CN113642255A - Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network - Google Patents

Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network Download PDF

Info

Publication number
CN113642255A
CN113642255A CN202111021128.2A CN202111021128A CN113642255A CN 113642255 A CN113642255 A CN 113642255A CN 202111021128 A CN202111021128 A CN 202111021128A CN 113642255 A CN113642255 A CN 113642255A
Authority
CN
China
Prior art keywords
layer
output
gru
power generation
photovoltaic power
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
Application number
CN202111021128.2A
Other languages
Chinese (zh)
Inventor
马武彬
吴亚辉
邓苏
周浩浩
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.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
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 National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202111021128.2A priority Critical patent/CN113642255A/en
Publication of CN113642255A publication Critical patent/CN113642255A/en
Pending legal-status Critical Current

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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • 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
    • G06F2119/06—Power analysis or power optimisation
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • General Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Marketing (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • General Business, Economics & Management (AREA)
  • Computing Systems (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Quality & Reliability (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Operations Research (AREA)

Abstract

The invention discloses a photovoltaic power generation power prediction method based on a multi-scale convolution cyclic neural network, which comprises the following steps of: constructing a photovoltaic power generation power prediction model based on a multi-scale convolution cyclic neural network; training the photovoltaic power generation power prediction model by using training set data; and inputting the test set data into the trained photovoltaic power generation power prediction model, and calculating to obtain a predicted value of the photovoltaic power generation power. The method introduces the multi-scale convolutional layer into the recurrent neural network, and attention mechanisms are distributed from different scales, so that the model can acquire historical information from different scales; the bidirectional GRU layer can more fully acquire context information of sequence data, the whole model adopts a convolution structure to fuse identification outputs of attention mechanisms of different scales, and the convolution connection is used for screening and identifying the outputs of different scales, so that the prediction of the photovoltaic power generation power can be more accurately acquired.

