CN112529275A - Wind power climbing event prediction method based on feature extraction and deep learning - Google Patents

Wind power climbing event prediction method based on feature extraction and deep learning Download PDF

Info

Publication number
CN112529275A
CN112529275A CN202011398774.6A CN202011398774A CN112529275A CN 112529275 A CN112529275 A CN 112529275A CN 202011398774 A CN202011398774 A CN 202011398774A CN 112529275 A CN112529275 A CN 112529275A
Authority
CN
China
Prior art keywords
climbing
wind power
prediction
event
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011398774.6A
Other languages
Chinese (zh)
Other versions
CN112529275B (en
Inventor
韩丽
乔妍
王晓静
李梦洁
鲁盼盼
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Mining and Technology CUMT
Original Assignee
China University of Mining and Technology CUMT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China University of Mining and Technology CUMT filed Critical China University of Mining and Technology CUMT
Priority to CN202011398774.6A priority Critical patent/CN112529275B/en
Publication of CN112529275A publication Critical patent/CN112529275A/en
Application granted granted Critical
Publication of CN112529275B publication Critical patent/CN112529275B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Marketing (AREA)
  • General Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Development Economics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a wind power climbing event prediction method based on feature extraction and deep learning, which comprises the steps of firstly, utilizing an improved revolving door (OpSDA) algorithm to perform climbing recognition on historical wind power to obtain a historical climbing feature value; then inputting the historical characteristic value and the historical power as prediction models, outputting the prediction power, establishing a CNN-LSTM wind power prediction model, and excavating a coupling relation between wind power climbing characteristics and wind power through deep learning, wherein the CNN is responsible for secondary extraction of data characteristics, and the LSTM is responsible for predicting the wind power; and finally, carrying out climbing identification to obtain a climbing event prediction result. The deep learning network is adopted to effectively extract and learn the climbing characteristics in the wind power, and a more accurate climbing prediction result can be obtained.

