CN110782024A - Photo-thermal electric field solar direct normal radiation prediction method based on convolutional neural network - Google Patents
Photo-thermal electric field solar direct normal radiation prediction method based on convolutional neural network Download PDFInfo
- Publication number
- CN110782024A CN110782024A CN201911117657.5A CN201911117657A CN110782024A CN 110782024 A CN110782024 A CN 110782024A CN 201911117657 A CN201911117657 A CN 201911117657A CN 110782024 A CN110782024 A CN 110782024A
- Authority
- CN
- China
- Prior art keywords
- layer
- prediction method
- normal radiation
- direct normal
- electric field
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 230000005855 radiation Effects 0.000 title claims abstract description 20
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 19
- 230000005684 electric field Effects 0.000 title claims abstract description 14
- 238000011176 pooling Methods 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000010248 power generation Methods 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 abstract description 2
- 230000007547 defect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 20
- 238000010586 diagram Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004146 energy storage Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- 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
- G01—MEASURING; TESTING
- G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
- G01J1/00—Photometry, e.g. photographic exposure meter
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Photovoltaic Devices (AREA)
Abstract
A photo-thermal electric field solar direct normal radiation intensity prediction method based on a convolutional neural network utilizes CNN in deep learning to design a CSP electric field DNI prediction method so as to overcome the defects of the traditional prediction method and accurately obtain a predicted value, so that a CSP power station is easy to schedule, and the impact on the existing power system caused by new energy power generation grid connection is further reduced. Firstly, the characteristics of direct normal radiation of the sun are analyzed, a convolutional neural network is selected according to the obtained characteristics, parameters in the network are modified and debugged, and finally a prediction method is obtained to reduce negative influence caused when the photo-thermal power station is connected into a power grid. The prediction method can accurately predict the direct normal radiation intensity of the sun of the photothermal electric field.
Description
Technical Field
The present invention relates to a Direct Normal radiation (DNI) prediction technique, and more particularly, to a Direct Normal radiation (CNN) prediction technique based on a Convolutional Neural Network (CNN).
Background
In recent years, a novel solar power generation mode, namely a solar thermal energy storage power (CSP) mode, appears in a historical stage, and the characteristic that the system output can be quickly adjusted by means of a special thermal energy storage subsystem, so that the impact on an existing power system when new energy power generation is connected to a grid is reduced becomes a research hotspot at the present stage. Because CSP power station needs a large amount of solar energy-DNI, so need to build in the northwest of China with above characteristics. The complicated and changeable climate conditions in northwest of China cause certain difficulties in the scheduling of CSP power stations and the prediction of output thereof. The existing prediction method aiming at DNI mainly utilizes the prediction of a neural network to estimate solar radiation. However, the traditional shallow prediction network generates gradient loss during working, falls into the problems of local minimum and the like, and reduces the accuracy of the whole network.
Disclosure of Invention
The invention aims to provide a photo-thermal electric field solar direct normal radiation prediction method based on a convolutional neural network.
The invention relates to a photo-thermal electric field solar direct normal radiation prediction method based on a convolutional neural network, which comprises the following steps:
step 1: from the essence, the direct normal radiation of the sun belongs to a time sequence, and the prediction of the time sequence can refer to the standard and the method of the time sequence prediction; the evaluation indexes adopted by the method are root mean square error, average relative error and average accuracy;
step 2: the overall framework of the prediction method is designed, and the main program of the solar direct normal radiation prediction method based on the convolutional neural network mainly comprises the following steps: preprocessing data, setting basic parameters of a network, initializing a subprogram, training a subprogram and testing the subprogram; the setting of the network basic parameters mainly comprises the following steps: the number of neurons of an input and output layer, the arrangement and the number of convolutional layers and pooling layers, the size of a convolutional kernel, a pooling mode, the number of output characteristic graphs of the convolutional layers, the types of an activation function and a cost function, and the distribution of training and testing samples;
and step 3: the invention relates to a solar direct normal radiation prediction method for designing a photo-thermal electric field, which takes time as input and takes solar direct normal radiation as output; therefore, the number of the neurons in the input layer is 4, the number of the neurons in the output layer is 1, namely the input sample is four-dimensional, and the output sample is one-dimensional;
and 4, step 4: combining the setting of the four-dimensional input vector in the last step, adding a convolution layer behind the input layer, and selecting one-dimensional convolution, wherein the size of a convolution kernel is set to be 1 x 2;
and 5: combining the setting of the convolution layer in the last step, adding a layer of pooling layer after the convolution layer, and selecting average pooling to retain more hidden information;
step 6: selecting a hyperbolic tangent function, namely a tanh curve as an activation function of the full connection layer of the model so as to accelerate the convergence speed of the method during training;
and 7: selecting a mean square error function as a cost function of the method to represent the accumulated value of all sample errors in time;
and 8: in the hidden layer, the number of convolutional layer output characteristic maps is preferably selected from a {100,200,300,400,500 }. When the number of the output characteristic graphs respectively takes the elements in the set A, corresponding cost function curves are obtained, the number of different output characteristic graphs and the corresponding cost function curves are compared, and the fact that when the parameter takes the value of 300, the convergence effect of the cost function meets the common requirement is found, so that the number of the output characteristic graphs takes the value of 300;
and step 9: training samples and test samples were assigned at 5: 1. After the network parameters are set according to the steps, the evaluation indexes are specifically as follows: RMSE 0.25527, MRE 0.20252, MA 0.79748, and training time 3.60690 s.