Description

Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network
Technical Field
The invention belongs to the technical field of photovoltaic power generation power prediction, and particularly relates to a photovoltaic power generation power prediction method based on a multi-scale convolution cyclic neural network.
Background
The output of the photovoltaic power generation power has obvious intermittent fluctuation characteristics, and the large-scale photovoltaic power generation access can bring certain impact to the safe and stable operation of a power grid. With the increase of the proportion of renewable energy sources such as wind power, photovoltaic and the like in various regions, the phenomena of wind abandonment and light abandonment are further increased.
The photovoltaic power generation power prediction is one of key technologies for solving the problem, and the development of a photovoltaic power station power generation power prediction method and system research has important academic and application values. Therefore, how to accurately predict the power of the photovoltaic power station becomes a research hotspot in recent years, and is popular among domestic and foreign scholars.
The prediction accuracy of the photovoltaic power generation power is still insufficient at present. The traditional machine learning method is difficult to capture long-term dependency relationship, and the time sequence importance among all factors is not well acquired. Recently, learners adopt a deep learning method to predict the photovoltaic power generation power, so that a good effect is achieved. However, both the conventional machine learning method and the deep learning method which is popular in recent years do not capture the correlation characteristics between the elements from the time sequence, and the prediction accuracy is not ideal.
Disclosure of Invention
In view of the above, in order to solve the problem of accurate prediction of photovoltaic power generation power, the present invention aims to provide a method for predicting photovoltaic power generation power based on a multi-scale convolution cyclic neural network, which provides a method with higher prediction accuracy and precision by using meteorological data such as irradiation intensity, temperature, humidity, air pressure, altitude, etc., so as to predict the photovoltaic power generation power.
Based on the purpose, the photovoltaic power generation power prediction method based on the multi-scale convolution cyclic neural network comprises the following steps:
step 1, constructing a photovoltaic power generation power prediction model based on a multi-scale convolution cyclic neural network;
step 2, training the photovoltaic power generation power prediction model by using training set data, wherein the training set data comprises influence factor data and known photovoltaic power generation power data;
and 3, inputting the test set data into the trained photovoltaic power generation power prediction model, and calculating to obtain a predicted value of the photovoltaic power generation power.
Specifically, the photovoltaic power generation power prediction model comprises a first convolution layer, a first bidirectional GRU layer, a first multi-scale convolution layer, a second bidirectional GRU layer, a second multi-scale convolution layer, a first full-connection layer and a second full-connection layer, wherein the layers are sequentially connected, the output of the first convolution layer and the output of the first bidirectional GRU layer are connected and then simultaneously used as the input of the first multi-scale convolution layer and the second multi-scale convolution layer, the bidirectional GRU layer is formed by connecting a forward GRU model and a backward GRU model in parallel to form a bidirectional structure, the bidirectional GRU layer outputs two combined GRU signals, the output layer of the first full connection layer is 100, the output layer of the second full connection layer is 1, the input of the first convolution layer in the photovoltaic power generation power prediction model is an influence factor data sequence, and the output of the second full-connection layer is a photovoltaic power generation power value.
Specifically, the photovoltaic power generation power prediction model is
Figure RE-GDA0003267066530000021
