CN113673788A - Photovoltaic power generation power prediction method based on decomposition error correction and deep learning - Google Patents
Photovoltaic power generation power prediction method based on decomposition error correction and deep learning Download PDFInfo
- Publication number
- CN113673788A CN113673788A CN202111118285.5A CN202111118285A CN113673788A CN 113673788 A CN113673788 A CN 113673788A CN 202111118285 A CN202111118285 A CN 202111118285A CN 113673788 A CN113673788 A CN 113673788A
- Authority
- CN
- China
- Prior art keywords
- photovoltaic power
- modal
- power generation
- gru
- decomposition
- 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.)
- Pending
Links
- 238000010248 power generation Methods 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 58
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 238000012937 correction Methods 0.000 title claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 12
- 230000005855 radiation Effects 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 241000288105 Grus Species 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 4
- 125000004122 cyclic group Chemical group 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 9
- 238000004590 computer program Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 6
- 229910052799 carbon Inorganic materials 0.000 description 6
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 5
- 230000015654 memory Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 101150029915 eIF3j gene Proteins 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000006386 neutralization reaction Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N3/08—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
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Marketing (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a photovoltaic power generation power prediction method based on decomposition error correction and deep learning, which comprises the following steps of: s1, forming a historical load characteristic library and an influence factor characteristic library; s2, obtaining K photovoltaic power modal components; s3, classifying the K photovoltaic power modal components obtained in the step S2; s4, training the CNN-GRU-AM network to obtain a prediction result of the photovoltaic power modal component; and S5, constructing an XGboost model to obtain a prediction result of the photovoltaic power generation power. The method can comprehensively consider the fluctuation characteristics of different modal components and improve the photovoltaic power generation power prediction precision.
Description
Technical Field
The invention belongs to the technical field of photovoltaic power generation power prediction, relates to a photovoltaic power generation power prediction method, and particularly relates to a photovoltaic power generation power prediction method based on decomposition error correction and deep learning.
Background
With the aggravation of the climate crisis, the establishment of a green low-carbon cyclic development system has become a common consensus of all mankind. Therefore, the strategic goals of 'carbon peak reaching and carbon neutralization' are put forward in China, and the emission of CO2 strives to reach the peak value 2030 years ago, and strives to realize carbon neutralization 2060 years ago. The vigorous development of new energy power generation is the key to controlling carbon emission and realizing the aim of carbon neutralization. However, the photovoltaic power generation system is influenced by natural factors such as day and night, weather, seasons and the like, the output has strong fluctuation and intermittency, the photovoltaic power generation system is difficult to participate in power grid dispatching, and the safe, reliable and economic operation of the power system is seriously influenced. Therefore, accurate photovoltaic power generation power prediction is crucial to operation and planning of the power grid.
Because the photovoltaic power generation system is influenced by complex external factors and physical modeling is difficult, the data-driven method is widely applied to the problem of photovoltaic power generation power prediction. In the aspect of machine learning, a support vector regression, a gaussian process regression, an XGBoost and the like are representative methods. In the aspect of deep learning, the method mainly comprises a BP neural network, an Elman neural network, a deep confidence network, a long-term and short-term memory network and the like. The long-time and short-time memory network has long-time memory capability and can be used as a complex nonlinear unit for constructing a larger deep neural network, so that better performance is achieved. However, the long-time and short-time memory network has a complex structure and a large number of over-parameters, so that the efficiency is low when large-scale photovoltaic power generation power prediction is carried out.
In addition to direct prediction of photovoltaic power generation power, more and more methods advocate decomposition of the original power curve and then prediction of the components. The common sequence decomposition algorithm comprises empirical mode decomposition and variational mode decomposition, and can decompose the photovoltaic power generation power sequence into a plurality of regular modal components, so that the prediction difficulty is reduced. However, the component sum after modal decomposition has an error with the original sequence, which limits the accuracy of photovoltaic power generation prediction. In addition, the method adopts the deep learning network with the same structure to predict different modal components, and the fluctuation characteristics of the different modal components are not considered.
Through searching, no prior art publication which is the same as or similar to the present invention is found.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a photovoltaic power generation power prediction method based on decomposition error correction and deep learning, can comprehensively consider the fluctuation characteristics of different modal components, and improves the photovoltaic power generation power prediction precision.