Description

Wind power climbing event prediction method based on feature extraction and deep learning
Technical Field
The invention relates to a wind power climbing event prediction method based on feature extraction and deep learning, and belongs to the field of wind power uncertainty analysis and new energy prediction.
Background
Wind energy is widely applied to China as renewable clean energy, and according to statistics released by the State energy agency, the newly increased grid-connected installed capacity of the wind power in 2019 in the whole nation is 2574 ten thousand kW, and the accumulated installed capacity is 2.1 hundred million kW. And as the wind power grid-connected scale is increased, the influence of uncertainty of wind power on safe and stable operation of the power system is larger. Especially, the wind power climbing event can cause the problems of unbalanced power generation and supply of the power grid and the like, and great potential safety hazard or serious economic loss is brought to the operation of the power grid. Therefore, for safe operation and economic dispatch of the power system, it is important to obtain an accurate climbing event prediction result.
The climbing prediction is different from the wind power prediction in that the climbing is a small-probability event in a wind power sequence, the event is used as a prediction object in the climbing prediction, and the prediction method comprises a direct prediction method and an indirect prediction method. The direct prediction needs to be carried out through a large number of historical climbing data training models, learners model the multi-attribute combined statistical characteristics of climbing, a day-ahead climbing event sequence prediction algorithm is provided, the relevance of the specific time sequence characteristics of wind power and climbing events is ignored, and the completeness of historical data influences the prediction accuracy of the statistical model. The current research based on wind power climbing event prediction mainly adopts an indirect prediction method, and climbing recognition is carried out on the basis of wind power prediction to obtain a climbing prediction result. The difficulty of analyzing the wind power output characteristics can be increased due to a climbing event, and the prediction accuracy is improved by taking the relevant characteristic data of wind power as the input of a model in most of the existing researches, for example, long-term trend characteristics of wind power signals are captured through a physical model before prediction. On the basis of considering the spatial characteristics of wind power, a learner adopts a method combining a deep neural network and multi-task learning, and simultaneously receives input from a plurality of wind power plants for prediction based on the spatial correlation of the wind power plants. Still, the scholars propose to extract features through a Convolutional Neural Network (CNN), but the features of a climbing event are rarely considered in the existing research aiming at climbing prediction, and the climbing feature extraction process does not pass through strict climbing definition, but directly utilizes the CNN to extract the sudden change features in the wind power time sequence as the climbing features, and does not combine the characteristics of the wind power signal in a prediction model.
The conventional wind power climbing event prediction theoretical system is still in the improvement process, and the conventional prediction method mainly has the problem that the input part of a prediction model does not fully consider the climbing characteristics of wind power, but directly inputs the wind power and other influence factors into the model. However, the climbing event is used as a small sample event in the wind power sequence, the characteristic information of the climbing event is easily submerged in the wind power signal, the characteristic information is complex, and if the climbing characteristic information is not extracted in advance as input, the model is difficult to learn the complete climbing characteristic condition, so that the wind power climbing event cannot be effectively predicted.
Disclosure of Invention
The purpose of the invention is as follows: in order to overcome the defects of the prior art, the invention provides the wind power climbing event prediction method based on feature extraction and deep learning, climbing is identified and features are extracted through an improved revolving door algorithm (OpSDA), and the climbing features are secondarily extracted by using a CNN-LSTM model and then predicted, so that the precision of climbing event prediction can be effectively improved.
The technical scheme is as follows: the wind power climbing event prediction method based on feature extraction and deep learning comprises the following steps:
step 1: climbing recognition is carried out on historical wind power by adopting an OpSDA algorithm, and 4 climbing characteristic values are extracted: climbing rate RRClimbing amplitude RSWStart time RSTAnd duration RD；
Step 2: dividing input data consisting of wind power and climbing characteristic values by using a sliding window, inputting the divided input data into a CNN-LSTM prediction model, extracting climbing characteristics again by the CNN, predicting by the LSTM to obtain a plurality of prediction results which are advanced by 16 steps, and obtaining a final rolling multistep wind power prediction result by extracting a first prediction point of multistep prediction;
and step 3: and carrying out climbing identification on the predicted power to obtain a prediction result of the climbing event.
Has the advantages that: according to the wind power climbing event prediction method based on feature extraction and deep learning, the OpSDA algorithm is used for recognizing wind power climbing extraction climbing features and inputting the climbing features into the prediction model, so that on one hand, the prediction model is favorable for learning the relation between wind power and climbing events in the training process and improving the prediction precision, on the other hand, the safe, stable and economic operation of wind power integration is favorable, the standby allowance during power grid scheduling can be reduced, and the impact of wind power climbing on power generation and supply balance of a power grid can be reduced. Secondly, as the wind power climbing event is an emergency, the climbing feature data has non-continuity for continuous wind power sequence data. The LSTM network model can reflect the long-term time sequence correlation problem of information, but cannot dig out the coupling relation between discontinuous climbing characteristics. And the convolution kernel of the CNN network can move on the original characteristic diagram to carry out convolution operation, so that the potential relation of each data in the characteristic diagram can be effectively extracted to form a characteristic vector, and the climbing characteristics in the whole data set can be extracted layer by layer from a deep structure from a small region through deep learning. And finally, the LSTM network can learn the relation of data in time sequence, and the wind power and the climbing characteristics extracted by the CNN network are jointly used as the input of the LSTM network, so that the model can learn the relation between the sudden climbing event and the wind power time sequence. Important climbing features are reserved through a memory cell group (cell) specific to the LSTM, unimportant features are forgotten by a forgetting gate, and the learning of the climbing features by the network is enhanced.