x0,...,xTTo influence the factor sequence data, (y)0,...,yK),K<T is a known photovoltaic power generation power value, (y)K+1,...,yT) In order to require a predicted photovoltaic power generation power value,
Figure RE-GDA0003267066530000022
for the corresponding estimated value, the input is x0,...,xT,y0,...,yKAnd variables are sequentially input into the photovoltaic power generation power prediction model to start training, the loss function adopts standard normalized MSE, and the activation function adopts Relu function.
Specifically, the analytical expression of the photovoltaic power generation power prediction model is as follows:
Figure RE-GDA0003267066530000023
Figure RE-GDA0003267066530000024
Figure RE-GDA0003267066530000025
Figure RE-GDA0003267066530000026
Figure RE-GDA0003267066530000031
Figure RE-GDA0003267066530000032
C4 t＝η2([C2 t,C3 t])
Figure RE-GDA0003267066530000033
wherein x istIs an input to the model at time t, η1(. eta.) and eta2(. is) two convolution operations, [, ]]For merge join operations, mutiscalaconv (·, Scale1) and mutiscalaconv (·, Scale2) are two multi-Scale convolution operations with scales Scale1 and Scale2, respectively, and the specific process of fusion convolution is as follows:
first convolution layer η1(xt) Accepting sequence data xtIs transported byIn and out are
Figure RE-GDA0003267066530000034
Simultaneously as inputs to the first multi-scale convolutional layer and the second multi-scale convolutional layer;
Figure RE-GDA0003267066530000035
is the output of the first bi-directional GRU layer,
Figure RE-GDA0003267066530000036
indicating the output to be GRU in forward direction
Figure RE-GDA0003267066530000037
And backward GRU output
Figure RE-GDA0003267066530000038
Carrying out merging connection;
Figure RE-GDA0003267066530000039
is to multiply the first bidirectional GRU layer by a weight vector
Figure RE-GDA00032670665300000310
And adding the offset vector
Figure RE-GDA00032670665300000311
The result of (1);
will be provided with
Figure RE-GDA00032670665300000312
And η1(xt) Output of (2)
Figure RE-GDA00032670665300000326
Are combined into Pt 1As input to the first multi-scale convolutional layer;
Figure RE-GDA00032670665300000314
is the output of the first multi-Scale convolution layer with Scale1Out, connected to a second bidirectional GRU layer;
Figure RE-GDA00032670665300000315
is the output of the second bidirectional GRU layer,
Figure RE-GDA00032670665300000316
indicating the output of the GRU in forward direction therein
Figure RE-GDA00032670665300000317
And backward GRU output
Figure RE-GDA00032670665300000318
Carrying out merging connection;
Figure RE-GDA00032670665300000319
is to multiply the second bidirectional GRU layer by a weight vector
Figure RE-GDA00032670665300000320
And adding the offset vector
Figure RE-GDA00032670665300000321
The result of (1);
by analogy, the expression is obtained
Figure RE-GDA00032670665300000322
By a convolution operation with Scale2, for [ C2 t,C3 t]Extracting to reserve the more important scale information of the target and obtain output
Figure RE-GDA00032670665300000323
Then obtaining output O through full connection operationt；
Wherein,
Figure RE-GDA00032670665300000324
and
Figure RE-GDA00032670665300000325
all are obtained by learning and training.
Specifically, the bidirectional GRU layer is a bidirectional structure formed by connecting two GRU models of a forward GRU and a backward GRU, and the first layer of the forward GRU is forgotten to be output by a gate: f. of1 t＝σ(W1 f[H1 t-1,x”t]+B1 f)，σ(x)＝1/(1+e-x) In the forward GRU update gate, the first output is: z is a radical of1 t＝σ(W1 z[H1 t-1,x”t]+B1 z) And the second output is:
Figure RE-GDA0003267066530000041
similarly, the corresponding first output to the GRU is: z is a radical of2 t＝σ(W2 z[H2 t-1,x”t]+B2 z) And a second output:
Figure RE-GDA0003267066530000042
intermediate output of forward GRU
Figure RE-GDA0003267066530000043
And backward GRU intermediate output
Figure RE-GDA0003267066530000044
Obtaining an output by an aggregation operation on the intermediate output
Figure RE-GDA0003267066530000045
Figure RE-GDA0003267066530000046
Indicating the output to be GRU in forward direction
Figure RE-GDA00032670665300000411
And backward GRU output
Figure RE-GDA0003267066530000047
Performing merged connection as output of bidirectional GRU layer
Figure RE-GDA0003267066530000048