The invention solves the practical problem by adopting the following technical scheme:
a photovoltaic power generation power prediction method based on decomposition error correction and deep learning comprises the following steps:
s1, obtaining influence factor characteristics considering influences on the power generation power of the photovoltaic system, and forming a historical load characteristic library and an influence factor characteristic library;
s2, determining the optimal variational modal decomposition number K by using a center frequency method, and decomposing the photovoltaic power generation power sequence by using a variational modal decomposition VMD algorithm to obtain K photovoltaic power modal components;
s3, classifying the K photovoltaic power modal components obtained in the step S2 according to the central frequency of the photovoltaic power modal components;
s4, classifying the photovoltaic power modal components in the step S3, constructing a corresponding CNN-GRU-AM network, and training the CNN-GRU-AM network by utilizing the historical load feature library and the influence factor feature library formed in the step S1 to obtain a prediction result of the photovoltaic power modal components;
and S5, constructing an XGboost model, taking the photovoltaic power modal component prediction result obtained in the step S4 as the input of the XGboost model, taking a real photovoltaic power generation power sequence as a label, training the XGboost model, correcting the VMD decomposition error in the photovoltaic power modal component, and further obtaining the photovoltaic power generation power prediction result.
Moreover, the historical load characteristic library in the step S1 is a historical photovoltaic power generation power sequence; the influence factor feature library in the step S1 includes a meteorological factor feature sequence in which the temperature, the direct solar radiation, the diffused radiation, and the like are closely influenced by the photovoltaic power generation.
In step S2, the specific step of determining the number K of variational modal decompositions using the center frequency method includes:
initializing, and enabling the number M of the variational modal decomposition to be 2;
performing variation modal decomposition on the historical photovoltaic power generation power sequence to obtain M modal components;
thirdly, calculating the center frequency of each modal component;
and fourthly, judging whether the M modal components have repeated center frequencies. If not, making M equal to M +1, and repeating the step of the second step to the fourth step; if the optimal variation modal decomposition number K is M-1, the optimal variation modal decomposition number is output.
Wherein K is a natural number in step S2.
In step S3, the method for classifying the K photovoltaic power modal components obtained in step S2 includes: dividing K photovoltaic power modal components into: the low-frequency I-type component, the medium-frequency II-type component and the high-frequency III-type component;
wherein K is a natural number in step S3.
The specific method of step S4 is:
constructing a Convolutional Neural Network (CNN), and performing feature extraction on the K photovoltaic power modal components obtained in the step S2; and constructing three types of gated cyclic unit networks (GRUs) behind the CNN network, adding an Attention Mechanism (AM) at the tail end of the network to form three types of CNN-GRU-AM networks, and respectively predicting the I, II and III types of photovoltaic power modal components to obtain a prediction result of the photovoltaic power modal components.
Furthermore, the three types of CNN-GRU-AM models in step S4 are different in the number of layers of the GRU network: the three-layer GRU network is used in the III type CNN-GRU-AM model, and the three-layer GRU network is respectively used for training and predicting the I type modal component, the II type modal component and the III type modal component.
Moreover, the CNN-GRU-AM network training method in step S4 is:
constructing K CNN-GRU-AM models, respectively taking the historical load characteristic set and the influence factor characteristic set of K photovoltaic power modal components as the input of the CNN-GRU-AM models, taking the true values of the photovoltaic power modal components as labels, and training the network;
wherein the model uses "Adam" as the activation function and the mean absolute error as the loss function.
The invention has the advantages and beneficial effects that:
1. the invention provides an ultrashort-term photovoltaic power generation power prediction method based on VMD error correction and CNN-GRU-AM, which comprises the steps of firstly decomposing a photovoltaic power generation power sequence by using VMD, determining the decomposition number by adopting a central frequency method, and decomposing a non-stable photovoltaic power generation power sequence into multiple modes; then, three CNN-GRU-AM networks are constructed to predict modal components by utilizing the nonlinear local feature extraction capability of CNN and the high efficiency of the gated cyclic unit network in processing time sequence prediction problems so as to adapt to the fluctuation characteristics of different modal components; and finally, a modal component prediction result is used as the input of the XGboost, the photovoltaic power generation power can be predicted by using the advantage that the XGboost can prevent overfitting, so that the decomposition error of the VMD is corrected, and the photovoltaic power generation power prediction precision is improved.