Drawings
FIG. 1 is a wind power climbing event feature diagram;
FIG. 2 is a schematic diagram of a turnstile algorithm;
FIG. 3 is a flow chart of OpSDA identifying climbing;
FIG. 4 is a diagram illustrating the result of the OpSDA hill climbing recognition;
FIG. 5 is a diagram of the structure of the CNN-LSTM model;
FIG. 6 is a comparison of prediction model performance evaluation indicators under different parameters;
fig. 7 is a diagram illustrating a result of the hill climbing prediction.
Detailed Description
The invention is further explained below with reference to the drawings.
The wind power climbing event prediction method based on feature extraction and deep learning comprises the following steps:
step 1: climbing recognition is carried out on historical wind power by adopting an OpSDA algorithm, and 4 climbing characteristic values are extracted: climbing rate RRClimbing amplitude RSWStart time RSTAnd duration RDAs shown in fig. 1. The step 1 comprises the following steps:
the wind power climbing event belongs to a small probability event for the whole wind power time sequence, so when the wind power is predicted by utilizing a deterministic model, the influence of climbing characteristics on prediction needs to be fully considered. The four important characteristics of the climbing event are important indexes for measuring whether climbing occurs or not and measuring the severity of the climbing, the greater the climbing rate is when the climbing event occurs, the stronger the nonlinearity of the wind power is, and the prediction error is larger compared with that of a non-climbing section. Therefore, the influence of the climbing characteristics is considered in the wind power prediction process, the OpSDA algorithm is adopted to identify the climbing and extract the characteristics as prediction input, and therefore the prediction accuracy is improved.
Step 1.1: and (3) constructing a parallelogram according to the adjustable parameter gate width epsilon by using a revolving door algorithm (SDA) to screen the sample data to obtain SDA segmentation points, as shown in figure 2.
Step 1.2: and (3) constructing an objective function P (i, j) on any section of wind power time sequence with the time interval (i, j) based on the segmentation points obtained in the step 1.1, and identifying the climbing by solving the maximum value of the objective function based on the climbing definition.
Wherein, the climbing is defined as:
|Pt+Δt-Pt|＞Pthreshold
in the formula, Pt+ΔtPower at time (t + Δ t); ptIs the power at time t; pthresholdFor a given thresholdThe value is obtained.
According to different climbing directions, climbing events can be divided into an ascending event and a descending event, wherein the ascending event refers to a sudden increase of wind power in a period of time and the descending event refers to a sudden decrease of wind power in a period of time.
The objective function and its constraints are:
Figure BDA0002811762070000041
S(i,j)＞S(i,k)+S(k+1,j)
S(i,j)＝(j-i)2R(i,j)
wherein S (i, k) is a score value of the corresponding section (i, k); and R (·) is a climbing criterion, and a climbing event occurs when the power in the time interval (i, j) meets a climbing definition formula, wherein R (i, j) is 1, and otherwise R (i, j) is 0.
In order to improve the accuracy of the overall climbing identification, in the identification process, some events (bump events) which have small change amplitude and are opposite to the change direction of the adjacent climbing in the wind power time sequence need to be identified. The uphill slope identification determines whether the bump event occurs through the following relation:
[pk+1-pk]×[1-B(k,k+1)]≥0
in the formula, pk+1And pkWind power values at (k +1) and k moments respectively; b (·) is a bump event criterion, and in the time interval (k, k +1), when a bump event occurs, B (k, k +1) becomes 1, and otherwise, B (k, k +1) becomes 0. When an uphill event is identified, when pk+1-pkAnd when the value is more than 0, the algorithm continues to execute, namely the climbing is identified according to the target function P (i, j) and the constraint condition thereof.
The downhill slope identification determines whether the bump event occurs through the following relation:
[pk+1-pk]×[1-B(k,k+1)]≤0
when identifying a downhill event, pk+1-pkIf the value is less than 0, the algorithm continues to execute, namely the climbing is identified according to the target function P (i, j) and the constraint condition of the target function P (i, j).
Step 1.3: and (3) after a climbing recognition result is obtained according to the step 1.1 and the step 1.2, extracting a characteristic value of the climbing event.
The flow of identifying the wind power climbing event and extracting the characteristic value by the OpSDA algorithm is shown in FIG. 3. For indirect climbing prediction, the quality of the early climbing recognition determines the accuracy of climbing prediction. Taking wind power data of a wind power plant 2019 of a Belgian company of Elia at 4 months, 4 days, 5:30 to 4 months, 6 days, 2:00 as an example, an identification result obtained by utilizing an OpSDA algorithm is shown in FIG. 4, the improved algorithm divides adjacent events with the same climbing direction into the same event according to an SDA segmentation point, a bump event is identified in the 1 st and the 3 rd climbing events and divided into the same event with the adjacent climbing, and the defect that the conventional SDA algorithm cannot identify long-period climbing is overcome, so that the OpSDA algorithm can effectively identify the climbing trend of the wind power sequence, and a data basis is provided for secondary feature extraction.
Wind power data with the installed capacity of 3796MW in 1 month of 2020 of a wind power plant in Elia Belgium is adopted, MATLAB statistics is carried out, a wind power climbing event is identified through an OpSDA algorithm, a characteristic value is extracted, and the identification result is shown in table 1. The start time in table 1 is converted into a digital format representation using a "datenum" function, and only the characteristic values of 22 wind power climbing events before 1 month are shown due to space limitation.
TABLE 1 climbing event and its eigenvalues
Figure BDA0002811762070000051
Note: "meshed" indicates climbing up; "↓" represents downward climbing.
Step 2: dividing input data consisting of wind power and climbing characteristic values by using a sliding window, inputting the divided input data into a CNN-LSTM prediction model, extracting climbing characteristics again by the CNN, predicting by the LSTM to obtain a plurality of prediction results which are advanced by 16 steps, and obtaining a final rolling multistep wind power prediction result by extracting a first prediction point of multistep prediction. The model structure is shown in fig. 5, and the prediction process includes the following steps:
step 2.1: a model input part: the wind power is combined with 4 typesThe hill climbing characteristic values constitute a data set X as input. The wind power P adopts wind power data of a wind power plant at a certain Belgium of Elia company, and one data point is arranged at intervals of 15 minutes; the four climbing characteristics are respectively the climbing rate RRClimbing amplitude RSWStart time RSTAnd duration RD(ii) a T is a specific moment in the wind power data; m is the predicted step length in advance; n is the width of the sliding window and is set to 32, namely, the model predicts 32 multiplied by 5 data of input historical power and climbing characteristics each time, and the input data set is { XT,XT-1,…,XT-n+1,XT-nIn which X isT＝{PT,(RR)T,(RSW)T,(RST)T,(RD)T}。
Step 2.2: CNN secondary extraction features: the CNN model part is provided with 3 convolutional layers and a maximum pooling layer for deep learning of the data set and extracting features. Wherein, the number of convolution kernels is set to be 4,16 and 32 respectively, and the size of the convolution kernel is 2 multiplied by 2. The CNN extracts data features by convolution operation through the movement of convolution kernels on the feature maps, the number of the convolution kernels determines the number of the feature maps, namely the depth of a next layer of convolution network, and as the number of the feature maps of each layer of the network increases, more complex features can be extracted from low-level features by multilayer convolution.
The convolutional layer performs feature extraction on input data through convolution operation, and the calculation formula is as follows:
Figure BDA0002811762070000061
in the formula (I), the compound is shown in the specification,
Figure BDA0002811762070000062
outputting a jth feature map of the ith layer;
Figure BDA0002811762070000063
the ith profile output representing the (l-1) th layer; "+" indicates convolution operation;
Figure BDA0002811762070000064
a convolution kernel weight matrix for connecting the jth characteristic diagram of the ith layer and the ith characteristic diagram of the (l-1) th layer;
Figure BDA0002811762070000065
is a bias matrix; n is a radical ofjIs a collection of input feature maps; f (-) activate function.
The pooling layer performs secondary feature extraction and information filtering on the output of the convolutional layer, thereby retaining the most significant features. And in the wind power climbing prediction based on the climbing characteristic value, selecting the most important information in the maximum pooling extraction climbing characteristic value. The formula of the pooling layer is as follows:
Figure BDA0002811762070000066
in the formula, down (·) represents a down-sampling function.
The model adopts a nonlinear activation function ReLU to construct a sparse matrix, and adopts an all-zero filling mode to keep characteristic dimensionality.
Step 2.3: LSTM prediction: the LSTM model part takes the output of the CNN model as input, adopts a single-layer LSTM network, sets the number of the neurons to be 128, and outputs a wind power prediction result through a full connection layer. The internal structure of the basic unit of the LSTM model comprises three control gates: the system comprises an input gate, a forgetting gate and an output gate, wherein the activation function formula of each gate is as follows:
Figure BDA0002811762070000071
in the formula: sigma is sigmoid function or tanh function; x is the number oftThe input vector at the current time t is obtained; wxi,Whi,Wci,Wxf,Wxo,Who,Wco,Wxc,WhcIs a weight parameter matrix; bi,bf,bo,bcIs a bias vector; c. CtVectors that are state units and instantaneous states; h istIs a status sheetOutputting the current t moment of the element; f. oftOutputting for a forgetting gate; i.e. itIs an input gate output; otIs output from the output gate.
In order to illustrate the influence of feature extraction and deep learning on prediction accuracy in wind power climbing event prediction, the method adopts Python as a programming language, the compiling environment is Pycharm Community Edition 2020, the RAM is 16GB, and the processor is AMD Ryzen 74800H. The following 2 cases were set:
1) case 1: the prediction model consists of an OpSDA algorithm and a CNN-LSTM network. And analyzing the performance of the climbing multi-step prediction model under different prediction step lengths, different gate widths epsilon and different input data according to the evaluation index of the model.
2) Case 2: the prediction model consists of an OpSDA algorithm and a CNN-LSTM network, and is compared with the performance of wind power climbing event prediction by using 2 prediction models of an LSTM neural network and a BP neural network. And comparing the optimal prediction result obtained by the model with the optimal prediction result of the recent climbing prediction research.
And (3) simulation results:
case1 analysis:
considering that in the process of climbing identification, the revolving door algorithm firstly compresses the wind power time-series data through the door width epsilon to obtain a plurality of SDA segmentation points, and the selection of the door width epsilon can bring certain influence on the climbing identification and the subsequent climbing prediction. In the multi-step prediction model, different prediction results can be obtained when the prediction step length is selected differently.
In summary, in order to evaluate different influences of parameters set by the model on the model and show the effect of considering the climbing feature value on improving the precision of the model in the input part of the prediction model, taking data in month 1 2020 as an example, the recall ratio R is adoptedCPrecision ratio FAFrequency deviation index BSKey success index CSIAnd analyzing the performance of the climbing multi-step prediction model under different prediction step lengths, different gate widths epsilon and different input data by using the four evaluation indexes.
Recall ratio RC:
Figure BDA0002811762070000081
Precision ratio FA:
Figure BDA0002811762070000082
Frequency deviation index BS:
Figure BDA0002811762070000083
Key success index CSI：
Figure BDA0002811762070000084
In the formula, NTPPredicting the times of occurrence and actual occurrence of the climbing event; n is a radical ofFNTo predict the number of times a hill climbing event does not occur but actually occurs; n is a radical ofFPTo predict the number of times a hill climbing event occurs but does not actually occur.
The prediction step length is respectively advanced by 4 steps (1h), 16 steps (4h) and 32 steps (8 h); the gate widths epsilon are respectively set to 10,25 and 50; the input of the prediction model is divided into two cases of wind power data only and wind power data plus a climbing characteristic value. The relationship of the four evaluation indexes is plotted as an evaluation index graph as shown in fig. 6 according to the following formula.
Figure BDA0002811762070000085
Figure BDA0002811762070000086
