CN114201924A - Solar irradiance prediction method and system based on transfer learning - Google Patents

Solar irradiance prediction method and system based on transfer learning Download PDF

Info

Publication number
CN114201924A
CN114201924A CN202210142291.2A CN202210142291A CN114201924A CN 114201924 A CN114201924 A CN 114201924A CN 202210142291 A CN202210142291 A CN 202210142291A CN 114201924 A CN114201924 A CN 114201924A
Authority
CN
China
Prior art keywords
data
solar irradiance
model
target
module
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210142291.2A
Other languages
Chinese (zh)
Other versions
CN114201924B (en
Inventor
严珂
黄晶
张伟
钟宜国
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Jingwei Information Technology Co ltd
Original Assignee
Hangzhou Jingwei Information Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangzhou Jingwei Information Technology Co ltd filed Critical Hangzhou Jingwei Information Technology Co ltd
Priority to CN202210142291.2A priority Critical patent/CN114201924B/en
Publication of CN114201924A publication Critical patent/CN114201924A/en
Application granted granted Critical
Publication of CN114201924B publication Critical patent/CN114201924B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/12—Timing analysis or timing optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Photovoltaic Devices (AREA)

Abstract

The invention discloses a solar irradiance prediction method and a solar irradiance prediction system based on transfer learning, and belongs to the technical field of solar irradiance prediction. The invention adopts CEEMDAN algorithm to construct the time sequence of the historical solar irradiance data of the target region and the source region
Figure 884462DEST_PATH_IMAGE002
、
Figure 354758DEST_PATH_IMAGE004
Correspondingly decomposing the data into a plurality of components, and carrying out sun processing on the input data by using a target model corresponding to each component obtained after model parameter migrationAnd predicting the irradiance, and finally superposing the solar irradiance prediction result of each target model into a solar irradiance prediction result corresponding to the input data, wherein the prediction effect and the prediction stability of the method are obviously improved compared with the conventional solar irradiance prediction method. According to the method, the target model is assisted by the transfer learning technology to acquire the knowledge learned by the source model so as to improve the prediction performance of the model, and the problem of low prediction precision of solar irradiance due to insufficient data volume in the prior art is solved.

