CN117556379B - Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint - Google Patents
Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint Download PDFInfo
- Publication number
- CN117556379B CN117556379B CN202410045987.2A CN202410045987A CN117556379B CN 117556379 B CN117556379 B CN 117556379B CN 202410045987 A CN202410045987 A CN 202410045987A CN 117556379 B CN117556379 B CN 117556379B
- Authority
- CN
- China
- Prior art keywords
- feature
- power generation
- photovoltaic power
- data
- features
- 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.)
- Active
Links
- 238000010248 power generation Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000004927 fusion Effects 0.000 title claims abstract description 38
- 230000007246 mechanism Effects 0.000 claims abstract description 20
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 230000003993 interaction Effects 0.000 claims abstract description 16
- 230000003321 amplification Effects 0.000 claims abstract description 15
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 15
- 238000004458 analytical method Methods 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 40
- 230000004913 activation Effects 0.000 claims description 15
- 230000002452 interceptive effect Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 230000005855 radiation Effects 0.000 claims description 6
- 230000002277 temperature effect Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 5
- 238000004140 cleaning Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000007774 longterm Effects 0.000 abstract description 5
- 230000009471 action Effects 0.000 abstract description 2
- 239000000284 extract Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- 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)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Human Resources & Organizations (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Marketing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Power Engineering (AREA)
- Photovoltaic Devices (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the field of photovoltaic power prediction, and discloses a photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint. By collecting historical data related to photovoltaic power generation, carrying out principle analysis and optimizing knowledge in the photovoltaic power generation field, carrying out feature amplification on original data by utilizing a sliding window feature amplification mechanism, and constructing a parallel feature extraction network to capture short-term and long-term dependency relationship. Then, the local and global information are effectively fused using a feature interaction and feature intersection fusion module. Finally, the domain knowledge is introduced into the prediction model, the theoretical knowledge is utilized to guide the model to train, the action improves the physical interpretability and accuracy of the model to a certain extent, and the problems of inaccurate and unreasonable power prediction are well solved.
Description
Technical Field
The invention belongs to the technical field of photovoltaic power prediction, and particularly relates to a photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint.
Background
In the global energy transformation large background, the photovoltaic power generation technology will become one of the most dominant new energy power generation technologies in the future. However, since photovoltaic power generation is affected by weather conditions, solar altitude, solar radiation intensity, temperature, humidity, photovoltaic power generation power has significant intermittence, fluctuation, which causes unstable power generation output thereof. Therefore, the improvement of the accuracy of photovoltaic power generation power prediction has important practical significance.
At present, a photovoltaic system generation power prediction method is mainly researched based on artificial intelligence technologies such as machine learning, deep learning and the like. The traditional method comprises the following steps: and predicting by using a support vector machine, a random forest and a differential integration moving average autoregressive model. In addition, a great deal of study is also being conducted on a photovoltaic power generation power combination prediction method by a great deal of students, such as: CNN-LSTM combination prediction method, stacking prediction method, XGBoost-LSTM combination prediction model. The traditional method has lower prediction accuracy, the combined prediction method has a certain improvement on the prediction accuracy, but ignores short-term and long-term dependence between data, and does not combine the actual situation of a photovoltaic system, and whether the prediction result is reasonable is considered.
Therefore, the depth feature fusion is utilized to mine short-term and long-term dependency relationship between related data of photovoltaic power generation, and the method has important significance for solving the problem that the accuracy of the current prediction result is low. Meanwhile, interaction among solar radiation, weather conditions and characteristic factors of the photovoltaic module is considered through principle analysis to mine field knowledge, the field knowledge is brought into a training process of the model, physical interpretability and accuracy of the model are improved, and a reliable tool can be provided for optimization and management of a photovoltaic power generation system.
Disclosure of Invention
The invention aims to provide a photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint so as to realize accurate prediction of photovoltaic power generation power.