For the input of the bidirectional GRU layer, [ W ]1 f,B1 f]， [W1 Z,B1 Z]，[W1 h,B1 h]For forward GRU model parameters, [ W ]2 f,B2 f],[W2 Z,B2 Z]，[W2 h,B2 h]For the inverse GRU model parameters, [ W ]12 o,B12 o]Are output layer parameters.
Figure RE-GDA00032670665300000410
The output of the first bidirectional GRU layer, correspondingly,
Figure RE-GDA0003267066530000049
is the output of the second bidirectional GRU layer.
Preferably, the convolutional layer is a 1-dimensional convolutional network.
Specifically, the influencing factor data comprises irradiation intensity, temperature, humidity and air pressure.
The photovoltaic power generation power prediction model in the method mainly comprises a multi-scale convolution layer, a bidirectional GRU layer and a full-connection layer, the multi-scale convolution layer is introduced into a recurrent neural network, a plurality of improved attention units are connected in series, attention mechanisms are distributed on different scales, therefore, the model can collect historical information from different scales, distinguish the influence of different input elements on a prediction result, the bidirectional GRU layer can more fully acquire the context information of sequence data on the basis of GRUs, the whole model adopts a convolution structure to fuse the recognition output of attention mechanisms of different scales, and the output is screened and identified by different scales through convolution connection, so that the scale information which is more important to the target can be reserved, and the prediction of the photovoltaic power generation power obtains better accuracy.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a data processing flow diagram of the method of the present invention;
FIG. 3 is a schematic diagram of a 1-dimensional convolutional network in the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, a photovoltaic power generation power prediction method based on a multi-scale convolution cyclic neural network is provided, which includes the following steps:
step 1, constructing a photovoltaic power generation power prediction model based on a multi-scale convolution cyclic neural network;
step 2, training the photovoltaic power generation power prediction model by using training set data, wherein the training set data comprises influence factor data and known photovoltaic power generation power data;
and 3, inputting the test set data into the trained photovoltaic power generation power prediction model, and calculating to obtain a predicted value of the photovoltaic power generation power.
Specifically, the photovoltaic power generation power prediction model comprises a first convolution layer, a first bidirectional GRU layer, a first multi-scale convolution layer, a second bidirectional GRU layer, a second multi-scale convolution layer, a first full-connection layer and a second full-connection layer, wherein the layers are sequentially connected, the output of the first convolution layer and the output of the first bidirectional GRU layer are connected and then simultaneously used as the input of the first multi-scale convolution layer and the second multi-scale convolution layer, the bidirectional GRU layer is formed by connecting a forward GRU model and a backward GRU model in parallel to form a bidirectional structure, the bidirectional GRU layer outputs two combined GRU signals, the output layer of the first full connection layer is 100, the output layer of the second full connection layer is 1, the input of the first convolution layer in the photovoltaic power generation power prediction model is an influence factor data sequence, and the output of the full-connection layer is a photovoltaic power generation power value.
Specifically, the photovoltaic power generation power prediction model is
Figure RE-GDA0003267066530000051
x0,...,xTTo influence the factor sequence data, (y)0,...,yK),K<T is a known photovoltaic power generation power value, (y)K+1,...,yT) In order to require a predicted photovoltaic power generation power value,
Figure RE-GDA0003267066530000061
for the corresponding estimated value, the input is x0,...,xT,y0,...,yKAnd variables are sequentially input into the photovoltaic power generation power prediction model to start training, the loss function adopts standard normalized MSE, and the activation function adopts Relu function.