FIG. 6 is a graph showing the precision F on the abscissaAThe ordinate represents the recall ratio RCThe curve represents the key success index CSIThe slope line represents the frequency deviation index BS. DieThe prediction effect of the model can be visually shown according to the evaluation index points in fig. 6, and the better the prediction effect, the closer the evaluation index is to the upper right corner. The input power and the characteristic value after recall ratio are generally superior to the case of only inputting power, RCThe number of times reaches more than 0.9. When epsilon is 25, the evaluation points are mainly distributed near the diagonal line, and the recall ratio and the precision ratio can be satisfied more simultaneously than that when epsilon is 50 and epsilon is 10. The evaluation index result of the step length predicted in advance to be 16 steps (4h) is more compact, the prediction result is more stable, and by combining the existing statistical result, the duration of more than 95% of wind power climbing events is lower than 4.04h, so that the prediction step length is set to be 16 steps (4h) to be optimal.
In fig. 6, the round solid evaluation index point is the optimal parameter point, the predicted step length is 16 steps (4h), the gate width is set to be epsilon 25, and the model input is added with a climbing characteristic value. In this case, the precision ratio is 0.8587, the probability that the occurrence of the climbing event occupies all predicted climbing results is accurately predicted to be higher, and the accuracy of accurately predicting the actual occurrence of the climbing event reaches similar FAOptimum of the situation, RCThe value reaches 0.9240, and the B is higherSAnd CSI. In conclusion, the invention selects the situation with the optimal parameters for the CNN-LSTM prediction model to simulate. In actual engineering application, different optimal parameters can be set for prediction according to different wind power data sets.
Case2 analysis:
the section carries out wind power climbing event prediction performance comparison by using 2 prediction models of an LSTM neural network and a BP neural network. Data of an Elia website from 10 months to 9 months in 2019 are selected, the data are identified by an OpSDA algorithm in four quarters, the climbing is performed, characteristic values are extracted, the data and wind power data are used as model input, the model input is substituted into a CNN-LSTM network model, an LSTM model and a BP neural network model for training, 1000 data points are taken in each quarter, the first 500 data points are used as a training set, and the last 500 data points are used as a test set. In order to evaluate the performance of the prediction model, the average absolute percentage error I of the 3 climbing prediction models is respectively obtainedMAPEAccuracy ACCThe rate M of missing reportsIFalse alarm rate ERRecall ratio RCPrecision ratio FAOn predictionClimbing correct rate SNRAnd predicting the accuracy of downhill climbing SRAs shown in table 2.
TABLE 2 comparison of evaluation indexes of different prediction models
Figure BDA0002811762070000091
According to Table 2, the mean absolute percentage error I of 4 quaternary CNN-LSTM model ramp segmentsMAPEAre all less than 0.1, IMAPEThe smaller the wind power is, the better the prediction effect is. Wherein the wind power prediction effect of the third quarter is optimal, IMAPE0.0603 only, I of LSTM and BP modelsMAPEIs significantly larger than the prediction model proposed by the present invention. In addition, the seven evaluation indexes mainly reflect the prediction accuracy of the prediction model on the climbing event, and the climbing prediction accuracy A of the CNN-LSTM model in the first 3 quartersCCAre all above 0.82, although quarter 4ACCIs slightly insufficient but still higher than the other two prediction models, and has good effect of predicting the occurrence of the downward climbing, SR0.8041 is reached. Although the report missing rate M of the BP model and the LSTM model existsIOr false alarm rate ERLower than in the CNN-LSTM model, e.g., quarter 3, but A for these 2 methodsCCAre far less than the CNN-LSTM model. The CNN-LSTM model can ensure that higher accuracy is obtained under the condition of relatively low missing report rate and false report rate. And the CNN-LSTM model has higher R in 4 quartersCAnd FAAnd both indexes in the 2 nd quarter reach more than 0.9, namely the number of times of correctly predicting the occurrence of the climbing event accounts for more than nine times of the total number of times of predicting the occurrence of the climbing event and the total number of times of actually occurring the climbing event. Generally speaking, the accuracy of the prediction of the downward climbing event by each model is higher than that of the upward climbing event, and the upward climbing event is more difficult to predict than the downward climbing event.
In order to further verify the wind power climbing performance predicted by the model, based on the same evaluation index, the optimal prediction result obtained by the model is compared with the optimal prediction result of the recent climbing prediction research, and the result is shown in table 3. Although the overall prediction accuracy A of the CNN-LSTM model of the inventionCCIs slightly lower than an MLP-BT model and an MLP-MSAR model, but the model can better balance each evaluation index and ensure the recall ratio RCHigher and simultaneously better precision ratio FAEach index is higher than 0.81, and RC、FAAnd SRThe prediction performance of the model of the invention for predicting the up-and-down climbing events is better than 0.9.
TABLE 3 comparison of evaluation indexes of other literature models
Figure BDA0002811762070000101
The prediction of the climbing event by the CNN-LSTM model in 1 month 2020 is shown in FIG. 7. Fig. 7(a) is an actual wind power curve. Each rectangle in fig. 7(b), fig. 7(c) and fig. 7(d) represents a climbing event, an ascending climbing event is represented in the first quadrant, a descending climbing event is represented in the fourth quadrant, the abscissa can display the duration of each climbing event and the sequence number point of the wind power sample at the beginning of climbing, and the ordinate can display the amplitude of each climbing event. The method for displaying and predicting the climbing event in the invention, which is provided in figure 7, can more intuitively see the prediction effect of each model, and can provide better scheduling basis for a power grid by predicting the climbing amplitude and the climbing time.
The wind turbine generator can adjust the reserve margin according to the climbing prediction result during scheduling, so that the prediction result has certain continuity and stability. According to fig. 7, the prediction results of the LSTM model and the BP model both generate a severe fluctuation phenomenon, and particularly, the BP model predicts a long-time climbing event as a plurality of events with short climbing duration and large climbing amplitude. Fig. 7(b) can obviously show that the CNN-LSTM model has more excellent performance in the prediction of the climbing amplitude and the climbing duration, and after the climbing characteristics are extracted for the second time by the CNN, the LSTM model can more effectively learn the continuity of the wind power signal and the climbing event, obtain the climbing prediction result on the long-time wind power signal, and is more favorable for the safe operation and the economic scheduling of the power system.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims (3)