Description

Solar irradiance prediction method and system based on transfer learning
Technical Field
The invention relates to the technical field of solar irradiance prediction, in particular to a solar irradiance prediction method and a solar irradiance prediction system based on transfer learning.
Background
Photovoltaic power generation is closely related to solar irradiance, and the power of photovoltaic power generation is directly proportional to the solar irradiance, so the solar irradiance is a main influence factor of photovoltaic power generation. However, due to instability of atmospheric environment and weather conditions, direct and scattered solar irradiance reaching the ground is not stable, so that fluctuation exists in photovoltaic power generation. In order to reduce the uncertainty of photovoltaic power generation, in the prior art, the approximate power generation amount of the photovoltaic equipment is known in advance by predicting the solar irradiance of the photovoltaic power generation equipment, so that the accurate prediction of the solar irradiance becomes the premise of accurate prediction of the photovoltaic power generation amount.
At present, the prediction of solar irradiance at home and abroad is mainly divided into a classical statistical method and a physical method. Classical statistical methods such as an autoregressive integrated moving average (ARIMA) and a multiple linear regression analysis prediction (MLR) can effectively simulate the mathematical relationship between the solar irradiance and historical data such as meteorological parameters, but if the historical solar irradiance and meteorological parameters at the installation point of the photovoltaic power generation equipment are lack, missing and incomplete, the robustness of a prediction model constructed by the historical data is poor.
The physical method is based on meteorological variables, establishes a conservation equation and predicts the atmospheric motion state and weather phenomenon at a certain time in the future. However, the problems of complex modeling, complex parameter solving and the like exist in the physical modeling, and the physical modeling is difficult to apply when the solar irradiance needs to be predicted in a short period or temporarily.
Disclosure of Invention
The invention provides a solar irradiance prediction method and a solar irradiance prediction system based on transfer learning, which are different from a physical method and aim to solve the problem that the prediction is not accurate enough due to insufficient data in the current solar irradiance prediction.
In order to achieve the purpose, the invention adopts the following technical scheme:
the method for predicting the solar irradiance based on the transfer learning comprises the following steps:
s1, collecting historical solar irradiance data of a plurality of days of a target area and a source area, constructing a time sequence, and respectively recording the historical solar irradiance data of the target area and the historical solar irradiance data of the source area in one day as the time sequence
Figure DEST_PATH_IMAGE001
，
Figure DEST_PATH_IMAGE002
，
Figure DEST_PATH_IMAGE003
、
Figure DEST_PATH_IMAGE004
Respectively representing time series
Figure DEST_PATH_IMAGE005
、
Figure DEST_PATH_IMAGE006
To (1)
Figure DEST_PATH_IMAGE007
Is first and second
Figure DEST_PATH_IMAGE008
The solar irradiance data is obtained by comparing the solar irradiance data,
Figure DEST_PATH_IMAGE009
、
Figure DEST_PATH_IMAGE010
respectively representing time series
Figure 901806DEST_PATH_IMAGE005
And
Figure 22209DEST_PATH_IMAGE006
the number of the elements in the (A) is,
Figure DEST_PATH_IMAGE011
，
Figure DEST_PATH_IMAGE012
time series of several days
Figure 552678DEST_PATH_IMAGE005
、
Figure 834973DEST_PATH_IMAGE006
Respectively forming a target data set and a source data set;
s2, after data normalization processing is carried out on each time sequence in the target data set and the source data set, each time sequence is subjected to each time sequence by adopting a CEEMDAN algorithm
Figure 993553DEST_PATH_IMAGE005
、
Figure 284857DEST_PATH_IMAGE006
Respectively decomposing the data to obtain each time sequence
Figure 630518DEST_PATH_IMAGE005
Corresponding to
Figure 298260DEST_PATH_IMAGE009
Each component and each time sequence
Figure 311347DEST_PATH_IMAGE006
Corresponding to
Figure DEST_PATH_IMAGE013
A component;
s3, performing data reconstruction on each component obtained by decomposing the source data set, dividing the data obtained by reconstruction into a training set and a test set, and then constructing an independent bidirectional long-short term memory network prediction model for each component based on the divided training set and the test set to serve as a source model;
s4, transferring the model parameters of each source model to a pre-constructed initial target model, then performing data reconstruction on each component obtained by decomposing the target data set, dividing the data obtained by reconstruction into a training set and a test set, respectively inputting the training set and the test set into the initial target model for model parameter optimization adjustment, and outputting a target model corresponding to each component obtained by decomposing the target data set;
and S5, respectively predicting the solar irradiance of the input data by using each target model obtained in the step S4, outputting a prediction result corresponding to each target model, and superposing the prediction results output by each target model to finally obtain a solar irradiance prediction result of the input data.
In a preferred embodiment of the present invention, in step S1, historical solar irradiance data is collected for 30 consecutive days of the target area.
In a preferred embodiment of the present invention, in step S1, historical solar irradiance data is collected for 365 consecutive days of the source region.
As a preferred embodiment of the present invention, the prediction accuracy of each target model is evaluated by using any one or more evaluation methods of an average absolute value error, a root mean square error, an average percentage error, and an average absolute ratio error.
The invention also provides a solar irradiance prediction system based on transfer learning, which can realize the solar irradiance prediction method, and the solar irradiance prediction system comprises:
the data acquisition module is used for acquiring historical solar irradiance data of a target area and a source area for a plurality of days;
a time sequence constructing module connected with the data acquisition module and used for constructing the acquired historical solar irradiance data in one day into the target area and the source area according to the data acquisition time sequenceRespectively recorded as a time sequence of historical solar irradiance data
Figure DEST_PATH_IMAGE014
，
Figure DEST_PATH_IMAGE015
，
Figure DEST_PATH_IMAGE016
、
Figure DEST_PATH_IMAGE017
Respectively representing time series
Figure DEST_PATH_IMAGE018
、
Figure 884804DEST_PATH_IMAGE006
To (1)
Figure DEST_PATH_IMAGE019
Is first and second
Figure DEST_PATH_IMAGE020
The solar irradiance data is obtained by comparing the solar irradiance data,
Figure DEST_PATH_IMAGE021
、
Figure DEST_PATH_IMAGE022
respectively representing time series
Figure DEST_PATH_IMAGE023
And
Figure DEST_PATH_IMAGE024
the number of the elements in the (A) is,
Figure DEST_PATH_IMAGE025
，
Figure DEST_PATH_IMAGE026
time series of several days
Figure 335071DEST_PATH_IMAGE023
、
Figure 947449DEST_PATH_IMAGE024
Respectively forming a target data set and a source data set;
the data normalization module is connected with the time sequence construction module and is used for carrying out data normalization processing on each time sequence in the target data set and the source data set; the data decomposition module is connected with the data normalization module and is used for adopting a CEEMDAN algorithm to carry out normalization on each time sequence
Figure 674097DEST_PATH_IMAGE023
、
Figure 182570DEST_PATH_IMAGE024
Respectively decomposing the data to obtain each time sequence