Specifically, the analytical expression of the photovoltaic power generation power prediction model is as follows:
Figure RE-GDA0003267066530000062
Figure RE-GDA0003267066530000063
Figure RE-GDA0003267066530000064
Figure RE-GDA0003267066530000065
Figure RE-GDA0003267066530000066
Figure RE-GDA0003267066530000067
C4 t＝η2([C2 t,C3 t])
Figure RE-GDA0003267066530000068
wherein x istIs an input to the model at time t, η1(. eta.) and eta2(. is) two convolution operations, [, ]]For merge join operations, mutiscalaconv (·, Scale1) and mutiscalaconv (·, Scale2) are two multi-Scale convolution operations with scales Scale1 and Scale2, respectively, and the specific process of fusion convolution is as follows:
first convolution layer η1(xt) Accepting sequence data xtInput and output of
Figure RE-GDA0003267066530000069
Simultaneously as inputs to the first multi-scale convolutional layer and the second multi-scale convolutional layer;
Figure RE-GDA00032670665300000610
is the output of the first bi-directional GRU layer,
Figure RE-GDA00032670665300000611
indicating the output to be GRU in forward direction
Figure RE-GDA00032670665300000612
And backward GRU output
Figure RE-GDA00032670665300000621
Carrying out merging connection;
Figure RE-GDA00032670665300000613
is to multiply the first bidirectional GRU layer by a weight vector
Figure RE-GDA00032670665300000614
And adding the offset vector
Figure RE-GDA00032670665300000615
The result of (1);
will be provided with
Figure RE-GDA00032670665300000616
And η1(xt) Output of (2)
Figure RE-GDA00032670665300000617
Are combined into Pt 1As input to the first multi-scale convolutional layer;
Figure RE-GDA00032670665300000618
is the output of the first multi-Scale convolutional layer with Scale1, connected to the second bidirectional GRU layer;
Figure RE-GDA00032670665300000619
is the output of the second bidirectional GRU layer,
Figure RE-GDA00032670665300000620
indicating the output of the GRU in forward direction therein
Figure RE-GDA0003267066530000071
And backward GRU output
Figure RE-GDA0003267066530000072
Carrying out merging connection;
Figure RE-GDA0003267066530000073
is to multiply the second bidirectional GRU layer by a weight vector
Figure RE-GDA0003267066530000074
And adding the offset vector
Figure RE-GDA0003267066530000075
The result of (1);
by analogy, the expression is obtained
Figure RE-GDA0003267066530000076
By a convolution operation with Scale2, for [ C2 t,C3 t]Extracting to reserve the more important scale information of the target and obtain output
Figure RE-GDA0003267066530000077
Then obtaining output O through full connection operationt(ii) a The data processing flow of the method of the invention is shown in FIG. 2;
wherein,
Figure RE-GDA0003267066530000078
and
Figure RE-GDA0003267066530000079
all are obtained by learning and training.
Specifically, the bidirectional GRU layer is a bidirectional structure formed by connecting two GRU models of a forward GRU and a backward GRU, and the first layer of the forward GRU is forgotten to be output by a gate: f. of1 t＝σ(W1 f[H1 t-1,x”t]+B1 f)，σ(x)＝1/(1+e-x) In the forward GRU update gate, the first output is: z is a radical of1 t＝σ(W1 z[H1 t-1,x”t]+B1 z) And the second output is:
Figure RE-GDA00032670665300000710
similarly, the corresponding first output to the GRU is: z is a radical of2 t＝σ(W2 z[H2 t-1,x”t]+B2 z) And a second output:
Figure RE-GDA00032670665300000711
intermediate output of forward GRU
Figure RE-GDA00032670665300000712
And an intermediate output H to the GRU2 t＝(1-z2 t)·h2 i-1+h2 t·z2 tObtaining an output by an aggregation operation on the intermediate output
Figure RE-GDA00032670665300000713
Figure RE-GDA00032670665300000714
Indicating the output to be GRU in forward direction
Figure RE-GDA00032670665300000715
And backward GRU output
Figure RE-GDA00032670665300000716
Performing merged connection as output of bidirectional GRU layer
Figure RE-GDA00032670665300000717
x”tFor the input of the bidirectional GRU layer, [ W ]1 f,B1 f]， [W1 Z,B1 Z]，[W1 h,B1 h]For forward GRU model parameters, [ W ]2 f,B2 f],[W2 Z,B2 Z]，[W2 h,B2 h]For the inverse GRU model parameters, [ W ]12 o,B12 o]Are output layer parameters.
Figure RE-GDA00032670665300000718
The output of the first bidirectional GRU layer, correspondingly,
Figure RE-GDA00032670665300000719
is the output of the second bidirectional GRU layer.