Step 1: the photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint comprises the following steps:
step S10, collecting historical data related to photovoltaic power generation, including meteorological data, photovoltaic panel parameters, power generation, solar radiation and cleaning data; further considering the mechanism and the characteristic of the photovoltaic power generation, analyzing the temperature effect of the photovoltaic module, grid-connected access of the photovoltaic system and the physical principle of the photovoltaic power generation to determine the domain knowledge constraint;
s20, inputting the collected data into a characteristic amplification module; extracting overall trend features in data in each window by utilizing a sliding window feature amplification mechanism, and serially connecting the extracted features to increase the number of features;
step S30, designing a parallel feature extraction network, wherein the parallel feature extraction network comprises two sub-networks CNN and a Transformer, the sub-network 1 (CNN framework) extracts key local features, and the sub-network 2 (Transformer framework) extracts overall global features; constructing a feature interaction module by utilizing a linear self-attention mechanism and convolution position coding, and constructing a feature cross fusion module by zooming, splicing and linearizing features, so as to effectively fuse the feature information extracted from two frame trunks;
and S40, excavating knowledge in the photovoltaic power generation field according to a photovoltaic power generation mechanism, constructing an activation function and a loss function to realize field knowledge constraint, and updating model parameters by adopting a random gradient descent method to obtain optimal learning parameters.
Step 2: according to the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint of the step 1, in the step S20, a sliding window feature amplification mechanism is specifically designed as follows: first, let p i Time sequence data of the ith time point, wherein i epsilon (1, n), wherein n is the length of the time sequence data; second, a single window size is set toDividing the data according to the size of the window to obtain a window w i Defined as->Wherein W is a window set, and m is the number of windows; then, the sliding step is set to ω, where +.>For the data samples in each window, extracting overall trend information in time series data related to the photovoltaic power generation power by adopting an average value, analyzing long-term change and trend of the data, realizing smooth processing of abnormal values, and reducing noise influence; finally, the features extracted from the adjacent windows are connected in series, so that feature amplification is realized.
Step 3: according to the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint in the step 1, in the step S30, local features and global features of data are obtained by utilizing the constructed parallel feature extraction network, interactive propagation between the two network features is realized through a linear self-attention mechanism, convolution position coding query interactive information is introduced, and the method is defined as follows:
Conv_Att(Q,K,V)=Linear_Att(Q,K,V)+CRPE(Q,V)
the Conv_Att (Q, K, V) represents a characteristic interaction module, the Q, K, V is obtained by carrying out Linear transformation on an input sequence, the linear_Att (Q, K, V) is the information attention and extraction of Linear attention to different positions in the input sequence, and the method specifically comprises the following steps:
d is the dimension of K, the K is subjected to softmax operation, the K is converted into probability distribution, T is transposed operation, dot product operation is carried out through transposed softmax (K) and V, and the weight of each key is distributed to a corresponding value; the interaction information is queried by using CRPE (Q, V), which is specifically as follows:
wherein,DepthwiseConv is the depth-separable convolution function, depthwiseConv (V) is the value-wise after depth-separable convolution, which is the Hadamard productAn amount of;
the feature cross fusion module is defined as:
f=ReLU(BN(Conv(Concat(μp(m low ),α·m hight ))))
wherein f is the characteristic after cross fusion, alpha is the attention coefficient, m low 、m hight A low-scale feature map and a high-scale feature map, up (m) low ) For m low Performing up-sampling operation, wherein Concat is a serial splicing function, conv is a convolution function, BN is a batch normalization function, reLU is an activation function, and m is calculated by dot multiplication alpha hight Scaled and associated with m low Serial splicing, performing convolution operation on the spliced features by Conv, normalizing each channel in the feature map by BN operation, and performing nonlinear activation on the feature map subjected to batch normalization by using ReLU.