Figure 893037DEST_PATH_IMAGE023
Corresponding to
Figure 371423DEST_PATH_IMAGE021
Each component and each time sequence being
Figure 359101DEST_PATH_IMAGE024
Corresponding to
Figure 569634DEST_PATH_IMAGE022
A component;
the data reconstruction module is connected with the data decomposition module and is used for performing data reconstruction on each component obtained by decomposing the source data set and the target data set to obtain reconstruction data corresponding to the source data set and the target data set respectively;
the sample set dividing module is connected with the data reconstruction module and is used for dividing the reconstruction data corresponding to the source data set into a training set and a test set and dividing the reconstruction data corresponding to the target data set into the training set and the test set;
the sample set input module is connected with the sample set dividing module and is used for inputting a training set and a test set which are divided aiming at the reconstruction data of the source data set into the source model building module;
the source model building module is connected with the sample set dividing module and used for building an independent bidirectional long-short term memory network prediction model for each component obtained by decomposing the source data set based on input data to serve as a source model;
the model parameter migration module is connected with the source model construction module and used for acquiring the model parameters of each source model and migrating the model parameters to a pre-constructed initial target model;
the model parameter adjusting module is connected with the sample set dividing module and the model parameter transferring module and is used for inputting a training set and a test set which aim at the reconstruction data division of the target data set into the initial target model after the model parameter transferring is completed respectively to carry out model parameter adjustment by taking the model parameter transferring of the initial target model as a starting instruction;
the target model output module is connected with the model parameter adjusting module and used for outputting the decomposed time sequence after the model parameter adjustment is finished
Figure 767397DEST_PATH_IMAGE023
Obtaining a target model corresponding to each component;
and the solar irradiance prediction module is connected with the target model output module and used for respectively predicting the solar irradiance of input data by utilizing each target model and outputting a prediction result corresponding to each target model, and then superposing the prediction results output by each target model to finally obtain the solar irradiance prediction result of the input data.
The invention has the following beneficial effects:
1. historical solar irradiance data time sequence of target region and source region by adopting CEEMDAN algorithm
Figure 924840DEST_PATH_IMAGE023
、
Figure DEST_PATH_IMAGE027
Before data decomposition, the sequence of time series is carried out
Figure 829342DEST_PATH_IMAGE023
、
Figure 804252DEST_PATH_IMAGE027
Firstly, data normalization processing is carried out, so that the subsequent solar irradiance prediction speed is favorably improved;
2. adopting CEEMDAN algorithm to normalize the time sequence
Figure 657737DEST_PATH_IMAGE023
、
Figure 743505DEST_PATH_IMAGE027
The solar irradiance prediction method comprises the steps of correspondingly decomposing the solar irradiance into a plurality of components, performing solar irradiance prediction on input data by using a target model corresponding to each component obtained after model parameter migration, and finally superposing a solar irradiance prediction result of each target model into a solar irradiance prediction result corresponding to the input data, wherein the prediction effect and the prediction stability of the method are obviously improved compared with those of the conventional solar irradiance prediction method;
3. the target model is assisted by the transfer learning technology to acquire the knowledge learned by the source model so as to improve the prediction performance of the model, and the problem of low prediction precision of solar irradiance due to insufficient data volume in the prior art is solved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments of the present invention will be briefly described below. It is obvious that the drawings described below are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 is a schematic diagram of an implementation of a transfer learning-based solar irradiance prediction method according to an embodiment of the present invention;
FIG. 2 is a network structure diagram of a two-way long-short term memory network prediction model employed in an embodiment of the present invention;
FIG. 3 is a graph comparing the predicted effect of a transfer learning-based solar irradiance prediction method employed in embodiments of the present invention with other existing solar irradiance prediction methods;
FIG. 4 is a diagram of implementation steps of a method for predicting solar irradiance based on transfer learning according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a solar irradiance prediction system based on transfer learning according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained by the specific implementation mode in combination with the attached drawings.
Wherein the showings are for the purpose of illustration only and are shown by way of illustration only and not in actual form, and are not to be construed as limiting the present patent; to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numerals in the drawings of the embodiments of the present invention correspond to the same or similar components; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "inner", "outer", etc. are used for indicating the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, it is only for convenience of description and simplification of description, but it is not indicated or implied that the referred device or element must have a specific orientation, be constructed in a specific orientation and be operated, and therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and are not to be construed as limitations of the present patent, and the specific meanings of the terms may be understood by those skilled in the art according to specific situations.
In the description of the present invention, unless otherwise explicitly specified or limited, the term "connected" or the like, if appearing to indicate a connection relationship between the components, is to be understood broadly, for example, as being fixed or detachable or integral; can be mechanically or electrically connected; they may be directly connected or indirectly connected through intervening media, or may be connected through one or more other components or may be in an interactive relationship with one another. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
The solar irradiance prediction method based on transfer learning provided by the embodiment of the invention is shown in fig. 4 and fig. 1, and comprises the following steps:
step S1, collecting historical solar irradiance data of a target area (an area with less historical solar irradiance data volume and to-be-subjected solar irradiance prediction) and a source area (an area with sufficient historical solar irradiance data volume) for a plurality of days, constructing a time sequence, and respectively recording the historical solar irradiance data time sequence of the target area in one day as
Figure DEST_PATH_IMAGE028
The historical solar irradiance time sequence in one day of the source region is
Figure DEST_PATH_IMAGE029
，
Figure DEST_PATH_IMAGE030
、
Figure DEST_PATH_IMAGE031
Respectively representing time series
Figure 190929DEST_PATH_IMAGE018
、
Figure 477685DEST_PATH_IMAGE006
To (1)
Figure 118882DEST_PATH_IMAGE019
Is first and second
Figure 883707DEST_PATH_IMAGE020
The solar irradiance data is obtained by comparing the solar irradiance data,
Figure DEST_PATH_IMAGE032
、
Figure 28380DEST_PATH_IMAGE010
respectively representing time series
