical wind power information and meteorological information of the wind farm station; Constructing a deep state space model based on the attention mechanism, wherein the deep state space model includes a state transfer equation and an observation model; Input the training set into the deep state space model, perform iterative training in an end-to-end manner based on the pytorch framework, and determine the deep state space model as the first target model when the deep state space model converges; Testing the first target model based on the test set to determine a second target model; The wind power ramp data is predicted by the second target model, and the predicted wind power ramp data is analyzed according to a preset extreme value method to determine whether a ramp event occurs.
2. The method for predicting wind power ramp events based on a deep state space model according to claim 1, characterized in that: Before dividing the determined target data information into a training set and a test set according to a preset ratio, the wind power ramp event prediction method based on the deep state space model further includes: Acquiring initial data information of a wind farm, wherein the initial data information includes historical wind power data and historical meteorological data; Performing outlier processing and normalization processing on the historical wind power data in the initial data information to obtain first intermediate data information; One-hot processing is performed on the historical meteorological data to obtain second intermediate data information; The first intermediate data information and the second intermediate data information are determined to be target data information.
3. The method for predicting wind power ramp events based on a deep state space model according to claim 1 or 2, characterized in that: The deep state space model satisfies the expression: y′ t+h =G(X t-L+1 ,X t-L+2 ,X t-L+3 ,...X t-1 ,e t-L+1 ,e t-L+2 ,e t-L+3 ,...e t-1 ) In the formula, y′ t+h is the wind power forecast value at time t+h, h represents the forecast interval, G is the formulaic expression of the deep state space model, X t-L+i represents the wind power at the historical moment, e t-L+i It represents the external eigenvector at a certain moment, 1≤i≤L-1, and L represents the length of recent historical data used for wind power prediction.
4. The method for predicting wind power ramp events based on a deep state space model according to claim 1 or 2, characterized in that: In the deep state space model, the observation model O cnn Determined based on convolutional neural network and satisfies the expression: O cnn =g cnn (X1,X2,...,X T ) In the formula, g cnn (·) uses layer normalization and ReLU activation function, and is used to represent two combinations of convolution, regularization, and activation functions. d c express The dimension of X is 1≤t≤T. t Represents the wind power matrix at a historical moment.
5. The method for predicting wind power ramp events based on a deep state space model according to claim 4, characterized in that: The state transfer equation is determined based on Kalman filtering, and the Kalman filtering includes two fully connected layers. The process of determining the observation value and the observation vector through the Kalman filtering satisfies the expression: In the formula, z t is the observed value, l t is the observation vector, is O at time t cnn , e t is the external eigenvector corresponding to time t, d k Indicates z t and l t The dimension of and Represents the learning parameters of the fully connected layer.
6. The method for predicting wind power ramp events based on a deep state space model according to claim 5, characterized in that: The output value of the Kalman filter Satisfies the expression: In the formula, is the output value of the Kalman filter at time t, w t and b t They represent learnable parameters respectively; The output value of the observation model Satisfies the expression: In the formula, w ar and b ar Respectively represent the weight and bias of the observation model; The wind power ramp data y′ predicted by the second target model t+h Satisfies the expression: y′ t+h =Relu(in T ABOUT t +b) In the formula, Indicates that the dimension at time t is d e The external feature vector of , w and b are both learnable parameters.
7. The method for predicting wind power ramp events based on a deep state space model according to claim 1 or 2, characterized in that: The wind power ramp data is used to characterize the ramp amplitude and ramp rate; the predicted wind power ramp data is analyzed according to a preset extreme value method to determine whether a ramp event occurs, including: Statistically predict the wind power ramp data to determine a target change range of the wind power ramp data within 4 hours; When the target variation amplitude reaches at least 50% of the installed capacity, it is determined that a ramp event occurs in the wind farm.
8. A wind power ramp event prediction system based on a deep state space model, characterized in that: include: A division module, used to divide the determined target data information into a training set and a test set according to a preset ratio; wherein the target data information is used to characterize the historical wind power information and meteorological information of the wind farm station; A construction module, used to construct a deep state space model based on the attention mechanism, wherein the deep state space model includes a state transfer equation and an observation model; A training module, used for inputting the training set into the deep state space model, performing iterative training in an end-to-end manner based on a pytorch framework, and determining the deep state space model as a first target model when the deep state space model converges; A testing module, configured to test the first target model based on the test set to determine a second target model; A prediction module is used to predict wind power climbing data through the second target model, analyze the predicted wind power climbing data according to a preset extreme value method, and determine whether a climbing event occurs.
9. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for predicting wind power ramping events based on a deep state space model as described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the method for predicting wind power ramping events based on a deep state space model as described in any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Wind power climbing event early warning method considering field-network factor
CN110348637A
Network traffic prediction method based on depth state space model
CN114372561A
Wind power plant climbing event prediction method based on data enhancement
CN117117968A
Wind power short-term power climbing prediction method
CN117578404A
Power load probability prediction method and system, medium, equipment and program product
CN117977584A
Cited By
Photovoltaic climbing event identification method and system based on improved SDA and multi-dimensional coupling discrimination
CN121663623A
A photovoltaic climbing event recognition method and system based on improved SDA and multi-dimensional coupling discrimination
CN121663623B
Wind power prediction method and system based on event-driven variable state space
CN122436965A