The invention has the advantages that: a series of problems generated in the prediction process of the traditional shallow neural network are solved, and meanwhile, the prediction precision is improved. Meanwhile, the method is combined with a static model of the CSP power station, and a certain technical basis can be provided for the development prospect of photo-thermal power generation in the future and the optimized operation of a new energy interconnection system containing the photo-thermal power station.
Drawings
Fig. 1 is a flow chart of a main program of the present invention, fig. 2 is a cost function curve when an output characteristic diagram takes different values, and fig. 3 is an error curve when a network is tested.
Detailed Description
The invention relates to a photo-thermal electric field solar direct normal radiation prediction method based on a convolutional neural network, which designs a DNI prediction method of a CSP electric field by utilizing CNN in deep learning so as to overcome the defects of the traditional prediction method and accurately obtain a predicted value, thereby facilitating the scheduling of a CSP power station and further reducing the impact on the existing power system when new energy power generation is connected to the grid.
The invention discloses a CSP electric field solar DNI prediction method based on CNN. In order to reduce the negative influence caused when the CSP power station is connected into a power grid, a prediction method of DNI (DNI) which is a main variable influencing the output of the CSP power station is designed by utilizing the excellent characteristic extraction and generalization capability of CNN (CNN), so as to achieve the aim of accurately predicting the DNI of the CSP power station, and the specific invention steps are as follows:
step 1: essentially, DNI belongs to a time series, and prediction thereof can be made with reference to time series prediction standards and methods. The evaluation indexes adopted by the invention are Root Mean Square Error (RMSE), Mean Relative Error (MRE) and Mean Accuracy (MA), and the calculation formula is as follows:
MA 1-MRE (formula III)
Wherein y (i) and net.o (i) represent actual values and predicted values, respectively; n represents the number of test samples.
Step 2: the overall framework of the prediction method is designed, and a flow chart of a main program of the CNN-based solar DNI prediction method is shown in fig. 1 and mainly comprises the following steps: preprocessing of data, setting of basic parameters of a network, an initialization subprogram, a training subprogram and a testing subprogram. The setting of the network basic parameters mainly comprises the following steps: the number of neurons in the input and output layer, the arrangement and the number of convolutional layers and pooling layers, the size of a convolutional kernel, a pooling mode, the number of output feature maps of the convolutional layers, the types of an activation function and a cost function, and the distribution of training and testing samples.
And step 3: the invention relates to a DNI prediction method for designing a CSP electric field, which takes time (year, month, day and hour) as input and DNI as output. Therefore, the number of input layer neurons is 4, the number of output layer neurons is 1, i.e., the input sample is four-dimensional, and the output sample is one-dimensional.
And 4, step 4: combining the setting of the four-dimensional input vector in the last step, adding a convolution layer behind the input layer, and selecting one-dimensional convolution, wherein the size of a convolution kernel is set to be 1 x 2, and the calculation method of the one-dimensional convolution comprises the following steps of;
wherein, represents convolution; (x) represents a convolution kernel; g (x) represents the input feature map.
And 5: and combining the setting of the convolution layer in the last step, adding a layer of pooling layer after the convolution layer, and selecting the mean pooling layer to reserve more hidden information. The mathematical expression of the pooling operation is:
in the formula, x
iRepresenting pooling layer input data; p is a preset parameter, when p is 1, the operation is mean pooling, when p → ∞ the operation is maximum pooling, and the present invention sets p to 1.