Figure 486038DEST_PATH_IMAGE018
And
Figure 614531DEST_PATH_IMAGE006
the number of the elements in the (A) is,
Figure DEST_PATH_IMAGE033
，
Figure DEST_PATH_IMAGE034
time series of several days
Figure 589571DEST_PATH_IMAGE018
、
Figure 932959DEST_PATH_IMAGE006
Respectively forming a target data set and a source data set;
it is emphasized here that the time series
Figure 420572DEST_PATH_IMAGE018
、
Figure 177306DEST_PATH_IMAGE006
The data in (1) is historical solar irradiance data of the same day, and a time sequence
Figure 549513DEST_PATH_IMAGE018
、
Figure 532426DEST_PATH_IMAGE006
The data acquisition time interval in (1) is preferably 15 minutes, i.e.
Figure DEST_PATH_IMAGE035
And
Figure DEST_PATH_IMAGE036
the data acquisition time interval is preferably 15 minutes, and since the historical solar irradiance data volume of the target area is small, the historical solar irradiance data of the target area for 30 days is selected, namely 30 time sequence sequences are selected
Figure 144935DEST_PATH_IMAGE018
As a target data set, historical solar irradiance data of 365 days per year in the source region is selected, namely 365 time sequence sequences are selected
Figure 513600DEST_PATH_IMAGE006
As a source data set.
Step S2, perform data normalization on each time sequence in the target data set and the source data set (there are many existing methods for performing data normalization on time sequences, so the specific data normalization method adopted in this embodiment is not described here), and use the CEEMDAN algorithm (fully adaptive noise set empirical mode decomposition) to perform data normalization on each time sequence after normalization
Figure 689497DEST_PATH_IMAGE018
、
Figure 460007DEST_PATH_IMAGE006
Respectively decomposing the data to obtain each time sequence
Figure 899210DEST_PATH_IMAGE018
Corresponding to
Figure 20750DEST_PATH_IMAGE032
A component (comprising
Figure DEST_PATH_IMAGE037
A natural mode function component and 1 residual component having different scales) and each time series
Figure 672442DEST_PATH_IMAGE006
Corresponding to
Figure 766300DEST_PATH_IMAGE010
A component (comprising
Figure DEST_PATH_IMAGE038
A natural mode function component having different scales and 1 remaining residual component); the "different scale" here refers to the time width between two local consecutive zero crossings in the timing signal.
Decomposition of time sequence by CEEMDAN algorithm
Figure 110825DEST_PATH_IMAGE018
、
Figure 188502DEST_PATH_IMAGE006
Can be expressed by the following formula (1) and formula (2):
Figure DEST_PATH_IMAGE039
Figure DEST_PATH_IMAGE040
in the formulae (1) to (2),
Figure DEST_PATH_IMAGE041
、
Figure DEST_PATH_IMAGE042
respectively, the CEEMDAN algorithm is shown in
Figure DEST_PATH_IMAGE043
Time resolved time series
Figure 801143DEST_PATH_IMAGE006
And time sequence
Figure 902171DEST_PATH_IMAGE018
，
Figure DEST_PATH_IMAGE044
、
Figure DEST_PATH_IMAGE045
Respectively represent
Figure 152018DEST_PATH_IMAGE043
Time of day decomposition time sequence
Figure DEST_PATH_IMAGE046
Obtained the first
Figure DEST_PATH_IMAGE047
Is first and second
Figure DEST_PATH_IMAGE048
A natural modal function component;
Figure DEST_PATH_IMAGE049
respectively represent
Figure 218457DEST_PATH_IMAGE043
Time of day decomposition time sequence
Figure 133323DEST_PATH_IMAGE046
The resulting residual component represents the dc component of the signal or the trend of the signal. Time sequence after data decomposition by CEEMDAN algorithm
Figure 936194DEST_PATH_IMAGE006
Is decomposed into
Figure DEST_PATH_IMAGE050
Each component, e.g. being represented as
Figure DEST_PATH_IMAGE051
、
Figure DEST_PATH_IMAGE052
、…、
Figure DEST_PATH_IMAGE053
、
Figure DEST_PATH_IMAGE054
Wherein, in the step (A),
Figure DEST_PATH_IMAGE055
to decompose the resulting natural modal function components,
Figure DEST_PATH_IMAGE056
the residual components obtained by decomposition. Likewise, time series
Figure 717461DEST_PATH_IMAGE018
Is decomposed into
Figure DEST_PATH_IMAGE057
And (4) a component.
Step S3, performing data reconstruction on each component obtained by decomposing the source data set (the reconstruction method is that the data dimension of the component is converted into the data dimension suitable for model input), dividing the data obtained by reconstruction into a training set and a test set (the data obtained by reconstruction is divided into the training set and the test set according to the ratio of 8: 2), and then constructing an individual two-way long and short term memory network prediction model for each component based on the divided training set and test set as a source model, and please refer to fig. 2 for the network structure of the two-way long and short term memory network prediction model adopted in this embodiment;
it should be noted that, in this embodiment, an individual bidirectional long-term and short-term memory network prediction model is constructed for each component obtained by decomposing the source data set, and each component includes local feature signals of different time scales of the original signal and is more stable than the original signal, which is beneficial to improving prediction accuracy.
Step S4, transferring the model parameters of each source model to a pre-constructed initial target model, then performing data reconstruction on each component obtained by decomposing a target data set (the reconstruction method is that the data dimension of the component is converted into the data dimension suitable for model input), dividing the data obtained by reconstruction into a training set and a test set, respectively inputting the training set and the test set into the initial target model for model parameter optimization and adjustment, and outputting the target model corresponding to each component obtained by decomposing the target data set;
the parameter optimization and adjustment specifically comprises the steps of training a target model on a target training set, fixing parameters of the previous layers, and finely adjusting the last two full-connection layers.
Step S5, respectively predicting the solar irradiance of the input data by using each target model obtained in step S4, outputting a prediction result corresponding to each target model, and then stacking the prediction results output by each target model to finally obtain a solar irradiance prediction result for the input data (there are many existing methods for stacking prediction results, and therefore, the method for stacking prediction results specifically adopted in this embodiment is not described here).
Since one of the purposes of the present invention is to help the target model obtain the knowledge learned by the source model through the transfer learning technique to improve the prediction performance of the self model, the historical solar irradiance time series sequence collected in step S1
Figure 51622DEST_PATH_IMAGE024
Data size ratio time series sequence of
Figure DEST_PATH_IMAGE058
The data amount of the target area needs much, and historical solar irradiance data of the target area for 30 days is selected, namely 30 time sequence sequences are selected
Figure 571510DEST_PATH_IMAGE058
As a target data set, historical solar irradiance data of 365 days per year in the source region is selected, namely 365 time sequence sequences are selected
Figure 760046DEST_PATH_IMAGE024
As a source data set.
In order to evaluate the prediction performance of the target model on the solar irradiance, the embodiment of the invention adopts any one or more of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), mean percentage error (MAPE) and mean absolute proportion error (MASE) evaluation methods to evaluate the prediction performance of each target model. The evaluation procedure of each error evaluation method is expressed by the following formulas (3) to (6):
Figure DEST_PATH_IMAGE059
Figure DEST_PATH_IMAGE060
Figure DEST_PATH_IMAGE061
Figure DEST_PATH_IMAGE062
in the formulae (3) to (6),
Figure DEST_PATH_IMAGE063
respectively representing the real value and the predicted value of the target model;
Figure DEST_PATH_IMAGE064
respectively represent
Figure DEST_PATH_IMAGE065
True value and