Preferably, the convolutional layer is a 1-dimensional convolutional network.
Specifically, the influencing factor data comprises irradiation intensity, temperature, humidity and air pressure.
Preferably, the convolutional layer is a 1-dimensional convolutional network. Convolutional neural networks generally include 1-dimensional convolution, 2-dimensional convolution, and 3-dimensional convolution networks. The one-dimensional convolution network is mainly used for sequence data such as audio data, equipment maintenance sampling data and the like, the two-dimensional convolution is mainly used for image processing such as image classification, target recognition, image segmentation and the like, and the three-dimensional convolution network is mainly used for video processing such as medical image video, motion detection and the like. In this embodiment, the time series data is mainly analyzed, and a 1-dimensional convolution network result is adopted. A typical 1-dimensional convolutional network results are shown in fig. 3. This includes a one-dimensional convolution kernel vector with a filter size k of 4. Convolution factors (convolution factors) d is 1.
For the element s that currently needs to be convolved, the mathematical expression of the one-dimensional convolution operation is:
Figure RE-GDA0003267066530000081
wherein f (i) represents a convolution kernel, Xs-d·iIndicating that sample values at interval d are taken forward.
In order to better show the details of the present embodiment, the present embodiment provides an experimental effect description. The data adopted by the invention is actually measured data from a certain foreign photovoltaic power generation company. Data was collected once every hour and ten hours per day. The experimental data of this example included 3000 sets of data for 300 days, of which 2900 sets were subjected to classification training, 100 sets of data were used as test data,
the experimental background used in this example is: the computer is mainly configured as follows: pentium (R) Dual-core 3.06 CPU, 8G RAM memory.
And (3) effect evaluation: parameters adopted for performance evaluation of the method in this embodiment include RMSE, MAE, MAPE, and CC:
RMSE (Root Mean Square Error) is calculated as:
Figure RE-GDA0003267066530000082
MAE (Mean absolute Error) is calculated as:
Figure RE-GDA0003267066530000083
MAPE (Mean absolute percent Error) was calculated as:
Figure RE-GDA0003267066530000091
r2 (coeffient of Determination), determining the coefficient by the following calculation method:
Figure RE-GDA0003267066530000092
it should be noted that RMSE, MAE and MAPE are measures of prediction errors, and the smaller the value is, the more accurate the value is, and the R2 parameter represents the coefficient for determining the number of two sequences, and the larger the value is, the more relevant the two sequence data is, the better the prediction effect is.
Experiments are carried out on the data, and the prediction accuracy of the photovoltaic power generation power can be predicted as shown in table 1.
Table 1: predicted result values of different algorithms
Figure RE-GDA0003267066530000093
The result shows that the method has better precision compared with other recurrent neural network models.
According to the invention content and the embodiment, the photovoltaic power generation power prediction model in the method mainly comprises a multi-scale convolution layer, a bidirectional GRU layer and a full connection layer, the multi-scale convolution layer is introduced into a recurrent neural network, a plurality of improved attention units are connected in a series connection mode and a jump connection mode, attention mechanisms are distributed on different scales, therefore, the model can collect historical information from different scales, distinguish the influence of different input elements on a prediction result, the bidirectional GRU layer can more fully acquire the context information of sequence data on the basis of GRUs, the whole model adopts a convolution structure to fuse the recognition output of attention mechanisms of different scales, and the output is screened and identified by different scales through convolution connection, so that the scale information which is more important to the target can be reserved, and the prediction of the photovoltaic power generation power obtains better accuracy.