2. The method divides the modal components into 3 types according to the fluctuation degree of the modal components, and provides three CNN-GRU-AM network structures for modal component prediction respectively so as to adapt to the fluctuation characteristics of different modal components, give full play to the advantages of different network structures, realize refined modal component prediction and improve the prediction effect;
3. the photovoltaic power generation prediction accuracy can be improved by using a strategy of decomposing a photovoltaic power generation power sequence by using the VMD, but the accuracy of the photovoltaic power generation prediction is limited by the VMD model, the modal component is difficult to accurately restore the original power data in the part with large photovoltaic power fluctuation, and the decomposition error exists. The error correction model based on the XGboost can correct the decomposition error of the VMD, and effectively improves the photovoltaic power prediction precision.
Drawings
Fig. 1 is a schematic diagram of a CNN-GRU-AM network structure provided in an embodiment of the present invention.
Detailed Description
The embodiments of the invention will be described in further detail below with reference to the accompanying drawings:
a photovoltaic power generation power prediction method based on decomposition error correction and deep learning is disclosed, as shown in FIG. 1, and comprises the following steps:
s1, obtaining influence factor characteristics considering influences on the power generation power of the photovoltaic system, and forming a historical load characteristic library and an influence factor characteristic library;
the historical load characteristic library in the step S1 is a historical photovoltaic power generation power sequence;
the influence factor feature library in the step S1 includes a meteorological factor feature sequence in which the temperature, the direct solar radiation, the diffused radiation, and the like are closely influenced by the photovoltaic power generation.
S2, determining the optimal variational modal decomposition number K by using a center frequency method, and decomposing the photovoltaic power generation power sequence by using a Variational Modal Decomposition (VMD) algorithm to obtain K photovoltaic power modal components;
the specific step of determining the number K of variational modal decompositions by using the center frequency method in step S2 includes:
initializing, and enabling the number M of the variational modal decomposition to be 2;
performing variation modal decomposition on the historical photovoltaic power generation power sequence to obtain M modal components;
thirdly, calculating the center frequency of each modal component;
and fourthly, judging whether the M modal components have repeated center frequencies. If not, making M equal to M +1, and repeating the step of the second step to the fourth step; if the optimal variation modal decomposition number K is M-1, the optimal variation modal decomposition number is output.
Wherein K is a natural number in step S2.
S3, classifying the K photovoltaic power modal components obtained in the step S2 according to the central frequency of the photovoltaic power modal components;
the method for classifying the K photovoltaic power modal components obtained in step S2 in step S3 includes: dividing K photovoltaic power modal components into: the low-frequency I-type component, the medium-frequency II-type component and the high-frequency III-type component are specifically as follows:
when the frequency of the modal component is lower than 10Hz, the modal component is a class I component, when the frequency of the modal component is higher than 100Hz, the modal component is a class III component, and the rest modal components are class II components.
Wherein K is a natural number in step S3.
S4, classifying the photovoltaic power modal components in the step S3, constructing a corresponding CNN-GRU-AM network, and training the CNN-GRU-AM network by utilizing the historical load feature library and the influence factor feature library formed in the step S1 to obtain a prediction result of the photovoltaic power modal components;
the specific method of step S4 is as follows: constructing a Convolutional Neural Network (CNN), and performing feature extraction on the K photovoltaic power modal components obtained in the step S2; and constructing three types of gated cyclic unit networks (GRUs) behind the CNN network, adding an Attention Mechanism (AM) at the tail end of the network to form three types of CNN-GRU-AM networks, and respectively predicting the I, II and III types of photovoltaic power modal components to obtain a prediction result of the photovoltaic power modal components.
The CNN network, GRU network and attention mechanism in step S4 are implemented using the tensierflow, keras deep learning toolkit in python programming language.
The three types of CNN-GRU-AM models in step S4 are different in the number of layers of the GRU network: the three-layer GRU network is used in the III type CNN-GRU-AM model, and the three-layer GRU network is respectively used for training and predicting the I type modal component, the II type modal component and the III type modal component.
The CNN-GRU-AM network training method in step S4 is:
constructing K CNN-GRU-AM models, respectively taking the historical load characteristic set and the influence factor characteristic set of K photovoltaic power modal components as the input of the CNN-GRU-AM models, taking the true values of the photovoltaic power modal components as labels, and training the network;
wherein the model uses "Adam" as the activation function and the mean absolute error as the loss function.