Step 4: according to the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint in the step 1, in the step S40, positive constraint is realized by using an activation function through principle analysis, specifically:
y′ i =ReLU(PV(x i ;θ))
wherein y' i To inject the positive constrained predictor, PV (x i The method comprises the steps of carrying out a first treatment on the surface of the θ) is an unconstrained predictor, x i For inputting data, it is defined asWherein R is a sample set, N is the number of samples, D is the sample dimension, θ is a learnable parameter, θ is updated by a random gradient descent method to obtain an optimal learning parameter, reLU is an activation function, and a negative value is prevented from being generated by a prediction result by using the ReLU function; according to the temperature effect analysis of the photovoltaic module, interval constraint is realized by using a loss function, and the model loss function is as follows:
Loss(θ)=βMSE pv +μMSE rl
where β, μ is a weight, β+μ=1, the initial value is set to 0.5, and the subsequent adjustment can be made by whether the interval constraint is violated and contains an error larger than the true value; MSE (mean square error) pv Is unconstrainedThe prediction loss term specifically includes:n is the number of samples, y i Is true, PV (x i The method comprises the steps of carrying out a first treatment on the surface of the θ) is an unconstrained predictor; MSE (mean square error) rl The constraint penalty term for the interval is specifically:
wherein P is max 、P min For maximum and minimum of generated power, limiting model output to fall into [ P ] by utilizing ReLU function max ,P min ]A range; in the training process, the loss function is minimized by adjusting the weight, so that the accuracy of the model is improved.
The beneficial effects of the invention are as follows:
according to the invention, through principle analysis, domain knowledge is introduced into a photovoltaic power generation power prediction model, and the action is of practical significance to the photovoltaic power generation industry. The physical interpretability and accuracy of the model are improved to a certain extent by utilizing theoretical guiding model training, so that more accurate and reasonable prediction results are provided. Meanwhile, the invention reasonably uses the parallel feature extraction network to capture short-term and long-term dependency relationship of data in a parallel mode, and realizes interactive propagation between features of the two networks through self-attention linearization. In addition, the feature cross fusion module is utilized to explore the global-local information of the data, so that the accurate and reasonable prediction of the photovoltaic power generation power is realized.
Drawings
FIG. 1 is an algorithm architecture;
fig. 2 is a feature interaction-intersection fusion illustration.
Detailed Description
Embodiments of the invention are described in detail below with reference to the attached drawings, but the invention can be implemented in a number of different ways, which are defined and covered by the claims.
Under the constraint of field knowledge, the photovoltaic power generation power prediction method utilizes global-local information of data to predict photovoltaic power generation power through feature fusion.
Firstly, collecting historical data related to photovoltaic power generation, including meteorological data, photovoltaic panel parameters, power generation power and solar radiation, and carrying out principle analysis after data cleaning to optimize knowledge in the photovoltaic power generation field; secondly, performing characteristic amplification on the collected data by utilizing a sliding window characteristic amplification mechanism; then, extracting local and global features by adopting a parallel feature extraction network, and realizing feature fusion by utilizing a feature interaction module and a feature cross fusion module so as to explore global-local information; and finally, according to principle analysis, the domain knowledge is injected into the prediction model, and the accuracy and rationality of the prediction result are improved by guiding the model training process.
Step 1: the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint is shown as a figure 1, and shows an algorithm system structure of the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint, and comprises the following steps:
step S10, collecting historical data related to photovoltaic power generation, including meteorological data, photovoltaic panel parameters, power generation, solar radiation and cleaning data; further considering the mechanism and the characteristic of the photovoltaic power generation, analyzing the temperature effect of the photovoltaic module, grid-connected access of the photovoltaic system and the physical principle of the photovoltaic power generation to determine the domain knowledge constraint;
s20, inputting the collected data into a characteristic amplification module; extracting overall trend features in data in each window by utilizing a sliding window feature amplification mechanism, and serially connecting the extracted features to increase the number of features;
step S30, designing a parallel feature extraction network, wherein the parallel feature extraction network comprises two sub-networks CNN and a Transformer, the sub-network 1 (CNN framework) extracts key local features, and the sub-network 2 (Transformer framework) extracts overall global features; constructing a feature interaction module by utilizing a linear self-attention mechanism and convolution position coding, and constructing a feature cross fusion module by zooming, splicing and linearizing features, so as to effectively fuse the feature information extracted from two frame trunks;
and S40, excavating knowledge in the photovoltaic power generation field according to a photovoltaic power generation mechanism, constructing an activation function and a loss function to realize field knowledge constraint, and updating model parameters by adopting a random gradient descent method to obtain optimal learning parameters.