1. The wind power climbing event prediction method based on feature extraction and deep learning is characterized by comprising the following steps of:
step 1: climbing recognition is carried out on historical wind power by adopting an OpSDA algorithm, and 4 climbing characteristic values are extracted: climbing rate RRClimbing amplitude RSWStart time RSTAnd duration RD；
Step 2: dividing input data consisting of wind power and climbing characteristic values by using a sliding window, inputting the divided input data into a CNN-LSTM prediction model, extracting climbing characteristics again by the CNN, predicting by the LSTM to obtain a plurality of prediction results which are advanced by 16 steps, and obtaining a final rolling multistep wind power prediction result by extracting a first prediction point of multistep prediction;
and step 3: and carrying out climbing identification on the predicted power to obtain a prediction result of the climbing event.
2. The wind power climbing event prediction method based on feature extraction and deep learning according to claim 1, wherein the step 1 comprises the following specific steps:
step 1.1: constructing a parallelogram according to the adjustable parameter gate width epsilon by using a revolving gate algorithm to screen sample data, and obtaining a revolving gate algorithm segmentation point;
step 1.2: constructing a target function P (i, j) on any section of wind power time sequence with a time interval (i, j) based on the segmentation points obtained in the step 1.1, and identifying the climbing by solving the maximum value of the target function based on the climbing definition;
the grade climbing is defined as:
|Pt+Δt-Pt|＞Pthreshold
in the formula, Pt+ΔtWork at time (t + Δ t)Rate, PtPower at time t, PthresholdIs a given threshold;
the objective function P (i, j) and its constraint are:
Figure FDA0002811762060000011
S(i,j)＞S(i,k)+S(k+1,j)
S(i,j)＝(j-i)2R(i,j)
wherein S (i, k) is a score value of the corresponding section (i, k); r (·) is a climbing criterion, a climbing event occurs when the power in the time interval (i, j) satisfies a climbing definition formula, R (i, j) is 1, otherwise R (i, j) is 0;
in order to improve the accuracy of the overall climbing identification, a bump event which has a small change amplitude in a wind power time sequence and is opposite to the change direction of an adjacent climbing needs to be identified in the identification process; the uphill slope identification is used for judging whether the bump event occurs or not through the following relational expression:
[pk+1-pk]×[1-B(k,k+1)]≥0
in the formula, pk+1And pkWind power values at (k +1) and k moments respectively; b (·) is a bump event criterion, and in a time interval (k, k +1), when a bump event occurs, B (k, k +1) is equal to 1, otherwise, B (k, k +1) is equal to 0; when an uphill event is identified, when pk+1-pkWhen the value is more than 0, the algorithm is continuously executed, namely climbing is identified according to the target function P (i, j) and the constraint condition of the target function P (i, j);
the downhill slope identification determines whether the bump event occurs through the following relation:
[pk+1-pk]×[1-B(k,k+1)]≤0
when identifying a downhill event, pk+1-pkIf the value is less than 0, the algorithm is continuously executed, namely climbing is identified according to the target function P (i, j) and the constraint condition of the target function P (i, j);
step 1.3: and (3) after a climbing recognition result is obtained according to the step 1.1 and the step 1.2, extracting a characteristic value of the climbing event.
3. The wind power climbing event prediction method based on feature extraction and deep learning according to claim 1, wherein the step 2 comprises the following specific steps:
step 2.1: a model input part: forming a data set X by using wind power and 4 types of climbing characteristic values as input; the method comprises the following steps that wind power data of a wind power plant are sampled at intervals of 15 minutes to obtain wind power P; input data set is { XT,XT-1,…,XT-n+1,XT-nIn which X isT＝{PT,(RR)T,(RSW)T,(RST)T,(RD)TT is a specific moment in the wind power data, n is the width of the sliding window, and n is set to be 32;
step 2.2: CNN secondary extraction features: the CNN model part is provided with 3 convolutional layers and a maximum pooling layer for deep learning of the data set and extracting features; wherein, the number of convolution kernels is set to be 4,16 and 32 respectively, and the size of the convolution kernel is 2 multiplied by 2; the CNN moves on the feature map through a convolution kernel to carry out convolution operation and extract data features;
the convolutional layer performs feature extraction on input data through convolution operation, and the calculation formula is as follows:
Figure FDA0002811762060000021
in the formula (I), the compound is shown in the specification,
Figure FDA0002811762060000022
outputting a jth feature map of the ith layer;
Figure FDA0002811762060000023
the ith profile output representing the (l-1) th layer; "+" indicates convolution operation;
Figure FDA0002811762060000024
for connecting the convolution kernel weight matrix between the jth feature map of the ith layer and the ith feature map of the (l-1) th layer；
Figure FDA0002811762060000025
Is a bias matrix; n is a radical ofjIs a collection of input feature maps; f (-) an activation function;
the pooling layer performs secondary feature extraction and information filtering on the output of the convolutional layer, so that the most significant features are reserved; the formula of the pooling layer is as follows:
Figure FDA0002811762060000031
wherein down (-) is a down-sampling function;
the model adopts a nonlinear activation function ReLU to construct a sparse matrix, and adopts an all-zero filling mode to keep characteristic dimensionality;
step 2.3: LSTM prediction: the LSTM model part takes the output of the CNN model as input, adopts a single-layer LSTM network, sets the number of neurons to be 128, and outputs a wind power prediction result through a full connection layer; the internal structure of the basic unit of the LSTM model comprises three control gates: the system comprises an input gate, a forgetting gate and an output gate, wherein the activation function formula of each gate is as follows:
Figure FDA0002811762060000032
in the formula: sigma is sigmoid function or tanh function; x is the number oftThe input vector at the current time t is obtained; wxi,Whi,Wci,Wxf,Wxo,Who,Wco,Wxc,WhcIs a weight parameter matrix; bi,bf,bo,bcIs a bias vector; c. CtVectors that are state units and instantaneous states; h istIs the output of the state unit at the current time t; f. oftOutputting for a forgetting gate; i.e. itIs an input gate output; otIs output from the output gate.
CN202011398774.6A 2020-12-02 2020-12-02 Wind power climbing event prediction method based on feature extraction and deep learning Active CN112529275B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011398774.6A CN112529275B (en) 2020-12-02 2020-12-02 Wind power climbing event prediction method based on feature extraction and deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011398774.6A CN112529275B (en) 2020-12-02 2020-12-02 Wind power climbing event prediction method based on feature extraction and deep learning