Claims (6)

1. The photovoltaic power generation power prediction method based on the multi-scale convolution cyclic neural network is characterized by comprising the following steps of:
step 1, constructing a photovoltaic power generation power prediction model based on a multi-scale convolution cyclic neural network;
step 2, training the photovoltaic power generation power prediction model by using training set data, wherein the training set data comprises influence factor data and known photovoltaic power generation power data;
step 3, inputting the test set data into the trained photovoltaic power generation power prediction model, and calculating to obtain a predicted value of the photovoltaic power generation power;
the photovoltaic power generation power prediction model comprises a first convolution layer, a first bidirectional GRU layer, a first multi-scale convolution layer, a second bidirectional GRU layer, a second multi-scale convolution layer, a first full-connection layer and a second full-connection layer, wherein the first convolution layer, the second bidirectional GRU layer, the second multi-scale convolution layer, the first full-connection layer and the second full-connection layer are sequentially connected, the output of the first convolution layer and the output of the first bidirectional GRU layer are connected and then simultaneously used as the input of the first multi-scale convolution layer and the input of the second multi-scale convolution layer, the bidirectional GRU layer is in a bidirectional structure formed by connecting a forward GRU model and a backward GRU model in parallel, the bidirectional GRU layer outputs two combined GRU signals, the output layer of the first full-connection layer is 100, the output layer of the second full-connection layer is 1, the input of the first convolution in the photovoltaic power generation power prediction model is an influence factor data sequence, and the output of the second full-connection layer is a photovoltaic power generation value.
2. The method according to claim 1, wherein the photovoltaic power generation power prediction model is
Figure RE-FDA0003267066520000011
x0,...,xTTo influence the factor sequence data, (y)0,...,yK),K<T is a known photovoltaic power generation power value, (y)K+1,...,yT) In order to require a predicted photovoltaic power generation power value,
Figure RE-FDA0003267066520000012
for the corresponding estimated value, the input is x0,...,xT,y0,...,yKAnd variables are sequentially input into the photovoltaic power generation power prediction model to start training, the loss function adopts standard normalized MSE, and the activation function adopts Relu function.
3. The method according to claim 1, wherein the photovoltaic power generation prediction model has the following analytical expression:
Figure RE-FDA0003267066520000021
Figure RE-FDA0003267066520000022
Figure RE-FDA0003267066520000023
Figure RE-FDA0003267066520000024
Figure RE-FDA0003267066520000025
Figure RE-FDA0003267066520000026
C4 t＝η2([C2 t,C3 t])
Figure RE-FDA0003267066520000027
wherein x istIs an input to the model at time t, η1(. eta.) and eta2(. is) two convolution operations, [, ]]For merge join operations, mutiscalaconv (·, Scale1) and mutiscalaconv (·, Scale2) are two multi-Scale convolution operations with scales Scale1 and Scale2, respectively, and the specific process of fusion convolution is as follows:
first convolution layer η1(xt) Accepting sequence data xtInput and output of
Figure RE-FDA0003267066520000028
Simultaneously as inputs to the first multi-scale convolutional layer and the second multi-scale convolutional layer;
Figure RE-FDA0003267066520000029
is firstThe output of the bi-directional GRU layer,
Figure RE-FDA00032670665200000210
indicating the output to be GRU in forward direction
Figure RE-FDA00032670665200000211
And backward GRU output
Figure RE-FDA00032670665200000212
Carrying out merging connection;
Figure RE-FDA00032670665200000213
is to multiply the first bidirectional GRU layer by a weight vector
Figure RE-FDA00032670665200000214
And adding the offset vector
Figure RE-FDA00032670665200000215
The result of (1);
will be provided with
Figure RE-FDA00032670665200000216
And η1(xt) Output of (2)
Figure RE-FDA00032670665200000217
Are combined into
Figure RE-FDA00032670665200000218
As an input to the first multi-scale convolutional layer;
Figure RE-FDA00032670665200000219
is the output of the first multi-Scale convolutional layer with Scale1, connected to the second bidirectional GRU layer;
Figure RE-FDA00032670665200000220
is the output of the second bidirectional GRU layer,
Figure RE-FDA00032670665200000221
indicating the output of the GRU in forward direction therein
Figure RE-FDA00032670665200000222
And backward GRU output
Figure RE-FDA00032670665200000223
Carrying out merging connection;
Figure RE-FDA00032670665200000224
is to multiply the second bidirectional GRU layer by a weight vector
Figure RE-FDA00032670665200000225
And adding the offset vector
Figure RE-FDA00032670665200000226
The result of (1);
by analogy, the expression is obtained
Figure RE-FDA00032670665200000227
By a convolution operation with Scale2, for [ C2 t,C3 t]Extracting to reserve the more important scale information of the target and obtain output
Figure RE-FDA0003267066520000031
Then obtaining output O through full connection operationt；
Wherein,
Figure RE-FDA0003267066520000032
and
Figure RE-FDA0003267066520000033
all are obtained by learning and training.
4. The multi-scale convolution cyclic neural network-based photovoltaic power generation power prediction method according to claim 1 or 3, wherein the bidirectional GRU layer is a bidirectional structure formed by connecting two GRU models, namely a forward GRU model and a backward GRU model, and a first layer in the forward GRU model is output through a forgetting gate: f. of1 t＝σ(W1 f[H1 t-1,x”t]+B1 f)，σ(x)＝1/(1+e-x) In the forward GRU update gate, the first output is: z is a radical of1 t＝σ(W1 z[H1 t-1,x”t]+B1 z) And the second output is:
Figure RE-FDA0003267066520000034
similarly, the corresponding first output to the GRU is: z is a radical of2 t＝σ(W2 z[H2 t-1,x”t]+B2 z) And a second output:
Figure RE-FDA0003267066520000035
intermediate output of forward GRU
Figure RE-FDA0003267066520000036
And backward GRU intermediate output
Figure RE-FDA0003267066520000037
Obtaining an output by an aggregation operation on the intermediate output
Figure RE-FDA0003267066520000038
Indicating the output to be GRU in forward direction
Figure RE-FDA0003267066520000039
And backward GRU output
Figure RE-FDA00032670665200000310
Performing merged connection as output of bidirectional GRU layer
Figure RE-FDA00032670665200000311
x”tFor the input of the bidirectional GRU layer, [ W ]1 f,B1 f]，[W1 Z,B1 Z]，[W1 h,B1 h]For forward GRU model parameters, [ W ]2 f,B2 f],[W2 Z,B2 Z]，[W2 h,B2 h]For the inverse GRU model parameters, [ W ]12 o,B12 o]Are output layer parameters.
Figure RE-FDA00032670665200000312
The output of the first bidirectional GRU layer, correspondingly,
Figure RE-FDA00032670665200000313
is the output of the second bidirectional GRU layer.
5. The multi-scale convolution cyclic neural network-based photovoltaic power generation power prediction method according to claim 4, wherein the convolution layer is a 1-dimensional convolution network.
6. The method according to claim 1, wherein the data of the influencing factors comprise irradiation intensity, temperature, humidity and air pressure.
CN202111021128.2A 2021-09-01 2021-09-01 Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network Pending CN113642255A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111021128.2A CN113642255A (en) 2021-09-01 2021-09-01 Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111021128.2A CN113642255A (en) 2021-09-01 2021-09-01 Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network