Step 2: according to the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint of the step 1, in the step S20, a sliding window feature amplification mechanism is specifically designed as follows: first, let p i Time sequence data of the ith time point, wherein i epsilon (1, n), wherein n is the length of the time sequence data; second, a single window size is set toDividing the data according to the size of the window to obtain a window w i Defined as->Wherein W is a window set, and m is the number of windows; then, the sliding step is set to ω, where +.>For the data samples in each window, adopting average value to extract characteristics, realizing the smooth processing of abnormal values and reducing noise influence; and finally, splicing the extracted features in the adjacent windows to realize feature amplification.
Step 3: according to the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint of the step 1, in the step S30, a constructed parallel feature extraction network is utilized to obtain data local features and global features; fig. 2 is a feature interaction-intersection fusion diagram shown in the application, and as shown in fig. 2, in order to effectively fuse local features and global features of data, a feature interaction module and a feature intersection fusion module are constructed, and a construction method thereof includes:
interactive propagation between two network features is achieved through a linear self-attention mechanism, defined as:
Conv_Att(Q,K,V)=Linear_Att(Q,K,V)+CRPE(Q,V)
the Conv_Att (Q, K, V) represents a characteristic interaction module, the Q, K, V is obtained by carrying out Linear transformation on an input sequence, the linear_Att (Q, K, V) is the information attention and extraction of Linear attention to different positions in the input sequence, and the method specifically comprises the following steps:
d is the dimension of K, the K is subjected to softmax operation, the K is converted into probability distribution, T is transposed operation, dot product operation is carried out through transposed softmax (K) and V, and the weight of each key is distributed to a corresponding value;
introducing convolution position coding inquiry interaction information, specifically:
wherein CRPE (Q, V) is the interactive information obtained by inquiry,the DepthwiseConv is a depth separable convolution function, and DepthwiseConv (V) is a value vector after the depth separable convolution;
the feature cross fusion module is defined as:
f=ReLU(BN(Conv(Concat(Up(m low ),α·m hight ))))
wherein f is the characteristic after cross fusion, alpha is the attention coefficient, m low 、m hight A low-scale feature map and a high-scale feature map, up (m) low ) For m low Performing up-sampling operation, wherein Concat is a serial splicing function, conv is a convolution function, BN is a batch normalization function, reLU is an activation function, and m is calculated by dot multiplication alpha hight Scaled and associated with m low Serial splicing, performing convolution operation on the spliced features by Conv, normalizing each channel in the feature map by BN operation, and performing nonlinear activation on the feature map subjected to batch normalization by using ReLU.
Step 4: according to the photovoltaic power generation power prediction method based on depth feature fusion under the domain knowledge constraint in the step 1, in the step S40, positive constraint is realized by using an activation function through principle analysis, specifically:
y′ i =ReLU(PV(x i ;θ))
wherein y' i To inject the positive constrained predictor, PV (x i The method comprises the steps of carrying out a first treatment on the surface of the θ) is an unconstrained predictor, x i For inputting data, it is defined asWherein R is a sample set, N is the number of samples, D is the sample dimension, θ is a learnable parameter, θ is updated by a random gradient descent method to obtain an optimal learning parameter, reLU is an activation function, and a negative value is prevented from being generated by a prediction result by using the ReLU function; according to the temperature effect analysis of the photovoltaic module, interval constraint is realized by using a loss function, and the model loss function is as follows:
Loss(θ)=βMSE pv +μMSE rl
where β, μ is a weight, β+μ=1, the initial value is set to 0.5, and the subsequent adjustment can be made by whether the interval constraint is violated and contains an error larger than the true value; MSE (mean square error) pv The method is used for unconstrained prediction loss items, and specifically comprises the following steps:n is the number of samples, y i Is true, PV (x i The method comprises the steps of carrying out a first treatment on the surface of the θ) is an unconstrained predictor; MSE (mean square error) rl The constraint penalty term for the interval is specifically:
wherein P is max 、P min For maximum and minimum of generated power, limiting model output to fall into [ P ] by utilizing ReLU function max ,P min ]A range; in the training process, the loss function is minimized by adjusting the weight, so that the accuracy of the model is improved.