Step 6: a hyperbolic tangent function, namely a tanh curve is selected as an activation function of the full connection layer of the method, so that the convergence speed of the method during training is accelerated. The expression of the hyperbolic tangent function is as follows:
and 7: a Mean Squared Error (MSE) function is selected as a cost function of the method to represent the accumulated value of all sample errors in time. The expression for MSE is as follows:
and 8: in the hidden layer, the number of the convolutional layer output characteristic graphs is selected from A ═ {100,200,300,400 and 500 }; when the number of the output feature maps respectively takes the elements in the set a, the corresponding cost function curve is shown in fig. 2. Comparing the number of different output characteristic diagrams in fig. 2 and the corresponding cost functions, when the value of the parameter is 300, the convergence effect of the cost function is the best, and the value of time after iteration is completed is the minimum, so that the number of the output characteristic diagrams is 300;
and step 9: training samples and test samples were assigned at 5: 1. After the parameters of the prediction network are set according to the steps, the evaluation indexes are specifically as follows: the error curves are shown in fig. 3, where RMSE is 0.25527, MRE is 0.20252, MA is 0.79748, and training time is 3.60690 s.
The above is one of the implementation methods of the present invention, and it is obvious to a person skilled in the art that various changes can be made to the above embodiments without any creative effort, and the object of the present invention can be achieved. It will be apparent that such variations are intended to be included within the scope of the invention as defined in the claims.
Claims (1)
1. The photo-thermal electric field solar direct normal radiation prediction method based on the convolutional neural network is characterized by comprising the following steps of:
step 1: from the essence, the direct normal radiation of the sun belongs to a time sequence, and the prediction of the time sequence can refer to the standard and the method of the time sequence prediction; the evaluation indexes adopted by the method are root mean square error, average relative error and average accuracy;
step 2: the overall framework of the prediction method is designed, and the main program of the solar direct normal radiation prediction method based on the convolutional neural network mainly comprises the following steps: preprocessing data, setting basic parameters of a network, initializing a subprogram, training a subprogram and testing the subprogram; the setting of the network basic parameters mainly comprises the following steps: the number of neurons of an input and output layer, the arrangement and the number of convolutional layers and pooling layers, the size of a convolutional kernel, a pooling mode, the number of output characteristic graphs of the convolutional layers, the types of an activation function and a cost function, and the distribution of training and testing samples;
and step 3: the invention relates to a solar direct normal radiation prediction method for designing a photo-thermal electric field, which takes time as input and takes solar direct normal radiation as output; therefore, the number of the neurons in the input layer is 4, the number of the neurons in the output layer is 1, namely the input sample is four-dimensional, and the output sample is one-dimensional;
and 4, step 4: combining the setting of the four-dimensional input vector in the last step, adding a convolution layer behind the input layer, and selecting one-dimensional convolution, wherein the size of a convolution kernel is set to be 1 x 2;
and 5: combining the setting of the convolution layer in the last step, adding a layer of pooling layer after the convolution layer, and selecting average pooling to retain more hidden information;
step 6: selecting a hyperbolic tangent function, namely a tanh curve as an activation function of the full connection layer of the model so as to accelerate the convergence speed of the method during training;
and 7: selecting a mean square error function as a cost function of the method to represent the accumulated value of all sample errors in time;
and 8: in the hidden layer, the number of the convolutional layer output characteristic graphs is selected from A ═ {100,200,300,400 and 500 }; when the number of the output characteristic graphs respectively takes the elements in the set A, corresponding cost function curves are obtained, the number of different output characteristic graphs and the corresponding cost function curves are compared, and the fact that when the parameter takes the value of 300, the convergence effect of the cost function meets the common requirement is found, so that the number of the output characteristic graphs takes the value of 300;
and step 9: training samples and test samples were assigned at 5: 1. After the network parameters are set according to the steps, the evaluation indexes are specifically as follows: RMSE 0.25527, MRE 0.20252, MA 0.79748, and training time 3.60690 s.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911117657.5A CN110782024A (en) | 2019-11-15 | 2019-11-15 | Photo-thermal electric field solar direct normal radiation prediction method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911117657.5A CN110782024A (en) | 2019-11-15 | 2019-11-15 | Photo-thermal electric field solar direct normal radiation prediction method based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110782024A true CN110782024A (en) | 2020-02-11 |
Family
ID=69391488