Figure DEST_PATH_IMAGE066
a predicted value of each target model;
Figure DEST_PATH_IMAGE067
indicating the length of the test set.
In step S4 of the solar irradiance prediction method based on transfer learning provided in this embodiment, according to the error evaluation of the prediction result of the initial target model, the internal parameters of the model are adjusted to achieve the lowest prediction error, that is, the average absolute value error (MAE), the Root Mean Square Error (RMSE), the average percentage error (MAPE), and the average absolute proportion error (MASE), to be the lowest. In order to compare the traditional machine learning model and the deep learning model, some models are selected for testing, the machine learning model comprises an Extreme Learning Machine (ELM) and a Back Propagation Neural Network (BPNN), and the deep learning model comprises a long-short term memory network (LSTM), a bidirectional long-short term memory network (BilsTM), an EMD algorithm combined with a bidirectional long-short term memory network (EMD-BilsTM) and a CEEMDAN algorithm combined with a bidirectional long-short term memory network (CEEMDAN-BilsTM), as shown in the error results in the following table 1 and shown in FIG. 3, compared with the traditional machine learning model and the deep learning model (ELM model, BPNN model, LSTM model, BilsTM model, EMD-BilsTM model and CEEMDAN-BilsTM model in the table 1), the target model based on the migration learning training (represented by 'proposed' in the table 1), has lower prediction error and better performance. "TRUE" in fig. 3 indicates a TRUE value.
Figure DEST_PATH_IMAGE069
TABLE 1
The present invention also provides a solar irradiance prediction system based on transfer learning, which can implement the above solar irradiance prediction method, as shown in fig. 5, the solar irradiance prediction system includes:
the data acquisition module is used for acquiring historical solar irradiance data of a target area and a source area for a plurality of days;
the time sequence construction module is connected with the data acquisition module and is used for constructing the acquired historical solar irradiance data in one day into time sequence sequences of the historical solar irradiance data of the target area and the historical solar irradiance data of the source area according to the data acquisition time sequence and respectively recording the time sequence sequences as the historical solar irradiance data of the target area and the historical solar irradiance data of the source area
Figure DEST_PATH_IMAGE070
，
Figure DEST_PATH_IMAGE071
，
Figure DEST_PATH_IMAGE072
、
Figure DEST_PATH_IMAGE073
Respectively representing time series
Figure DEST_PATH_IMAGE074
、
Figure DEST_PATH_IMAGE075
To (1)
Figure 979060DEST_PATH_IMAGE048
Is first and second
Figure DEST_PATH_IMAGE076
The solar irradiance data is obtained by comparing the solar irradiance data,
Figure DEST_PATH_IMAGE077
、
Figure 472621DEST_PATH_IMAGE010
respectively representing time series
Figure 994869DEST_PATH_IMAGE074
And
Figure 178857DEST_PATH_IMAGE075
the number of the elements in the (A) is,
Figure DEST_PATH_IMAGE078
，
Figure DEST_PATH_IMAGE079
time series of several days
Figure 82353DEST_PATH_IMAGE074
、
Figure 843636DEST_PATH_IMAGE075
Respectively forming a target data set and a source data set;
the data normalization module is connected with the time sequence construction module and is used for carrying out data normalization processing on the time sequence of each of the target data set and the source data set;
a data decomposition module connected with the data normalization module and used for adopting the CEEMDAN algorithm to carry out normalization on each time sequence
Figure 903996DEST_PATH_IMAGE074
、
Figure 676911DEST_PATH_IMAGE075
Respectively decomposing the data to obtain each time sequence
Figure 797313DEST_PATH_IMAGE074
Corresponding to
Figure 186837DEST_PATH_IMAGE077
Each component and each time sequence being
Figure 395096DEST_PATH_IMAGE075
Corresponding to
Figure 412731DEST_PATH_IMAGE010
A component;
the data reconstruction module is connected with the data decomposition module and is used for performing data reconstruction on each component obtained by decomposing the source data set and the target data set to obtain reconstruction data corresponding to the source data set and the target data set respectively;
the sample set dividing module is connected with the data reconstruction module and is used for dividing reconstruction data corresponding to the source data set into a training set and a test set and dividing reconstruction data corresponding to the target data set into the training set and the test set;
the sample set input module is connected with the sample set dividing module and is used for inputting the training set and the test set which are used for dividing the reconstruction data of the source data set into the source model building module;
the source model building module is connected with the sample set dividing module and used for building an independent bidirectional long-short term memory network prediction model for each component obtained by decomposing the source data set based on the input data to serve as a source model;
the model parameter migration module is connected with the source model construction module and used for acquiring the model parameters of each source model and migrating the model parameters to a pre-constructed initial target model;
the model parameter adjusting module is connected with the sample set dividing module and the model parameter transferring module and is used for inputting a training set and a testing set which are divided aiming at the reconstruction data of the target data set into the initial target model after the model parameter transferring is completed as a starting instruction to carry out model parameter adjustment;
the target model output module is connected with the model parameter adjusting module and used for outputting the decomposed time sequence after the model parameter adjustment is finished
Figure DEST_PATH_IMAGE080
Obtaining a target model corresponding to each component;
and the solar irradiance prediction module is connected with the target model output module and is used for respectively predicting the solar irradiance of the input data by utilizing each target model and outputting a prediction result corresponding to each target model, and then superposing the prediction results output by each target model to finally obtain the solar irradiance prediction result of the input data.
In conclusion, the invention helps the target model to acquire the knowledge learned by the source model through the transfer learning technology so as to improve the prediction performance of the model, solves the problem of low prediction precision of solar irradiance due to insufficient data volume in the prior art, and adopts the CEEMDAN algorithm to normalize the time sequence after normalization
Figure 770944DEST_PATH_IMAGE080
、
Figure DEST_PATH_IMAGE081
Correspondingly decomposing the data into a plurality of components, predicting the solar irradiance of the input data by using a target model corresponding to each component obtained after model parameter migration, and finally predicting the solar irradiance of each target modelThe solar irradiance prediction results are superposed to be the solar irradiance prediction results corresponding to the input data, and the prediction effect and the prediction stability of the method are obviously improved compared with the existing solar irradiance prediction method.
It should be understood that the above-described embodiments are merely preferred embodiments of the invention and the technical principles applied thereto. It will be understood by those skilled in the art that various modifications, equivalents, changes, and the like can be made to the present invention. However, such variations are within the scope of the invention as long as they do not depart from the spirit of the invention. In addition, certain terms used in the specification and claims of the present application are not limiting, but are used merely for convenience of description.