And S5, constructing an XGboost model, taking the photovoltaic power modal component prediction result obtained in the step S4 as the input of the XGboost model, taking a real photovoltaic power generation power sequence as a label, training the XGboost model, correcting the VMD decomposition error in the photovoltaic power modal component, and further obtaining the photovoltaic power generation power prediction result.
The XGBoost model in step S5 is implemented using the sklern machine learning toolkit in python programming language.
In this embodiment, the method for predicting the load by using the photovoltaic power generation power of the present invention, with the measured data of the photovoltaic power station in a certain area in the Guangdong as a research object, includes the following steps:
s1, obtaining influence factor characteristics considering influences on the power generation power of the photovoltaic system, and forming a historical load characteristic library and an influence factor characteristic library;
and S2, determining the variation modal decomposition number K by using a center frequency method, and decomposing the photovoltaic power generation power sequence by using a Variation Modal Decomposition (VMD) algorithm to obtain K photovoltaic power modal components. The specific implementation steps of the center frequency method are as follows:
initializing, and enabling the number M of the variational modal decomposition to be 2;
performing variation modal decomposition on the historical photovoltaic power generation power sequence to obtain M modal components;
thirdly, calculating the center frequency of each modal component;
and fourthly, judging whether the M modal components have repeated center frequencies. If not, making M equal to M +1, and repeating the step of the second step to the fourth step; if the optimal variation modal decomposition number K is M-1, the optimal variation modal decomposition number is output.
S3, dividing the K photovoltaic power modal components obtained in the step S2 into three types according to the size of the center frequency of the photovoltaic power modal components, wherein the three types are I type components with the frequency less than 10Hz, II type components with the frequency between 10Hz and 100Hz and III type components with the frequency more than 100 Hz;
s4, constructing three types of CNN-GRU-AM networks
In a CNN-GRU-AM network, a CNN module is responsible for feature extraction, a GRU module is responsible for predicting a modal component, and an AM module enhances the single-point prediction capability. The CNN module comprises two convolution layers and a pooling layer, a Dropout layer is added at the tail end to prevent overfitting, a full connection layer in the traditional CNN network is deleted, and a feature extraction result is directly input into the GRU module. The network structure of three types of CNN-GRU-AM models constructed by aiming at three types of photovoltaic power modal components in the embodiment is shown in FIG. 1, the curve of the type I component is relatively gentle, the fluctuation is small, and the model is suitable for a shallow neural network, so that the model 1 adopts a layer of GRU network; the fluctuation frequency of the class II component curve is increased, and the number of the neural network layers is also properly increased, so that the model 2 adopts a two-layer GRU network; the class iii component curve changes dramatically and is suitable for deeper neural networks, so model 3 uses three layers of GRU networks. The Adam activation function is used and the average absolute error is used as a loss function.
S5, constructing XGboost model
And (4) taking the photovoltaic power modal component prediction result obtained in the step (S5) as the input of the XGboost model, taking the real photovoltaic power generation power sequence as a label, training the XGboost model, correcting the VMD decomposition error existing in the photovoltaic power modal component, and further obtaining the photovoltaic power generation power prediction result. In the embodiment, a proper XGBoost key parameter is obtained through cross validation: the maximum depth of the tree is 6; the minimum leaf node sample weight sum is 1; the learning rate is 0.05; the number of iterations is 1000; the remaining parameters are default values.
For example, in this embodiment, measured data of a photovoltaic power station in a certain area in Guangdong is used as a research object, the data time range is 2 months from 2018 to 8 months from 2018, the photovoltaic power generation power is recorded every 5min, and 288 groups of data are recorded every day. In addition, 3 groups of meteorological data which are highly related to photovoltaic power generation are collected from a meteorological website and respectively comprise temperature, solar direct radiation and diffusion radiation. Because photovoltaic power generation can be carried out only in the daytime, 7: 00-18: 00 of each day is selected as a research object, original data is divided into a training set, a verification set and a test set according to the ratio of 5:1:1, model training and performance testing are carried out, and the prediction time scale is 5 min.