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911117657.5A Pending CN110782024A (en) | 2019-11-15 | 2019-11-15 | Photo-thermal electric field solar direct normal radiation prediction method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110782024A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114241272A (en) * | 2021-11-25 | 2022-03-25 | 电子科技大学 | Heterogeneous information fusion positioning method based on deep learning |
| CN114895380A (en) * | 2022-04-22 | 2022-08-12 | 中国科学院地理科学与资源研究所 | Solar radiation prediction method, device, equipment and medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955768A (en) * | 2014-04-30 | 2014-07-30 | 河北省电力勘测设计研究院 | CSP radiation and heat energy predication method based on BP neural network model |
| CN108053061A (en) * | 2017-12-08 | 2018-05-18 | 天津大学 | A kind of solar energy irradiation level Forecasting Methodology based on improvement convolutional neural networks |
| US20180218254A1 (en) * | 2017-02-02 | 2018-08-02 | International Business Machines Corporation | Solar power forecasting with volumetric convolutional neural network |
| CN110059878A (en) * | 2019-04-15 | 2019-07-26 | 中国计量大学 | Based on CNN LSTM photovoltaic power generation power prediction model and its construction method |
| CN110070226A (en) * | 2019-04-24 | 2019-07-30 | 河海大学 | Photovoltaic power prediction technique and system based on convolutional neural networks and meta learning |
| CN110429667A (en) * | 2019-09-06 | 2019-11-08 | 国网甘肃省电力公司经济技术研究院 | Photo-thermal photovoltaic bundling electricity generation system capacity collocation method based on cuckoo search |
-
2019
- 2019-11-15 CN CN201911117657.5A patent/CN110782024A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955768A (en) * | 2014-04-30 | 2014-07-30 | 河北省电力勘测设计研究院 | CSP radiation and heat energy predication method based on BP neural network model |
| US20180218254A1 (en) * | 2017-02-02 | 2018-08-02 | International Business Machines Corporation | Solar power forecasting with volumetric convolutional neural network |
| CN108053061A (en) * | 2017-12-08 | 2018-05-18 | 天津大学 | A kind of solar energy irradiation level Forecasting Methodology based on improvement convolutional neural networks |
| CN110059878A (en) * | 2019-04-15 | 2019-07-26 | 中国计量大学 | Based on CNN LSTM photovoltaic power generation power prediction model and its construction method |
| CN110070226A (en) * | 2019-04-24 | 2019-07-30 | 河海大学 | Photovoltaic power prediction technique and system based on convolutional neural networks and meta learning |
| CN110429667A (en) * | 2019-09-06 | 2019-11-08 | 国网甘肃省电力公司经济技术研究院 | Photo-thermal photovoltaic bundling electricity generation system capacity collocation method based on cuckoo search |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114241272A (en) * | 2021-11-25 | 2022-03-25 | 电子科技大学 | Heterogeneous information fusion positioning method based on deep learning |
| CN114895380A (en) * | 2022-04-22 | 2022-08-12 | 中国科学院地理科学与资源研究所 | Solar radiation prediction method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
| CN108734331B (en) | Short-term photovoltaic power generation power prediction method and system based on LSTM | |
| CN113128793A (en) | Photovoltaic power combination prediction method and system based on multi-source data fusion | |
| CN106529814B (en) | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain | |
| CN113538391A (en) | Photovoltaic defect detection method based on Yolov4 and thermal infrared image | |
| CN110909919A (en) | Photovoltaic power prediction method of depth neural network model with attention mechanism fused | |
| CN106485075B (en) | Photovoltaic model parameter identification method based on eagle strategy and self-adaptive NM simplex | |
| CN109783841A (en) | A kind of photovoltaic power output time series analogy method based on more scene state transfer matrixes and conditional probability sampling | |
| CN111582555B (en) | Photovoltaic power prediction method based on foundation cloud picture image characteristics | |
| CN103942736B (en) | A kind of wind power plant multimachine equivalent modeling method | |
| CN110782024A (en) | Photo-thermal electric field solar direct normal radiation prediction method based on convolutional neural network | |
| CN117613883A (en) | Method and device for predicting generated power, computer equipment and storage medium | |
| CN112465251A (en) | Short-term photovoltaic output probability prediction method based on simplest gated neural network | |
| CN111985678A (en) | Photovoltaic power short-term prediction method | |
| CN107273569A (en) | Aerodynamic configuration drag reduction optimization method based on mesh deformation technique | |
| CN114418174A (en) | Electric vehicle charging load prediction method | |
| CN115829126A (en) | Photovoltaic power generation power prediction method based on multi-view self-adaptive feature fusion | |
| CN110909310A (en) | Photovoltaic short-term power generation capacity prediction method and system based on model parameter optimization | |
| CN114091317A (en) | Photovoltaic power station power prediction method based on NWP irradiance correction and error prediction | |
| CN111259550B (en) | Grid search and improved NM simplex algorithm-based photovoltaic model updating method | |
| CN116384583A (en) | Photovoltaic power prediction method based on multiple neural networks | |
| CN106600055A (en) | Wind speed prediction method the basis of self excitation threshold autoregression model | |
| CN116205377A (en) | Distributed photovoltaic power station output prediction method, system, computer and storage medium | |
| CN109145503A (en) | A kind of Dynamic High-accuracy modeling method of photovoltaic plant cluster | |
| CN110188964B (en) | Photovoltaic output prediction method based on correlation |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200211 |