Claims (5)

1. A solar irradiance prediction method based on transfer learning is characterized by comprising the following steps:
s1, collecting historical solar irradiance data of a plurality of days of a target area and a source area, constructing a time sequence, and respectively recording the historical solar irradiance data of the target area and the historical solar irradiance data of the source area in one day as the time sequence
Figure 615188DEST_PATH_IMAGE001
，
Figure 904218DEST_PATH_IMAGE002
，
Figure 28339DEST_PATH_IMAGE003
、
Figure 111833DEST_PATH_IMAGE004
Respectively representing time series
Figure 80926DEST_PATH_IMAGE005
、
Figure 672182DEST_PATH_IMAGE006
To (1)
Figure 161063DEST_PATH_IMAGE007
Is first and second
Figure 182502DEST_PATH_IMAGE008
The solar irradiance data is obtained by comparing the solar irradiance data,
Figure 638891DEST_PATH_IMAGE009
、
Figure 269724DEST_PATH_IMAGE010
respectively representing time series
Figure 377226DEST_PATH_IMAGE005
And
Figure 458315DEST_PATH_IMAGE006
the number of the elements in the (A) is,
Figure 746208DEST_PATH_IMAGE011
，
Figure 682196DEST_PATH_IMAGE012
time series of several days
Figure 880090DEST_PATH_IMAGE005
、
Figure 240402DEST_PATH_IMAGE006
Respectively forming a target data set and a source data set;
s2, after data normalization processing is carried out on each time sequence in the target data set and the source data set, each time sequence is subjected to each time sequence by adopting a CEEMDAN algorithm
Figure 671383DEST_PATH_IMAGE005
、
Figure 909598DEST_PATH_IMAGE006
Are respectively provided withPerforming data decomposition to obtain each time sequence
Figure 729043DEST_PATH_IMAGE005
Corresponding to
Figure 496142DEST_PATH_IMAGE009
Each component and each time sequence
Figure 991583DEST_PATH_IMAGE006
Corresponding to
Figure 892543DEST_PATH_IMAGE010
A component;
s3, performing data reconstruction on each component obtained by decomposing the source data set, dividing the data obtained by reconstruction into a training set and a test set, and then constructing an independent bidirectional long-short term memory network prediction model for each component based on the divided training set and the test set to serve as a source model;
s4, transferring the model parameters of each source model to a pre-constructed initial target model, then performing data reconstruction on each component obtained by decomposing the target data set, dividing the data obtained by reconstruction into a training set and a test set, respectively inputting the training set and the test set into the initial target model for model parameter optimization adjustment, and outputting a target model corresponding to each component obtained by decomposing the target data set;
and S5, respectively predicting the solar irradiance of the input data by using each target model obtained in the step S4, outputting a prediction result corresponding to each target model, and superposing the prediction results output by each target model to finally obtain a solar irradiance prediction result of the input data.
2. The solar irradiance prediction method based on transfer learning of claim 1, wherein in step S1, historical solar irradiance data of the target region is collected for 30 consecutive days.
3. The solar irradiance prediction method based on transfer learning of claim 1, wherein historical solar irradiance data of the source region for 365 consecutive days is collected in step S1.
4. The solar irradiance prediction method based on the transfer learning of claim 1, wherein the prediction accuracy of each target model is evaluated by any one or more of an average absolute value error, a root mean square error, an average percentage error and an average absolute proportion error.
5. A solar irradiance prediction system based on transfer learning, which can implement the solar irradiance prediction method as claimed in any one of claims 1 to 4, wherein the solar irradiance prediction system comprises:
the data acquisition module is used for acquiring historical solar irradiance data of a target area and a source area for a plurality of days;
a time sequence construction module connected with the data acquisition module and used for constructing the acquired historical solar irradiance data in one day into time sequence sequences of the historical solar irradiance data of the target area and the historical solar irradiance data of the source area according to the data acquisition time sequence and respectively recording the time sequence sequences as the historical solar irradiance data of the target area and the historical solar irradiance data of the source area
Figure 189663DEST_PATH_IMAGE013
，
Figure 906426DEST_PATH_IMAGE014
，
Figure 797153DEST_PATH_IMAGE015
、
Figure 344546DEST_PATH_IMAGE016
Respectively representing time series
Figure 230594DEST_PATH_IMAGE017
、
Figure 729708DEST_PATH_IMAGE018
To (1)
Figure 468251DEST_PATH_IMAGE019
Is first and second
Figure 586379DEST_PATH_IMAGE020
The solar irradiance data is obtained by comparing the solar irradiance data,
Figure 497572DEST_PATH_IMAGE021
、
Figure 42954DEST_PATH_IMAGE022
respectively representing time series
Figure 891962DEST_PATH_IMAGE017
And
Figure 315246DEST_PATH_IMAGE018
the number of the elements in the (A) is,
Figure 582411DEST_PATH_IMAGE023
，
Figure 797229DEST_PATH_IMAGE024
time series of several days
Figure 133532DEST_PATH_IMAGE017
、
Figure 593464DEST_PATH_IMAGE018
Respectively forming a target data set and a source data set;
the data normalization module is connected with the time sequence construction module and is used for carrying out data normalization processing on each time sequence in the target data set and the source data set; data decompositionA module connected with the data normalization module and used for adopting a CEEMDAN algorithm to normalize each time sequence
Figure 951020DEST_PATH_IMAGE017
、
Figure 838205DEST_PATH_IMAGE018
Respectively decomposing the data to obtain each time sequence
Figure 35706DEST_PATH_IMAGE017
Corresponding to
Figure 158382DEST_PATH_IMAGE021
Each component and each time sequence being
Figure 524773DEST_PATH_IMAGE018
Corresponding to
Figure 947971DEST_PATH_IMAGE022
A component;
the data reconstruction module is connected with the data decomposition module and is used for performing data reconstruction on each component obtained by decomposing the source data set and the target data set to obtain reconstruction data corresponding to the source data set and the target data set respectively;
the sample set dividing module is connected with the data reconstruction module and is used for dividing the reconstruction data corresponding to the source data set into a training set and a test set and dividing the reconstruction data corresponding to the target data set into the training set and the test set;
the sample set input module is connected with the sample set dividing module and is used for inputting a training set and a test set which are divided aiming at the reconstruction data of the source data set into the source model building module;
the source model building module is connected with the sample set dividing module and used for building an independent bidirectional long-short term memory network prediction model for each component obtained by decomposing the source data set based on input data to serve as a source model;
the model parameter migration module is connected with the source model construction module and used for acquiring the model parameters of each source model and migrating the model parameters to a pre-constructed initial target model;
the model parameter adjusting module is connected with the sample set dividing module and the model parameter transferring module and is used for inputting a training set and a test set which aim at the reconstruction data division of the target data set into the initial target model after the model parameter transferring is completed respectively to carry out model parameter adjustment by taking the model parameter transferring of the initial target model as a starting instruction;
the target model output module is connected with the model parameter adjusting module and used for outputting the decomposed time sequence after the model parameter adjustment is finished
Figure 134233DEST_PATH_IMAGE005
Obtaining a target model corresponding to each component;
and the solar irradiance prediction module is connected with the target model output module and used for respectively predicting the solar irradiance of input data by utilizing each target model and outputting a prediction result corresponding to each target model, and then superposing the prediction results output by each target model to finally obtain the solar irradiance prediction result of the input data.
CN202210142291.2A 2022-02-16 2022-02-16 Solar irradiance prediction method and system based on transfer learning Active CN114201924B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210142291.2A CN114201924B (en) 2022-02-16 2022-02-16 Solar irradiance prediction method and system based on transfer learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210142291.2A CN114201924B (en) 2022-02-16 2022-02-16 Solar irradiance prediction method and system based on transfer learning