The foregoing is merely representative of embodiments of the present invention and is not intended to limit the scope of the present invention. Any equivalent structures or equivalent flow transformations based on the description of the present invention and the drawings, as well as direct or indirect applications in other related arts, are within the scope of the present invention.
Claims (4)
1. The photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint is characterized by comprising the following steps of:
step S10, collecting historical data related to photovoltaic power generation, including meteorological data, photovoltaic panel parameters, power generation, solar radiation and cleaning data; analyzing the temperature effect of the photovoltaic module, grid-connected access of the photovoltaic system and extraction of field knowledge constraint of the physical principle of photovoltaic power generation;
s20, inputting the collected data into a characteristic amplification module; extracting overall trend features in data in each window by utilizing a sliding window feature amplification mechanism, and serially connecting the extracted features to increase the number of features;
step S30, designing a parallel feature extraction network, wherein the parallel feature extraction network comprises two sub-networks CNN and a Transformer; the sub-network 1 adopts a CNN framework to extract key local features, and the sub-network 2 captures overall global features based on a Transformer framework; constructing a feature interaction module and a feature cross fusion module; the feature interaction module consists of a linear self-attention mechanism and a convolution position code, the feature interaction propagation is realized by using the linear self-attention mechanism, and feature interaction information is inquired by using the convolution position code; the feature cross fusion module comprises the steps of performing feature scaling by using an attention coefficient, and performing serial splicing, convolution, batch normalization and linearization operation on the features to realize the fusion of key local feature information and overall global feature information;
and S40, excavating knowledge in the photovoltaic power generation field according to a photovoltaic power generation mechanism, constructing an activation function and a loss function to realize field knowledge constraint, and updating model parameters by adopting a random gradient descent method to obtain optimal learning parameters.
2. The method for predicting the photovoltaic power generation power by depth feature fusion under the domain knowledge constraint according to claim 1, wherein in the step S20, the sliding window feature amplification mechanism is specifically designed as follows: first, let p i Time sequence data of the ith time point, wherein i epsilon (1, n), wherein n is the length of the time sequence data; second, a single window size is set toDividing the data according to the size of the window to obtain a window w i Defined as->Wherein W is a window set, and m is the number of windows; then, the sliding step is set to ω, where +.>For the data samples in each window, extracting features by adopting an average value; and finally, serially connecting the extracted features in the adjacent windows.
3. The method for predicting the photovoltaic power generation power by depth feature fusion under the domain knowledge constraint according to claim 1, wherein in the step S30, the local features and the global features of the data are obtained by using the constructed parallel feature extraction network, the interactive propagation between the two network features is realized by a linear self-attention mechanism, and the convolutional position coding query interactive information is introduced, which is defined as:
Conv_Att(Q,K,V)=Linear_Att(Q,K,V)+CRPE(Q,V)
the Conv_Att (Q, K, V) represents a characteristic interaction module, the Q, K, V is obtained by carrying out Linear transformation on an input sequence, the linear_Att (Q, K, V) is the information attention and extraction of Linear attention to different positions in the input sequence, and the method specifically comprises the following steps:
d is the dimension of K, the K is subjected to softmax operation, the K is converted into probability distribution, T is transposed operation, dot product operation is carried out through transposed softmax (K) and V, and the weight of each key is distributed to a corresponding value; the interaction information is queried by using CRPE (Q, V), which is specifically as follows:
wherein,the DepthwiseConv is a depth separable convolution function, and DepthwiseConv (V) is a value vector after the depth separable convolution;
the feature cross fusion module is defined as:
f=ReLU(BN(Conv(Concat(Up(m low ),α·m hight ))))
wherein f is the characteristic after cross fusion, alpha is the attention coefficient, m low 、m hight A low-scale feature map and a high-scale feature map, up (m) low ) For m low Performing up-sampling operation, wherein Concat is a serial splicing function, conv is a convolution function, BN is a batch normalization function, reLU is an activation function, and m is calculated by dot multiplication alpha hight Scaled and associated with m low Serial splicing, namely performing convolution operation on the characteristics after serial splicing by using Conv, normalizing each channel in the characteristic diagram through BN operation, and performing nonlinear activation on the characteristic diagram after batch normalization by using ReLu.