VMD decomposition is carried out on the photovoltaic power generation power sequence to obtain K modal components { u }1,u2,...,uk}. When for the ith modal component uiWhen the input characteristics of each CNN-GRU-AM network are predicted at the tth moment, the input characteristics of each CNN-GRU-AM network include:
phi 10 historical modal component data before the moment to be predicted ui,t-10,...,ui,t-2,ui,t-1}；
② 3 meteorological characteristic data { w) at time to be predicted1,t,w2,t,w3,t}；
And thirdly, time characteristic data of the moment to be predicted. This feature represents 132 time instants between 7:00 and 18:00 by {0, 1, 2, …, 131} respectively.
In order to accelerate the convergence of the deep learning network, the meteorological feature and the time feature variable are normalized by the formula
In the formula, x and x' represent the original feature vector and the normalized feature vector, respectively.
The variable mode decomposition is carried out on the photovoltaic power generation power sequence, and the optimal K is 8 obtained by a center frequency method. The 8 modal components obtained and their classification results are shown in table 1.
TABLE 1 center frequency of modal components
1. Predictive performance comparison
In order to verify the performance of the method provided by the present invention, this embodiment keeps the models of AM and XGBoost unchanged (AM is added to all methods, and XGBoost model is added to the method including VMD), and sets 8 sets of comparison experiments, which respectively are: LSTM, VMD-LSTM, GRU, VMD-GRU, SVR, VMD-SVM, CNN-GRU, and the method of the present invention. In this embodiment, MAE and Root Mean Square Error (RMSE) are used as model evaluation indexes, and the comparison of prediction errors is shown in table 2.
TABLE 2 comparison of prediction errors for different methods
As can be seen from table 2, overall, the prediction error based on the VMD method is smaller, which indicates that the prediction accuracy can be effectively improved by the method of predicting and summarizing the modal components after the variational modal decomposition. The prediction error of the CNN-GRU network is lower than that of the LSTM and GRU, which shows that the CNN carries out effective local feature extraction, so that the model can learn important information more easily. GRUs also exhibit certain advantages as an improved version of LSTM, with slightly higher prediction accuracy than LSTM. In addition, the prediction accuracy of the method provided by the invention is obviously higher than that of other methods, because the GRU networks with different structures are constructed aiming at modal components with different fluctuation characteristics, and the correctness of the scheme is verified by the prediction result.
2. Effect analysis of the proposed three types of GRU structures
For verifying the advantages of the proposed three types of GRU structures, the present embodiment sets three sets of contrast experiments, which are respectively: predicting all modal components by using a model 1 (a GRU layer); predicting all modal components by using a model 2 (two layers of GRUs); and thirdly, predicting all modal components by using a model 3 (three layers of GRU). The error comparison is shown in Table 3.
TABLE 3 prediction error comparison of different GRU architectures
As can be seen from table 3, if a single deep learning network is used for predicting all modal components, the prediction error will be significantly increased, which indicates that the method can fully volatilize the advantages of the learning network structures of different depths, and can well match the fluctuation characteristics of different modal components, thereby improving the accuracy of predicting the photovoltaic power generation power.
Analysis of AM and XGboost Effect
To verify the correctness of AM and XGBoost, this embodiment sets four sets of comparative experiments, which are respectively: firstly, AM and XGboost are not used; using only AM; using XGboost only and the method mentioned in the fourth step. The error comparison is shown in Table 4.
TABLE 4 prediction error comparison of AM and XGboost
As can be seen from table 4, compared with a method without using AM or XGBoost, the MAE is reduced by about 5% to 8% after adding AM, and the MAE is reduced by about 14% to 16% after adding XGBoost, which indicates that the prediction accuracy can be significantly improved by correcting the decomposition error of VMD using XGBoost, and the prediction performance of the model is further improved by using AM.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Claims (7)
1. A photovoltaic power generation power prediction method based on decomposition error correction and deep learning is characterized by comprising the following steps: the method comprises the following steps:
s1, obtaining influence factor characteristics considering influences on the power generation power of the photovoltaic system, and forming a historical load characteristic library and an influence factor characteristic library;
s2, determining the optimal variational modal decomposition number K by using a center frequency method, and decomposing the photovoltaic power generation power sequence by using a variational modal decomposition VMD algorithm to obtain K photovoltaic power modal components;
s3, classifying the K photovoltaic power modal components obtained in the step S2 according to the central frequency of the photovoltaic power modal components;
s4, classifying the photovoltaic power modal components in the step S3, constructing a corresponding CNN-GRU-AM network, and training the CNN-GRU-AM network by utilizing the historical load feature library and the influence factor feature library formed in the step S1 to obtain a prediction result of the photovoltaic power modal components;
and S5, constructing an XGboost model, taking the photovoltaic power modal component prediction result obtained in the step S4 as the input of the XGboost model, taking a real photovoltaic power generation power sequence as a label, training the XGboost model, correcting the VMD decomposition error in the photovoltaic power modal component, and further obtaining the photovoltaic power generation power prediction result.