Publications (2)

Publication Number Publication Date
CN112529275A true CN112529275A (en) 2021-03-19
CN112529275B CN112529275B (en) 2023-09-22

Family

ID=74997283

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011398774.6A Active CN112529275B (en) 2020-12-02 2020-12-02 Wind power climbing event prediction method based on feature extraction and deep learning

Country Status (1)

Country Link
CN (1) CN112529275B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766583A (en) * 2021-01-25 2021-05-07 湖北省气象服务中心（湖北省专业气象服务台） Wind power short-term prediction method and system based on climbing event feature recognition
CN113298290A (en) * 2021-04-27 2021-08-24 宁波职业技术学院 Wind power prediction method and system based on image generation
CN113358157A (en) * 2021-06-10 2021-09-07 国网甘肃省电力公司兰州供电公司 RST-PNN-GA-based power equipment temperature rise detection and early warning method
CN115712065A (en) * 2023-01-05 2023-02-24 湖南大学 Motor fault diagnosis method and system with perception matching of time-frequency revolving door and convolution kernel
CN116384598A (en) * 2023-06-06 2023-07-04 中国电力科学研究院有限公司 Wind power process prediction method and device based on time sequence event coding
CN116739172A (en) * 2023-06-19 2023-09-12 河海大学 Method and device for ultra-short-term prediction of offshore wind power based on climbing identification
CN116885691A (en) * 2023-05-30 2023-10-13 淮阴工学院 Wind power climbing event indirect prediction method
CN117196122A (en) * 2023-11-02 2023-12-08 湖南赛能环测科技有限公司 Wind power plant adjustment method and device based on wind power climbing time length
CN117293825A (en) * 2023-11-24 2023-12-26 天津电力工程监理有限公司 Wind speed and power hybrid prediction method and device considering wind power climbing
CN117408535A (en) * 2023-12-15 2024-01-16 国网山东省电力公司电力科学研究院 New energy climbing event rolling early warning method and system based on deep learning

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679282A (en) * 2013-09-30 2014-03-26 清华大学 Prediction method for wind power ramp
CN104820869A (en) * 2015-04-29 2015-08-05 武汉大学 Model-switching-mechanism-contained prediction method of wind power ramp event
CN106897789A (en) * 2017-01-03 2017-06-27 内蒙古电力(集团)有限责任公司 A kind of active power for wind power climbing event detecting method and device
CN106933778A (en) * 2017-01-22 2017-07-07 中国农业大学 A kind of wind power combination forecasting method based on climbing affair character identification
CN107909212A (en) * 2017-11-20 2018-04-13 武汉大学 Climbing event prediction method based on mesoscale wind power evolution serializing
CN110378504A (en) * 2019-04-12 2019-10-25 东南大学 A kind of photovoltaic power generation climbing probability of happening prediction technique based on higher-dimension Copula technology

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679282A (en) * 2013-09-30 2014-03-26 清华大学 Prediction method for wind power ramp
CN104820869A (en) * 2015-04-29 2015-08-05 武汉大学 Model-switching-mechanism-contained prediction method of wind power ramp event
CN106897789A (en) * 2017-01-03 2017-06-27 内蒙古电力(集团)有限责任公司 A kind of active power for wind power climbing event detecting method and device
CN106933778A (en) * 2017-01-22 2017-07-07 中国农业大学 A kind of wind power combination forecasting method based on climbing affair character identification
CN107909212A (en) * 2017-11-20 2018-04-13 武汉大学 Climbing event prediction method based on mesoscale wind power evolution serializing
CN110378504A (en) * 2019-04-12 2019-10-25 东南大学 A kind of photovoltaic power generation climbing probability of happening prediction technique based on higher-dimension Copula technology