4. The method for predicting the photovoltaic power generation power by depth feature fusion under domain knowledge constraint according to claim 1, wherein in the step S40, positive constraint is realized by using an activation function through principle analysis, specifically:
y′ i =ReLU(PV(x i ;θ))
wherein y' i To inject the positive constrained predictor, PV (x i The method comprises the steps of carrying out a first treatment on the surface of the θ) is an unconstrained predictor, x i For inputting data, it is defined asWherein R is a sample set, N is the number of samples, D is the sample dimension, θ is a learnable parameter, θ is updated by a random gradient descent method to obtain an optimal learning parameter, and ReLU is an activation function; according to the temperature effect analysis of the photovoltaic module, interval constraint is realized by using a loss function, and the model loss function is as follows:
Loss(θ)=βMSE pv +μMSE rl
wherein, beta and mu are weights, MSE pv The method is used for unconstrained prediction loss items, and specifically comprises the following steps:
where N is the number of samples, y i Is true, PV (x i The method comprises the steps of carrying out a first treatment on the surface of the θ) is an unconstrained predictor; MSE (mean square error) rl The constraint penalty term for the interval is specifically:
wherein P is max 、P min For maximum and minimum of generated power, limiting model output to fall into [ P ] by utilizing ReLU function max ,P min ]Range.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410045987.2A CN117556379B (en) | 2024-01-12 | 2024-01-12 | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410045987.2A CN117556379B (en) | 2024-01-12 | 2024-01-12 | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117556379A CN117556379A (en) | 2024-02-13 |
CN117556379B true CN117556379B (en) | 2024-04-09 |
Family
ID=89820944
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410045987.2A Active CN117556379B (en) | 2024-01-12 | 2024-01-12 | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117556379B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114021741A (en) * | 2021-09-23 | 2022-02-08 | 华能南京金陵发电有限公司 | Photovoltaic cell panel inspection method based on deep learning |
CN115829126A (en) * | 2022-12-07 | 2023-03-21 | 大连理工大学 | Photovoltaic power generation power prediction method based on multi-view self-adaptive feature fusion |
CN116227560A (en) * | 2023-02-06 | 2023-06-06 | 中国矿业大学 | Time sequence prediction model and method based on DTW-former |
CN116563681A (en) * | 2023-05-09 | 2023-08-08 | 安徽理工大学 | Gaze estimation detection algorithm based on attention crossing and two-way feature fusion network |
CN116911460A (en) * | 2023-07-20 | 2023-10-20 | 兰州理工大学 | Traffic flow prediction method combining multi-head attention with self-adaptive graph convolution |
CN116976505A (en) * | 2023-07-08 | 2023-10-31 | 河南大学 | Click rate prediction method of decoupling attention network based on information sharing |
CN116992986A (en) * | 2022-04-18 | 2023-11-03 | 北京航空航天大学 | Network traffic prediction system based on time sequence decomposition |
CN117273188A (en) * | 2023-04-14 | 2023-12-22 | 石河子大学 | Solar photovoltaic power generation power prediction method with incremental learning capability |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113128793A (en) * | 2021-05-19 | 2021-07-16 | 中国南方电网有限责任公司 | Photovoltaic power combination prediction method and system based on multi-source data fusion |
-
2024
- 2024-01-12 CN CN202410045987.2A patent/CN117556379B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114021741A (en) * | 2021-09-23 | 2022-02-08 | 华能南京金陵发电有限公司 | Photovoltaic cell panel inspection method based on deep learning |
CN116992986A (en) * | 2022-04-18 | 2023-11-03 | 北京航空航天大学 | Network traffic prediction system based on time sequence decomposition |
CN115829126A (en) * | 2022-12-07 | 2023-03-21 | 大连理工大学 | Photovoltaic power generation power prediction method based on multi-view self-adaptive feature fusion |
CN116227560A (en) * | 2023-02-06 | 2023-06-06 | 中国矿业大学 | Time sequence prediction model and method based on DTW-former |