2. The photovoltaic power generation power prediction method based on decomposition error correction and deep learning of claim 1, wherein: the historical load characteristic library in the step S1 is a historical photovoltaic power generation power sequence; the influence factor feature library in the step S1 includes a meteorological factor feature sequence in which the temperature, the direct solar radiation, the diffused radiation, and the like are closely influenced by the photovoltaic power generation.
3. The photovoltaic power generation power prediction method based on decomposition error correction and deep learning of claim 1, wherein: the specific step of determining the number K of variational modal decompositions by using the center frequency method in step S2 includes:
initializing, and enabling the number M of the variational modal decomposition to be 2;
performing variation modal decomposition on the historical photovoltaic power generation power sequence to obtain M modal components;
thirdly, calculating the center frequency of each modal component;
and fourthly, judging whether the M modal components have repeated center frequencies. If not, making M equal to M +1, and repeating the step of the second step to the fourth step; if the optimal variation modal decomposition number K is M-1, the optimal variation modal decomposition number is output.
Wherein K is a natural number in step S2.
4. The photovoltaic power generation power prediction method based on decomposition error correction and deep learning of claim 1, wherein: the method for classifying the K photovoltaic power modal components obtained in step S2 in step S3 includes: dividing K photovoltaic power modal components into: the low-frequency I-type component, the medium-frequency II-type component and the high-frequency III-type component;
wherein K is a natural number in step S3.
5. The photovoltaic power generation power prediction method based on decomposition error correction and deep learning of claim 1, wherein: the specific method of step S4 is as follows:
constructing a Convolutional Neural Network (CNN), and performing feature extraction on the K photovoltaic power modal components obtained in the step S2; and constructing three types of gated cyclic unit networks (GRUs) behind the CNN network, adding an Attention Mechanism (AM) at the tail end of the network to form three types of CNN-GRU-AM networks, and respectively predicting the I, II and III types of photovoltaic power modal components to obtain a prediction result of the photovoltaic power modal components.
6. The photovoltaic power generation power prediction method based on decomposition error correction and deep learning of claim 5, wherein: the three types of CNN-GRU-AM models in step S4 are different in the number of layers of the GRU network: the three-layer GRU network is used in the III type CNN-GRU-AM model, and the three-layer GRU network is respectively used for training and predicting the I type modal component, the II type modal component and the III type modal component.
7. The photovoltaic power generation power prediction method based on decomposition error correction and deep learning of claim 1, wherein: the CNN-GRU-AM network training method in step S4 is: constructing K CNN-GRU-AM models, respectively taking the historical load characteristic set and the influence factor characteristic set of K photovoltaic power modal components as the input of the CNN-GRU-AM models, taking the true values of the photovoltaic power modal components as labels, and training the network;
wherein the model uses "Adam" as the activation function and the mean absolute error as the loss function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111118285.5A CN113673788A (en) | 2021-09-23 | 2021-09-23 | Photovoltaic power generation power prediction method based on decomposition error correction and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111118285.5A CN113673788A (en) | 2021-09-23 | 2021-09-23 | Photovoltaic power generation power prediction method based on decomposition error correction and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113673788A true CN113673788A (en) | 2021-11-19 |
Family
ID=78549979
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111118285.5A Pending CN113673788A (en) | 2021-09-23 | 2021-09-23 | Photovoltaic power generation power prediction method based on decomposition error correction and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113673788A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114548845A (en) * | 2022-04-27 | 2022-05-27 | 北京智芯微电子科技有限公司 | Distribution network management method, device and system |
| CN115907131A (en) * | 2022-11-16 | 2023-04-04 | 国网宁夏电力有限公司经济技术研究院 | Method and system for building electric heating load prediction model in northern area |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516844A (en) * | 2019-07-25 | 2019-11-29 | 太原理工大学 | Multivariable based on EMD-PCA-LSTM inputs photovoltaic power forecasting method |
| CN111611859A (en) * | 2020-04-21 | 2020-09-01 | 河北工业大学 | Gait recognition method based on GRU |
| CN111860982A (en) * | 2020-07-06 | 2020-10-30 | 东北大学 | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU |
| CN112348271A (en) * | 2020-11-12 | 2021-02-09 | 华北电力大学 | Short-term photovoltaic power prediction method based on VMD-IPSO-GRU |
| CN112580899A (en) * | 2021-02-23 | 2021-03-30 | 国能日新科技股份有限公司 | Medium-and-long-term wind power generation prediction method and system fused with machine learning model |
-
2021
- 2021-09-23 CN CN202111118285.5A patent/CN113673788A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516844A (en) * | 2019-07-25 | 2019-11-29 | 太原理工大学 | Multivariable based on EMD-PCA-LSTM inputs photovoltaic power forecasting method |
| CN111611859A (en) * | 2020-04-21 | 2020-09-01 | 河北工业大学 | Gait recognition method based on GRU |
| CN111860982A (en) * | 2020-07-06 | 2020-10-30 | 东北大学 | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU |
| CN112348271A (en) * | 2020-11-12 | 2021-02-09 | 华北电力大学 | Short-term photovoltaic power prediction method based on VMD-IPSO-GRU |
| CN112580899A (en) * | 2021-02-23 | 2021-03-30 | 国能日新科技股份有限公司 | Medium-and-long-term wind power generation prediction method and system fused with machine learning model |
Non-Patent Citations (1)
| Title |
|---|
| 王鑫 等: "考虑风速波动特性的VMD-GRU短期风电功率预测", 《电力科学与技术学报》, no. 4, 31 August 2021 (2021-08-31), pages 20 - 28 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114548845A (en) * | 2022-04-27 | 2022-05-27 | 北京智芯微电子科技有限公司 | Distribution network management method, device and system |
| CN115907131A (en) * | 2022-11-16 | 2023-04-04 | 国网宁夏电力有限公司经济技术研究院 | Method and system for building electric heating load prediction model in northern area |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110942194A (en) | Wind power prediction error interval evaluation method based on TCN | |
| CN111461463B (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
| CN113673788A (en) | Photovoltaic power generation power prediction method based on decomposition error correction and deep learning | |
| CN110020712B (en) | Optimized particle swarm BP network prediction method and system based on clustering | |
| Li et al. | Multi-step ahead wind speed forecasting approach coupling maximal overlap discrete wavelet transform, improved grey wolf optimization algorithm and long short-term memory | |
| CN112149905A (en) | Photovoltaic power station short-term power prediction method based on wavelet transformation and wavelet neural network | |
| CN114282646B (en) | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement | |
| CN113361761A (en) | Short-term wind power integration prediction method and system based on error correction | |
| CN112149883A (en) | Photovoltaic power prediction method based on FWA-BP neural network | |
| CN107562992A (en) | A kind of Photovoltaic array maximum power tracking method based on SVM and particle cluster algorithm | |
| CN114048896A (en) | Method, system, equipment and medium for predicting photovoltaic power generation data | |
| CN114298377A (en) | Photovoltaic power generation prediction method based on improved extreme learning machine | |
| CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
| CN115271253A (en) | Water-wind power generation power prediction model construction method and device and storage medium | |
| CN112651576A (en) | Long-term wind power prediction method and device | |
| CN116702937A (en) | Photovoltaic output day-ahead prediction method based on K-means mean value clustering and BP neural network optimization | |
| CN111985712A (en) | Multi-step wind speed combined prediction model building method | |
| CN110276478B (en) | Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM | |
| CN115796327A (en) | Wind power interval prediction method based on VMD (vertical vector decomposition) and IWOA-F-GRU (empirical mode decomposition) -based models | |
| CN116628488A (en) | Training method of wind power prediction model, wind power prediction method and device | |
| CN115907131A (en) | Method and system for building electric heating load prediction model in northern area | |
| CN115545164A (en) | Photovoltaic power generation power prediction method, system, equipment and medium | |
| CN114862123A (en) | Comprehensive energy system scene generation method and device | |
| CN114386666A (en) | Wind power plant short-term wind speed prediction method based on space-time correlation | |
| CN114139783A (en) | Wind power short-term power prediction method and device based on nonlinear weighted combination |
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 |