Publications (2)

Publication Number Publication Date
CN114201924A true CN114201924A (en) 2022-03-18
CN114201924B CN114201924B (en) 2022-06-10

Family

ID=80645488

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210142291.2A Active CN114201924B (en) 2022-02-16 2022-02-16 Solar irradiance prediction method and system based on transfer learning

Country Status (1)

Country Link
CN (1) CN114201924B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116167465A (en) * 2023-04-23 2023-05-26 杭州经纬信息技术股份有限公司 Solar irradiance prediction method based on multivariate time series ensemble learning
CN116451598A (en) * 2023-06-20 2023-07-18 杭州经纬信息技术股份有限公司 Solar irradiance prediction method based on denoising diffusion probability model
CN116488158A (en) * 2023-04-24 2023-07-25 国能日新科技股份有限公司 Method and device for predicting medium-long-term electric quantity of wind power based on transfer learning

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559561A (en) * 2013-11-13 2014-02-05 上海电气集团股份有限公司 Super-short-term prediction method of photovoltaic power station irradiance
CN110635476A (en) * 2019-09-29 2019-12-31 合肥工业大学 Knowledge migration-based cross-regional interconnected power grid dynamic scheduling rapid optimization method
CN110766212A (en) * 2019-10-15 2020-02-07 哈尔滨工程大学 Ultra-short-term photovoltaic power prediction method for historical data missing electric field
CN111461410A (en) * 2020-03-10 2020-07-28 北京邮电大学 Air quality prediction method and device based on transfer learning
CN112100911A (en) * 2020-09-08 2020-12-18 淮阴工学院 Solar radiation prediction method based on deep BISLTM
CN112183877A (en) * 2020-10-13 2021-01-05 上海大学 Photovoltaic power station fault intelligent diagnosis method based on transfer learning
CN112257941A (en) * 2020-10-28 2021-01-22 福州大学 Photovoltaic power station short-term power prediction method based on improved Bi-LSTM
CN112434891A (en) * 2020-12-18 2021-03-02 中国计量大学上虞高等研究院有限公司 Method for predicting solar irradiance time sequence based on WCNN-ALSTM
CN113762417A (en) * 2021-10-15 2021-12-07 南京澄实生物科技有限公司 Method for enhancing HLA antigen presentation prediction system based on deep migration