CN117273188A (en) * | 2023-04-14 | 2023-12-22 | 石河子大学 | Solar photovoltaic power generation power prediction method with incremental learning capability |
CN116563681A (en) * | 2023-05-09 | 2023-08-08 | 安徽理工大学 | Gaze estimation detection algorithm based on attention crossing and two-way feature fusion network |
CN116976505A (en) * | 2023-07-08 | 2023-10-31 | 河南大学 | Click rate prediction method of decoupling attention network based on information sharing |
CN116911460A (en) * | 2023-07-20 | 2023-10-20 | 兰州理工大学 | Traffic flow prediction method combining multi-head attention with self-adaptive graph convolution |
Non-Patent Citations (4)
Title |
---|
A CNN-Transformer Hybrid Recognition Approach for sEMG-Based Dynamic Gesture Prediction;Yanhong Liu等;IEEE Transactions on Instrumentation and Measurement;20230508;第72卷;1-16 * |
CNN-Transformer混合模型在计算机视觉领域的研究综述;戴洋毅等;建模与仿真;20230715;第12卷(第4期);3657-3672 * |
Solar Energy Production Forecasting Based on a Hybrid CNN-LSTM-Transformer Model;Elham M. Al-Ali等;Mathematics;20230128;第11卷(第3期);1-19 * |
Transformer在计算机视觉场景下的研究综述;陈洛轩等;计算机科学;20230914;第50卷(第12期);130-147 * |
Also Published As
Publication number | Publication date |
---|---|
CN117556379A (en) | 2024-02-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
CN108053061B (en) | Solar irradiance prediction method based on improved convolutional neural network | |
Ibrahim et al. | An improved wind driven optimization algorithm for parameters identification of a triple-diode photovoltaic cell model | |
CN110288136B (en) | Wind power multi-step prediction model establishment method | |
CN111260126B (en) | Short-term photovoltaic power generation prediction method considering correlation degree of weather and meteorological factors | |
CN113205226B (en) | Photovoltaic power prediction method combining attention mechanism and error correction | |
CN111401755B (en) | Multi-new-energy output scene generation method, device and system based on Markov chain | |
CN111127246A (en) | Intelligent prediction method for transmission line engineering cost | |
CN111242355A (en) | Photovoltaic probability prediction method and system based on Bayesian neural network | |
CN114792156A (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
CN112990553A (en) | Wind power ultra-short-term power prediction method using self-attention mechanism and bilinear fusion | |
CN115169543A (en) | Short-term photovoltaic power prediction method and system based on transfer learning | |
Aliberti et al. | Forecasting Short-term Solar Radiation for Photovoltaic Energy Predictions. | |
Dou et al. | Elman neural network based short-term photovoltaic power forecasting using association rules and kernel principal component analysis | |
CN115456235A (en) | Photovoltaic power prediction system based on multi-mode fusion | |
Aishwarya et al. | Prediction of time series data using GA-BPNN based hybrid ANN model | |
CN117748495A (en) | Photovoltaic power station short-term generation power prediction method based on residual neural network | |
CN117556379B (en) | Photovoltaic power generation power prediction method based on depth feature fusion under domain knowledge constraint | |
Niu et al. | De-Trend First, Attend Next: A Mid-Term PV forecasting system with attention mechanism and encoder–decoder structure | |
CN117154690A (en) | Photovoltaic power generation power prediction method and system based on neural network | |
CN115496153A (en) | Multi-head self-attention deep convolution embedded clustering wind-light-load combined scene method | |
CN116128211A (en) | Wind-light-water combined short-term optimization scheduling method based on wind-light uncertainty prediction scene | |
CN115115139A (en) | Real-time prediction method for wind power | |
CN115660032A (en) | Building roof photovoltaic power prediction method based on BI-LSTM neural network fusion attention mechanism | |
CN112529285A (en) | Photovoltaic power generation power prediction method based on similar daily theory and PCA-PSO-BP |
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 |