Publications (1)

Publication Number Publication Date
CN113642255A true CN113642255A (en) 2021-11-12

Family

ID=78424841

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111021128.2A Pending CN113642255A (en) 2021-09-01 2021-09-01 Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network

Country Status (1)

Country Link
CN (1) CN113642255A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114123200A (en) * 2022-01-24 2022-03-01 国网山西省电力公司晋城供电公司 Photovoltaic power station dynamic modeling method based on data driving and storage device
CN116205666A (en) * 2022-12-22 2023-06-02 国网湖北省电力有限公司宜昌供电公司 RACNet-based multivariable power load prediction method
CN117556715A (en) * 2024-01-12 2024-02-13 湖南大学 Method and system for analyzing degradation of intelligent ammeter in typical environment based on information fusion

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114123200A (en) * 2022-01-24 2022-03-01 国网山西省电力公司晋城供电公司 Photovoltaic power station dynamic modeling method based on data driving and storage device
CN116205666A (en) * 2022-12-22 2023-06-02 国网湖北省电力有限公司宜昌供电公司 RACNet-based multivariable power load prediction method
CN117556715A (en) * 2024-01-12 2024-02-13 湖南大学 Method and system for analyzing degradation of intelligent ammeter in typical environment based on information fusion
CN117556715B (en) * 2024-01-12 2024-03-26 湖南大学 Method and system for analyzing degradation of intelligent ammeter in typical environment based on information fusion

Similar Documents

Publication Publication Date Title
CN113642255A (en) Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network
CN111784041B (en) Wind power prediction method and system based on graph convolution neural network
CN109766583A (en) Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique
CN110807554B (en) Generation method and system based on wind power/photovoltaic classical scene set
CN107562992B (en) Photovoltaic array maximum power tracking method based on SVM and particle swarm algorithm
CN110880369A (en) Gas marker detection method based on radial basis function neural network and application
CN116842459B (en) Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning
CN111814956A (en) Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
Li et al. Deep spatio-temporal wind power forecasting
CN114462718A (en) CNN-GRU wind power prediction method based on time sliding window
CN113642254A (en) Wind power generation power prediction method based on multi-scale attention mechanism
CN109617526A (en) A method of photovoltaic power generation array fault diagnosis and classification based on wavelet multiresolution analysis and SVM
CN115185937A (en) SA-GAN architecture-based time sequence anomaly detection method
CN115983087A (en) Method for detecting time sequence data abnormity by combining attention mechanism and LSTM and terminal
Tzoumpas et al. A data filling methodology for time series based on CNN and (Bi) LSTM neural networks
CN116245259B (en) Photovoltaic power generation prediction method and device based on depth feature selection and electronic equipment
CN117458480A (en) Photovoltaic power generation power short-term prediction method and system based on improved LOF
CN116760017A (en) Prediction method for photovoltaic power generation
CN117033923A (en) Method and system for predicting crime quantity based on interpretable machine learning
CN115907178A (en) Clean ecosystem CO 2 Method for predicting exchange amount
CN114065335A (en) Building energy consumption prediction method based on multi-scale convolution cyclic neural network
CN112348700B (en) Line capacity prediction method combining SOM clustering and IFOU equation
CN114219118A (en) Method and system for predicting service life of intelligent electric meter based on D-S evidence theory
Chen Brain Tumor Prediction with LSTM Method

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