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559561A (en) * 2013-11-13 2014-02-05 上海电气集团股份有限公司 Super-short-term prediction method of photovoltaic power station irradiance
CN110635476A (en) * 2019-09-29 2019-12-31 合肥工业大学 Knowledge migration-based cross-regional interconnected power grid dynamic scheduling rapid optimization method
CN110766212A (en) * 2019-10-15 2020-02-07 哈尔滨工程大学 Ultra-short-term photovoltaic power prediction method for historical data missing electric field
CN111461410A (en) * 2020-03-10 2020-07-28 北京邮电大学 Air quality prediction method and device based on transfer learning
CN112100911A (en) * 2020-09-08 2020-12-18 淮阴工学院 Solar radiation prediction method based on deep BISLTM
CN112183877A (en) * 2020-10-13 2021-01-05 上海大学 Photovoltaic power station fault intelligent diagnosis method based on transfer learning
CN112257941A (en) * 2020-10-28 2021-01-22 福州大学 Photovoltaic power station short-term power prediction method based on improved Bi-LSTM
CN112434891A (en) * 2020-12-18 2021-03-02 中国计量大学上虞高等研究院有限公司 Method for predicting solar irradiance time sequence based on WCNN-ALSTM
CN113762417A (en) * 2021-10-15 2021-12-07 南京澄实生物科技有限公司 Method for enhancing HLA antigen presentation prediction system based on deep migration

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
BIXUAN GAO；XIAOQIAO HUANG: "Hourly forecasting of solar irradiance based on CEEMDAN and multi-strategy CNN-LSTM neural networks", 《RENEWABLE ENERGY》 *
李兢等: "考虑时空相关性的光辐照度序列估计方法", 《电力系统自动化》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116167465A (en) * 2023-04-23 2023-05-26 杭州经纬信息技术股份有限公司 Solar irradiance prediction method based on multivariate time series ensemble learning
CN116488158A (en) * 2023-04-24 2023-07-25 国能日新科技股份有限公司 Method and device for predicting medium-long-term electric quantity of wind power based on transfer learning
CN116488158B (en) * 2023-04-24 2023-09-15 国能日新科技股份有限公司 Method and device for predicting medium-long-term electric quantity of wind power based on transfer learning
CN116451598A (en) * 2023-06-20 2023-07-18 杭州经纬信息技术股份有限公司 Solar irradiance prediction method based on denoising diffusion probability model
CN116451598B (en) * 2023-06-20 2023-09-05 杭州经纬信息技术股份有限公司 Solar Irradiance Prediction Method Based on Denoising Diffusion Probability Model

Also Published As

Publication number Publication date
CN114201924B (en) 2022-06-10

Similar Documents

Publication Publication Date Title
CN114201924B (en) Solar irradiance prediction method and system based on transfer learning
CN110807554B (en) Generation method and system based on wind power/photovoltaic classical scene set
CN111582551A (en) Method and system for predicting short-term wind speed of wind power plant and electronic equipment
US20210125200A1 (en) Method and system for predicting medium-long term water demand of water supply network
Liu et al. Heating load forecasting for combined heat and power plants via strand-based LSTM
CN112508244B (en) Multi-element load prediction method for user-level comprehensive energy system
CN112508246B (en) Photovoltaic power generation power prediction method based on similar days
Jiang et al. Day‐ahead renewable scenario forecasts based on generative adversarial networks
Xiao et al. A novel method based on time series ensemble model for hourly photovoltaic power prediction
CN114021830A (en) Multi-time-range wind speed prediction method based on CNN-LSTM
JP2015148934A (en) Power generation amount prediction device and power generation amount prediction method
CN115146842A (en) Multivariate time series trend prediction method and system based on deep learning
Wibawa et al. Long Short-Term Memory to Predict Unique Visitors of an Electronic Journal
Jayashankara et al. A novel approach for short-term energy forecasting in smart buildings
CN116707331B (en) Inverter output voltage high-precision adjusting method and system based on model prediction
Alharbi et al. Short-term wind speed and temperature forecasting model based on gated recurrent unit neural networks
Kumar et al. Forecasting indoor temperature for smart buildings with ARIMA, SARIMAX, and LSTM: A fusion approach
CN110555566B (en) B-spline quantile regression-based photoelectric probability density prediction method
Brabec et al. Statistical modeling of energy production by photovoltaic farms
CN112232714B (en) Deep learning-based risk assessment method for distribution network under incomplete structural parameters
CN116029419A (en) Deep learning-based long-term new energy daily average generation power prediction method and system
Bantupalli et al. Wind Speed forecasting using empirical mode decomposition with ANN and ARIMA models
CN112070320B (en) Ultra-short-term wind power prediction method and system based on dynamic harmonic regression
CN114970369A (en) Multi-task learning-based multi-element load prediction method for comprehensive energy system
Wan et al. Turbine location wind speed forecast using convolutional neural network

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