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LI HAN 等: "Wind Power Ramp Event Forecasting Based on Feature Extraction and Deep Learning", ENERGIES, pages 1 - 19 *
黄麒元;王致杰;杜彬;盛戈;刘三明;: "基于前置分解组合预测方法的风电功率爬坡预测研究", 可再生能源, no. 12, pages 28 - 38 *

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766583A (en) * 2021-01-25 2021-05-07 湖北省气象服务中心（湖北省专业气象服务台） Wind power short-term prediction method and system based on climbing event feature recognition
CN112766583B (en) * 2021-01-25 2024-02-02 湖北省气象服务中心（湖北省专业气象服务台） Wind power short-term prediction method and system based on climbing event feature recognition
CN113298290A (en) * 2021-04-27 2021-08-24 宁波职业技术学院 Wind power prediction method and system based on image generation
CN113358157A (en) * 2021-06-10 2021-09-07 国网甘肃省电力公司兰州供电公司 RST-PNN-GA-based power equipment temperature rise detection and early warning method
CN115712065B (en) * 2023-01-05 2023-04-07 湖南大学 Motor fault diagnosis method and system with sensing matching of time-frequency revolving door and convolution kernel
CN115712065A (en) * 2023-01-05 2023-02-24 湖南大学 Motor fault diagnosis method and system with perception matching of time-frequency revolving door and convolution kernel
CN116885691A (en) * 2023-05-30 2023-10-13 淮阴工学院 Wind power climbing event indirect prediction method
CN116885691B (en) * 2023-05-30 2024-02-23 淮阴工学院 Wind power climbing event indirect prediction method
CN116384598A (en) * 2023-06-06 2023-07-04 中国电力科学研究院有限公司 Wind power process prediction method and device based on time sequence event coding
CN116739172A (en) * 2023-06-19 2023-09-12 河海大学 Method and device for ultra-short-term prediction of offshore wind power based on climbing identification
CN116739172B (en) * 2023-06-19 2023-12-08 河海大学 Method and device for ultra-short-term prediction of offshore wind power based on climbing identification
CN117196122A (en) * 2023-11-02 2023-12-08 湖南赛能环测科技有限公司 Wind power plant adjustment method and device based on wind power climbing time length
CN117196122B (en) * 2023-11-02 2024-02-06 湖南赛能环测科技有限公司 Wind power plant adjustment method and device based on wind power climbing time length
CN117293825A (en) * 2023-11-24 2023-12-26 天津电力工程监理有限公司 Wind speed and power hybrid prediction method and device considering wind power climbing
CN117293825B (en) * 2023-11-24 2024-02-13 天津电力工程监理有限公司 Wind speed and power hybrid prediction method and device considering wind power climbing
CN117408535A (en) * 2023-12-15 2024-01-16 国网山东省电力公司电力科学研究院 New energy climbing event rolling early warning method and system based on deep learning
CN117408535B (en) * 2023-12-15 2024-04-12 国网山东省电力公司电力科学研究院 New energy climbing event rolling early warning method and system based on deep learning

Also Published As

Publication number Publication date
CN112529275B (en) 2023-09-22

Similar Documents

Publication Publication Date Title
CN112529275A (en) Wind power climbing event prediction method based on feature extraction and deep learning
CN112949945B (en) Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network
CN110766212B (en) Ultra-short-term photovoltaic power prediction method for historical data missing electric field
CN107102969A (en) The Forecasting Methodology and system of a kind of time series data
CN110348624B (en) Sand storm grade prediction method based on Stacking integration strategy
CN111723527B (en) Method for predicting residual life of gear based on cocktail long-short-term memory neural network
CN104008164A (en) Generalized regression neural network based short-term diarrhea multi-step prediction method
CN112434390B (en) PCA-LSTM bearing residual life prediction method based on multi-layer grid search
CN115063020A (en) Cascade hydropower station multidimensional safety scheduling device and method based on risk monitoring fusion
CN108964023A (en) A kind of busbar voltage situation short term prediction method and system for power grid
CN116822920B (en) Flow prediction method based on cyclic neural network
CN113516271A (en) Wind power cluster power day-ahead prediction method based on space-time neural network
CN115146842A (en) Multivariate time series trend prediction method and system based on deep learning
CN114548482A (en) Creep type landslide kinetic energy change rate face-slip early warning method
CN114266201B (en) Self-attention elevator trapping prediction method based on deep learning
CN116401516A (en) Power load abnormal data detection and correction method based on deep learning
CN116050547A (en) Landing gear performance prediction method based on self-attention integrated learning
CN117494586B (en) Mountain torrent space-time prediction method based on deep learning
CN116976227B (en) Storm water increasing forecasting method and system based on LSTM machine learning
CN110222910A (en) A kind of active power distribution network Tendency Prediction method and forecasting system
CN113111592A (en) Short-term wind power prediction method based on EMD-LSTM
CN105224801A (en) A kind of multiple-factor reservoir reservoir inflow short-time forecast evaluation method
CN116258260A (en) Probability power load prediction method based on gating double convolution neural network
CN112183847B (en) QRNN improved Stacking algorithm-based train running wind speed probability prediction method
CN115293249A (en) Power system typical scene probability prediction method based on dynamic time sequence prediction

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant