CN109447843A - A kind of photovoltaic power generation output forecasting method and device - Google Patents
A kind of photovoltaic power generation output forecasting method and device Download PDFInfo
- Publication number
- CN109447843A CN109447843A CN201811229686.6A CN201811229686A CN109447843A CN 109447843 A CN109447843 A CN 109447843A CN 201811229686 A CN201811229686 A CN 201811229686A CN 109447843 A CN109447843 A CN 109447843A
- Authority
- CN
- China
- Prior art keywords
- photovoltaic power
- moment
- predicted
- day
- power generating
- 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
Links
- 238000010248 power generation Methods 0.000 title claims abstract description 16
- 238000013277 forecasting method Methods 0.000 title claims abstract description 9
- 238000000034 method Methods 0.000 claims abstract description 39
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000003062 neural network model Methods 0.000 abstract 1
- 239000011159 matrix material Substances 0.000 description 27
- 238000012549 training Methods 0.000 description 14
- 238000013528 artificial neural network Methods 0.000 description 13
- 230000008569 process Effects 0.000 description 11
- 210000002569 neuron Anatomy 0.000 description 9
- 230000006870 function Effects 0.000 description 7
- 238000012545 processing Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 210000005036 nerve Anatomy 0.000 description 5
- 230000005611 electricity Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000012546 transfer Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000012417 linear regression Methods 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- 230000007704 transition Effects 0.000 description 3
- 241000529895 Stercorarius Species 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005183 dynamical system Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000002969 morbid Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000013082 photovoltaic technology Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000005309 stochastic process Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000000714 time series forecasting Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
Landscapes
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of photovoltaic power generation output forecasting method and devices, this method comprises: establishing the ESN prediction model at moment to be predicted；Obtain the highest temperature in nearest preset time period, temperature on average, the photovoltaic power generating value for obtaining the historical juncture, at the time of the historical juncture value with the moment to be predicted at the time of be worth it is identical；Determine the day index of the vertex degree at the moment to be predicted and prediction day where the moment to be predicted；Using the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture, the vertex degree and the day index as the input of the ESN prediction model, to obtain the photovoltaic power generating value at the moment to be predicted.Compared with the existing technology, the present embodiment by establishing ESN prediction model, using temperature, index of type, history photovoltaic power generating value as the input of ESN network, the photovoltaic power generating value of prediction time in prediction one day solves the problems, such as true using the few caused forecasting inaccuracy of sample data when neural network model.
Description
Technical field
The present invention relates to photovoltaic technology field, more particularly to a kind of photovoltaic power output method and device.
Background technique
Under impact of economic globalization, problem of environmental pollution and global energy crisis are two hang-ups to become increasingly conspicuous,
New energy is developed and used, the important measures of energy long run supply are to maintain.As one of the new energy for developing comparative maturity, the sun
Can be with the advantages such as its cleaning, pollution-free, from a wealth of sources, by the extensive concern of society, large-scale photovoltaic, which generates electricity by way of merging two or more grid systems, to be become too
The development trend that sun can generate electricity.Photovoltaic power generation is influenced by factors such as outside weather, self performances, and the fluctuation of each factor makes
It obtains photovoltaic and fluctuation and intermittence occurs, the operation of photovoltaic large-scale grid connection will affect power quality, influence electric system
The operation of safety and stability economy, thus to photovoltaic power output carry out prediction be very it is necessary to.
Currently used photovoltaic power generation output forecasting method has artificial neural network predicted method.Artificial neural network can imitate people
The intelligent processing process of brain has stronger autonomous learning and adaptive ability, learns in photovoltaic power prediction by research
The favor of person.But neural network work when sample data is insufficient it is undesirable, certain weather patterns (such as heavy rain, severe snow,
Heavy rain etc.) sample data volume accumulation it is less, models fitting effect is poor.
Summary of the invention
In order to solve the above technical problems, the embodiment of the invention provides a kind of photovoltaic power generation output forecasting method and device, technology
Scheme is as follows:
A kind of photovoltaic power generation output forecasting method, comprising:
Establish the ESN prediction model at moment to be predicted；
The highest temperature, the temperature on average, the photovoltaic power generating value for obtaining the historical juncture in nearest preset time period are obtained, it is described
At the time of historical juncture value with the moment to be predicted at the time of be worth it is identical；
Determine the day index of the vertex degree at the moment to be predicted and prediction day where the moment to be predicted；
By the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture, the vertex degree and described
The input of day index as the ESN prediction model, to obtain the photovoltaic power generating value at the moment to be predicted.
Preferably, further includes:
Predefine the weather pattern of the prediction day；
According to the photovoltaic power curve and the weather pattern in the nearest preset time period, determine the day index and
The vertex degree.
Preferably, in the nearest preset time period, the corresponding day index of the smallest weather pattern of photovoltaic power generating value is set
It is 1, the corresponding vertex degree of the maximum weather pattern of photovoltaic power generating value sets not 1.
Preferably, further includes:
From 6:00 to 19:00,14 ESN prediction models are established respectively to the prediction day；
Using any integral point moment of the 6:00 into 19:00 as the predetermined time.
Preferably, further includes:
It is modified by photovoltaic power generating value of the Markov Chain prediction model to the moment to be predicted.
A kind of photovoltaic power generation output forecasting device, comprising:
First establishing unit, for establishing the ESN prediction model at moment to be predicted；
Acquiring unit, for obtaining the highest temperature in nearest preset time period, temperature on average, the light for obtaining the historical juncture
Lie prostrate power generating value, at the time of the historical juncture value with the moment to be predicted at the time of be worth it is identical；
First determination unit, for determining the vertex degree at the moment to be predicted and predicting day where the moment to be predicted
Day index；
Computing unit, for by the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture, institute
The input of vertex degree and the day index as the ESN prediction model is stated, to obtain the photovoltaic power output at the moment to be predicted
Value.
Preferably, further includes:
Second determination unit, for predefining the weather pattern of the prediction day；
Third determination unit, for according in the nearest preset time period photovoltaic power curve and the weather class
Type determines the day index and the vertex degree.
Preferably, in the nearest preset time period, the corresponding day index of the smallest weather pattern of photovoltaic power generating value is set
It is 1, the corresponding vertex degree of the maximum weather pattern of photovoltaic power generating value sets not 1.
Preferably, further includes:
Second establishes unit from 6:00 to 19:00, establishes 14 ESN prediction models respectively to the prediction day；
Correspondingly, the first establishing unit is specifically used for:
Using any integral point moment of the 6:00 into 19:00 as the predetermined time.
Preferably, further includes:
Amending unit, for being repaired by photovoltaic power generating value of the Markov Chain prediction model to the moment to be predicted
Just.
Technical solution provided in an embodiment of the present invention, by establishing ESN prediction model, by temperature, index of type, history light
Input of the power generating value as ESN network is lied prostrate, the photovoltaic power generating value of prediction time, solves using neural network mould in prediction one day
The true problem of forecasting inaccuracy caused by sample data is few when type.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will to embodiment or
Attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
Some embodiments of the present invention, for those of ordinary skill in the art, without any creative labor,
It is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of a kind of flow diagram of photovoltaic power generation output forecasting method provided by the embodiment of the present invention；
Fig. 2 is a kind of a kind of structural schematic diagram of photovoltaic power generation output forecasting device provided by the embodiment of the present invention；
Fig. 3 is the network structure of echo state network provided by the embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that the described embodiment is only a part of the embodiment of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, every other reality obtained by those of ordinary skill in the art without making creative efforts
Example is applied, shall fall within the protection scope of the present invention.
Referring to Fig. 1, Fig. 1 is a kind of a kind of implementation process of photovoltaic power generation output forecasting method provided in an embodiment of the present invention
Figure, this method comprises:
Step S101, the ESN prediction model at moment to be predicted is established.
Step S102, the highest temperature in nearest preset time period, temperature on average, the photovoltaic of acquisition historical juncture is obtained to go out
Force value, at the time of the historical juncture value with the moment to be predicted at the time of be worth it is identical.
It can be with meteorological data daily in nearest 1 year of collection photovoltaics power station location, including highest gas in practical application
Temperature, the lowest temperature, weather pattern, daily 24 hours history power generating values in nearly 1 year of collection photovoltaics power station.Intercept 6:00 to 19:
00 power generating value is as photovoltaic power curve.The principal element for influencing electricity generation system output power has intensity of illumination, environment temperature
With the characteristic of photovoltaic plant itself etc., since the data of these factors are not easy to acquire, indirectly using weather pattern, the highest temperature,
Temperature on average, history photovoltaic power generating value are as the factor for influencing photovoltaic power generation.
The climatic factors such as solar radiation, cloud amount, wind speed, the temperature of different type weather be it is different, photovoltaic cell
Output power is also different.For example, fine day and the output power of rainy day differ greatly.Weather pattern is to photovoltaic plant power output result shadow
Sound is very big, these influences are embodied directly in the size of day generated output.Therefore, herein by the processing for going out force data to photovoltaic,
Weather pattern is mapped as to the index of type of a numeric type, the input as echo state network training and prediction.
In order to improve precision of prediction, this paper prediction point-by point, i.e., corresponding one day 6:00 to 19:00 builds together vertical 14 respectively
ESN prediction model.So input of the index of type as model, should reflect the influence that weather pattern contributes to whole day, also want
Embody the contribution in certain moment point to power output.In consideration of it, index of type is defined as a two-dimensional array, including day index and
Vertex degree two parts, i.e. (day index, vertex degree).
Step S103, it determines the vertex degree at the moment to be predicted and predicts referring to day for day where the moment to be predicted
Number.
Daily meteorological data in nearest 1 year of collection photovoltaics power station location, including the highest temperature, the lowest temperature, weather
Type, daily 24 hours history power generating values in nearly 1 year of collection photovoltaics power station.Intercept the power generating value conduct of 6:00 to 19:00
Photovoltaic power curve.The principal element for influencing electricity generation system output power has intensity of illumination, environment temperature and photovoltaic plant itself
Characteristic etc. use weather pattern, the highest temperature, temperature on average, history light indirectly since the data of these factors are not easy to acquire
Power generating value is lied prostrate as the factor for influencing photovoltaic power generation.
The climatic factors such as solar radiation, cloud amount, wind speed, the temperature of different type weather be it is different, photovoltaic cell
Output power is also different.For example, fine day and the output power of rainy day differ greatly.Weather pattern is to photovoltaic plant power output result shadow
Sound is very big, these influences are embodied directly in the size of day generated output.Therefore, herein by the processing for going out force data to photovoltaic,
Weather pattern is mapped as to the index of type of a numeric type, the input as echo state network training and prediction.
In order to improve precision of prediction, this paper prediction point-by point, i.e., corresponding one day 6:00 to 19:00 builds together vertical 14 respectively
ESN prediction model.So input of the index of type as model, should reflect the influence that weather pattern contributes to whole day, also want
Embody the contribution in certain moment point to power output.In consideration of it, index of type is defined as a two-dimensional array, including day index and
Vertex degree two parts, i.e. (day index, vertex degree).
Step S104, by the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture, the point
The input of index and the day index as the ESN prediction model, to obtain the photovoltaic power generating value at the moment to be predicted.
The prediction model based on echo state network that the present embodiment is established, prominent weather pattern is to photovoltaic generation power
It influences, considers this factor emphatically when modeling, increase the weight of day index, it need not be according to weather conditions by model decomposition when prediction
For submodel, it is suitable for various weather patterns, there is stronger applicability and preferable predictive ability.
In order to improve precision of prediction, this paper prediction point-by point, i.e., corresponding one day 6:00 to 19:00 builds together vertical 14 respectively
ESN prediction model.Parameter setting, training, the prediction process of each network model are as follows:
Input layer determines: prediction day first three days mutually power generating value in the same time, index of type, the highest temperature, temperature on average, in advance
Survey the highest temperature, temperature on average, the index of type of day.
Output layer determines: the power generating value at prediction day 6:00 to 19:00 a certain moment.
Preferably, after obtaining the photovoltaic power generating value at moment to be predicted, Markov Chain prediction model can also be passed through
The photovoltaic power generating value at the moment to be predicted is modified.
Markov Chain prediction is the present status and its variation tendency according to certain variables, predicts it in the following a certain spy
The state being likely to occur during fixed is suitble to the statement larger problem of stochastic volatility.Set forth herein Markov residual errors to echo
The characteristics of method that the prediction result of state network is modified is predicted to photovoltaic plant power output, agrees with photovoltaic plant, will
The two has complementary advantages, to obtain more accurate prediction conclusion.
Using lower threshold value in difference locating for the relative error of ESN neural network forecast value and actual value as state demarcation codomain.
In Prediction of Markov, a step of most critical is to seek state transition probability matrix.In the solution procedure of state probability, state
It is classified most important, common method has mean value-mean square deviation staging, clustering methodology and left and right split plot design.The present invention adopts
With mean value-mean square deviation staging.Basic thought: to the application of results Markov mould predicted by ESN echo state network
Type analyzes the fluctuating range and fluctuation development trend of its error, obtains the state transition probability matrix of error, and matrix pair accordingly
Neural network prediction result is modified.
Specific step is as follows:
A, the relative error ξ of ESN network test sample predictions value is calculated；
B, it using lower threshold value in difference locating for the relative error ξ of test sample predicted value as state demarcation codomain, establishes
State demarcation standard；
C, state-transition matrix P is found out according to relative error state(k)；
D, initial state vector V (0) is determined；
E, formula V (k)=V (0) P is shifted according to state(k)Find out the state transfer result of kth step；
F, predicted value is corrected P=P0(1-ξ*), wherein P0It is the prediction result of echo state network, ξ*For locating error
The average value of lower threshold value in state interval.
Technical solution provided in an embodiment of the present invention, by establishing ESN prediction model, by temperature, index of type, history light
Input of the power generating value as ESN network is lied prostrate, the photovoltaic power generating value of prediction time, solves using neural network mould in prediction one day
The true problem of forecasting inaccuracy caused by sample data is few when type.
Algorithm model involved in the present invention is illustrated below:
(1) echo state network model
Jaeger and Haas is in proposition echo state networks (echo state network, ESN) in 2001 and accordingly
Learning algorithm, open brand-new road for the research of recurrent neural network.This method is also referred to as so-called reserve pool
Calculating mode, it introduces the internal network for being referred to as reserve pool, when external list entries enters this internal network, just
Complicated and diversified non-linear state space has been excited wherein, it is then defeated to obtain network by a simple reading network again
Out.It is a difference in that in the training process with the maximum of recurrent neural network, the connection weight inside reserve pool is to immobilize
, adjustment is carried out only for network is read, and greatly reduces trained calculation amount, is in turn avoided most of gradients that are based on and is declined
Learning algorithm the difficult local minimum phenomenon avoided, good modeling accuracy can be obtained.
ESN constitutes random network structure by being randomly disposed the imictron of Large Scale Sparse connection, this is used for
The extensive Recursive Networks of the random partially connected of timing input signal are handled, referred to as " reserve pool " (Reservoirs),
Its structure is as shown in Figure 3.The linear combination that neuron (reserve pool internal element) exports x (n) forms the output signal y of system
(n).The connection weight W of connection weight W and input signal in reserve pool between neuron for neuron in reserve poolinAll
It is randomly generated, and is remained unchanged after generating, do not need to train.Trained connection weight is needed to only have reserve pool defeated to system
Connection weight W outout, training process generally only needs to solve a linear regression problem.
Further analyze the structure and mathematical model of ESN.Consider the network structure [4] of ESN as shown in Figure 1.Assuming that
System has M input unit, N number of interior processing unit (Processing Elements, PEs), i.e., N number of intrinsic nerve
Member, while there is L output unit.The value of moment n input unit, internal state and output unit is respectively as follows:
U (n)=[u1(n), u2(n) ..., uM(n)]T,
X (n)=[x1(n), x2(n) ..., xN(n)]T,
Y (n)=[y1(n), y2(n) ..., yL(n)]T。
From structure, ESN is a kind of recurrent neural network of specific type, and basic thought is using extensive random
The Recursive Networks of connection replace the middle layer in classical neural network, to simplify the training process of network.Echo state network
State equation and output equation can be provided by formula (1):
Wherein W, Win、WbackIt respectively indicates state variable, output and input the connection weight matrix to state variable；WoutTable
Show reserve pool, output and input the connection weight matrix for output,It indicates the bias term of output or can represent to make an uproar
Sound.F=[f1, f2..., fN] indicate intrinsic nerve member activation primitive, it is generally the case that fi(i=1,2 ..., N) takes and does hyperbolic
Tangent function.Indicate output function, under normal circumstances, output layer be it is linear, i.e.,Take identity function.In above various connection weight matrixs, it is connected to the connection weight matrix of reserve pool
Win、W、WbackIt is randomly generated, just immobilizes once generating.And it is connected to each connection weight matrix W of outputoutIt needs according to being
Input, the output data training of system obtain, because being linear relationship between state variable and output, these usual connection weights
It need to only be obtained by solving linear regression problem.W is a sparse connectivity matrix, and degree of rarefication generally takes 1%~5%.ESN
Whether network has the characteristic of echo state extremely important, echo state characteristic refer to input vector before network and
Influence of the original state of reserve pool to future state is smaller and smaller until fading away.In order to guarantee the echo effect of ESN network
It answers, the spectral radius of W must assure that less than 1.
1) selection of reserve pool parameter
The building method of ESN is very simple, but in the specific use process, it is necessary to carry out to some key parameters in network
Empirical selection and adjustment.
A) the scale N of reserve pool
Reserve pool scale refers to that the number of neuron in reserve pool, the selection of reserve pool scale and the number of training sample have
It closes, has larger impact to network performance.Under normal conditions, the scale of reserve pool is bigger, and the dynamical system that ESN can be indicated can
It can be more complicated.For given dynamical system, reserve pool scale is bigger, and ESN description thereof is more accurate.But reserve pool is advised
Mould cannot arbitrarily increase, because if reserve pool scale is excessive may to cause overfitting problem.Over-fitting will lead to model pair
Decline in the generalization ability of test data.Common selection principle is to be stepped up reserve pool scale, until network is for test
Until the processing capacity (such as classification error rate, prediction error etc.) of sample degenerates.
B) connection weight matrix spectral radius SR inside reserve pool
Connection weight spectral radius refers to the spy of the maximum absolute value of connection weight matrix W inside reserve pool inside so-called reserve pool
Value indicative is denoted as λmax.SR is the key parameter of reserve pool, it is generally the case that works as λmaxWhen < 1, ESN could have echo state
Attribute, so that it is guaranteed that the state of network and input are after the sufficiently long time, the influence to network can disappear.For specific
The selection of time series forecasting problem, parameter SR has tremendous influence for ESN performance, adapts in Journal of Sex Research in reserve pool, SR is
It can be optimized as key parameter.
C) reserve pool input unit scale IS
Reserve pool input unit scale parameter IS is exactly that the input signal of reserve pool is connected to reserve pool intrinsic nerve member
A preceding scale factor for needing to be multiplied.Because of neuron activation functions select in reserve pool difference and sample data feature
Input signal is not usually applied directly to reserve pool by difference, but by a scale factor IS, i.e., first to input signal
Carry out certain scaling.
D) the sparse degree SD of reserve pool
The sparse degree SD of reserve pool indicates the connection in reserve pool between neuron.Not all mind in reserve pool
All there is connection relationship between member, only there are connection relationships between partial nerve member, and what parameter SD was indicated is in reserve pool
The percentage of neuron Zhan interconnected total neuron number (N).The parameter can measure vector included in reserve pool
Abundant degree, the abundant degree of vector influences the None-linear approximation ability of network in reserve pool, and the vector of network is abundanter, non-
Linear approximation ability is stronger.
In addition, noticeable further includes the selection of intrinsic nerve member excitation function.Jaeger initially proves echo state
When use linear neuron.Typically, for linear neural metanetwork, can be provided while possessing echo state
Stronger short-term memory.But due to most of systems for needing neural network to be modeled all have it is very strong non-linear, most
Often take more common S-shaped excitation function in practical applications afterwards.
2) training of echo state network
The training process of echo state network is exactly according to given training sample
(u (n), y (n), n=1,2 ..., M) determines the output connection weight matrix W in systemoutProcess.In order to simple
For the sake of, it is assumed here that Wback=0, while being input to output connection weight also assumes that be 0, sample data (u (n), y (n), n=1,
2 ..., M) it is also sometimes referred to as teacher's data.The training process of echo state network can be divided into two stages: sampling
(Sampling) stage and weight computing (Weight Computation) stage.
A) it samples
The original state of sample phase network arbitrarily selected first, but the original state for choosing network under normal conditions is
0, i.e. x (0)=0.Training sample (u (n), n=1,2 ..., M) is by input connection weight Win, sample data y (n) is by feedback
Connection weight WbackIt is respectively added to reserve pool, according to system (1), is sequentially completed the calculating and corresponding output of system mode's
It calculates and collects.The calculating for paying attention to each moment system mode x (n) requires sample data y (n) being written to output unit.
In order to calculate output connection weight matrix, need to collect (sampling) internal state variable since a certain moment.It is assumed herein that when from m
It carves and starts collection system state, and with vector (x1(i), x2(i) ..., xN(i)) (i=m, m+1 ..., M) is that row constitutes matrix
B (M-m+1, N), while corresponding sample data y (n) is also collected, and constitutes a column vector T (M-m+1,1).Here it needs
Illustrate two o'clock:
If system includes the connection weight for being input to output, being output to output, then in the state matrix B of collection system
When, in addition it is also necessary to it collects and outputs and inputs part accordingly；
So that eliminate influence of the arbitrary initial state to system dynamic characteristic, always after a certain moment collect system
The state of system.From this moment, it is believed that system reflection is the mapping relations inputted, between output sample data.
B) weight computing
In order to realize the calculating of weight, need to count according to systematic observation matrix and sample data are collected into sample phase
Calculate output connection weight Wout.Because state variable x (n) and system exportBetween be linear relationship, and need the target realized
It is to utilize network reality outputDesired output y (n) is approached, i.e.,
Namely wish to calculate weight(For matrix WoutElement), the mean square error for meeting system is minimum, i.e.,
Need to solve following optimization problem:
From the viewpoint of mathematics, this is a linear regression problem, and problem, which can be attributed to, asks the inverse matrix of matrix B to ask
It inscribes, matrix B may be morbid state in practical application, and computationally the problem can be further processed as the pseudo- inverse problem of matrix B,
I.e.
(Wout)T=B-1T
So far, ESN network training has been completed.
(2) Markov Chain
Markov Chain, Yin Andelie Markov (1856-1922) are gained the name, and are to have Markov property in mathematics
The discrete time stochastic process of matter.During being somebody's turn to do, in the case where given current knowledge or information, the past is (i.e. current pervious
Historic state) it is in the future (i.e. current later future state) unrelated for prediction.
Markov Chain prediction is the present status and its tendency of changes according to certain variables, predicts it in the following a certain spy
The state being likely to occur during fixed is suitble to the description larger problem of stochastic volatility.
In Prediction of Markov, a step of most critical is to seek state transition probability matrix.In the solution procedure of state probability
In, the classification of state seems most important, and common method has mean value-mean square deviation staging.For sequence x1,…,xn,
Value isMean square deviation is s, after mean value-mean square deviation staging, generally sequence can be divided into 5 grades: Wherein a1,a4
Value value, a in [1.0,1.5]2,a3The value in [0.3,0.6].
According to Markov theory, state EiBecome E by k stepjProbability be
In formula:It is sample from EiTo EjTransfer number；NiFor the total degree that state occurs, then k walks state probability and turns
Moving matrix is
State probability transfer matrix is a n rank square matrix, and there are two features for tool:
Matrix each element nonnegativity；That is matrix
The sum of every row is 1.
The state vector of kth step is calculated with state probability transfer formula
V (k)=V (0) P(k) (5)
Wherein, V (0) is initial state vector.
Referring to Fig. 2, Fig. 2 is a kind of structural schematic diagram of photovoltaic power generation output forecasting device provided in an embodiment of the present invention, it should
The course of work of each module in structural schematic diagram referring to Fig.1 in corresponding embodiment method implementation procedure, which includes:
First establishing unit 210, for establishing the ESN prediction model at moment to be predicted；
Acquiring unit 220, for obtaining the highest temperature in nearest preset time period, temperature on average, obtaining the historical juncture
Photovoltaic power generating value, at the time of the historical juncture value with the moment to be predicted at the time of be worth it is identical；
First determination unit 230, the vertex degree and the moment place to be predicted for determining the moment to be predicted are pre-
Survey the day index of day；
Computing unit 240, for by the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture,
The input of the vertex degree and the day index as the ESN prediction model, is gone out with the photovoltaic for obtaining the moment to be predicted
Force value.
Technical solution provided in an embodiment of the present invention, by establishing ESN prediction model, by temperature, index of type, history light
Input of the power generating value as ESN network is lied prostrate, the photovoltaic power generating value of prediction time, solves using neural network mould in prediction one day
The true problem of forecasting inaccuracy caused by sample data is few when type.
In another embodiment of the invention, further includes:
Second determination unit, for predefining the weather pattern of the prediction day；
Third determination unit, for according in the nearest preset time period photovoltaic power curve and the weather class
Type determines the day index and the vertex degree.
In another embodiment of the invention, in the nearest preset time period, the smallest weather class of photovoltaic power generating value
The corresponding day index of type is set as 1, and the corresponding vertex degree of the maximum weather pattern of photovoltaic power generating value sets not 1.
In another embodiment of the invention, further includes:
Second establishes unit from 6:00 to 19:00, establishes 14 ESN prediction models respectively to the prediction day；
Correspondingly, the first establishing unit is specifically used for:
Using any integral point moment of the 6:00 into 19:00 as the predetermined time.
In another embodiment of the invention, further includes:
Amending unit, for being repaired by photovoltaic power generating value of the Markov Chain prediction model to the moment to be predicted
Just.
Description and claims of this specification and term " first ", " second ", " third " " in above-mentioned attached drawing
The (if present)s such as four " are to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should manage
The data that solution uses in this way are interchangeable under appropriate circumstances, so that the embodiment of the present invention described herein for example can be to remove
Sequence other than those of illustrating or describe herein is implemented.In addition, term " includes " and " having " and theirs is any
Deformation, it is intended that cover it is non-exclusive include, for example, containing the process, method of a series of steps or units, system, production
Product or equipment those of are not necessarily limited to be clearly listed step or unit, but may include be not clearly listed or for this
A little process, methods, the other step or units of product or equipment inherently.
For device or system embodiments, since it essentially corresponds to embodiment of the method, thus related place referring to
The part of embodiment of the method illustrates.Device or system embodiment described above is only schematical, wherein described
Unit may or may not be physically separated as illustrated by the separation member, and component shown as a unit can be with
It is or may not be physical unit, it can it is in one place, or may be distributed over multiple network units.It can
It is achieved the purpose of the solution of this embodiment with selecting some or all of the modules therein according to the actual needs.This field is common
Technical staff can understand and implement without creative efforts.
In several embodiments provided by the present invention, it should be understood that disclosed system, device and method are not having
It has more than in the spirit and scope of the present invention, can realize in other way.Current embodiment is a kind of demonstration
Example, should not be taken as limiting, given particular content should in no way limit the purpose of the present invention.For example, the unit or
The division of subelement, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple lists
First or multiple subelements combine.In addition, multiple units can with or component may be combined or can be integrated into another and be
System, or some features can be ignored or not executed.
In addition, described system, the schematic diagram of device and method and different embodiments, without departing from the scope of the present invention
It is interior, it can be with other systems, module, techniques or methods combination or integrated.Another point, shown or discussed mutual coupling
It closes or direct-coupling or communication connection can be through some interfaces, the indirect coupling or communication connection of device or unit can be with
It is electrically mechanical or other forms.
The above is only a specific embodiment of the invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of photovoltaic power generation output forecasting method characterized by comprising
Establish the ESN prediction model at moment to be predicted；
Obtain the highest temperature, the temperature on average, the photovoltaic power generating value for obtaining the historical juncture in nearest preset time period, the history
At the time of moment value with the moment to be predicted at the time of be worth it is identical；
Determine the day index of the vertex degree at the moment to be predicted and prediction day where the moment to be predicted；
By the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture, the vertex degree and refer to the day
Input of the number as the ESN prediction model, to obtain the photovoltaic power generating value at the moment to be predicted.
2. the method according to claim 1, wherein further include:
Predefine the weather pattern of the prediction day；
According to the photovoltaic power curve and the weather pattern in the nearest preset time period, the day index and described is determined
Vertex degree.
3. according to the method described in claim 2, it is characterized in that, photovoltaic power generating value is most in the nearest preset time period
The corresponding day index of small weather pattern is set as 1, and the corresponding vertex degree of the maximum weather pattern of photovoltaic power generating value sets not 1.
4. the method according to claim 1, wherein further include:
From 6:00 to 19:00,14 ESN prediction models are established respectively to the prediction day；
Using any integral point moment of the 6:00 into 19:00 as the predetermined time.
5. method according to claim 1-4, which is characterized in that further include:
It is modified by photovoltaic power generating value of the Markov Chain prediction model to the moment to be predicted.
6. a kind of photovoltaic power generation output forecasting device characterized by comprising
First establishing unit, for establishing the ESN prediction model at moment to be predicted；
Acquiring unit goes out for obtaining the highest temperature in nearest preset time period, temperature on average, obtaining the photovoltaic of historical juncture
Force value, at the time of the historical juncture value with the moment to be predicted at the time of be worth it is identical；
First determination unit, for determining the day of the vertex degree at the moment to be predicted and prediction day where the moment to be predicted
Index；
Computing unit, for by the highest temperature, the temperature on average, the photovoltaic power generating value of the historical juncture, the point
The input of index and the day index as the ESN prediction model, to obtain the photovoltaic power generating value at the moment to be predicted.
7. device according to claim 6, which is characterized in that further include:
Second determination unit, for predefining the weather pattern of the prediction day；
Third determination unit, for according to the photovoltaic power curve and the weather pattern in the nearest preset time period, really
The fixed day index and the vertex degree.
8. device according to claim 7, which is characterized in that in the nearest preset time period, photovoltaic power generating value is most
The corresponding day index of small weather pattern is set as 1, and the corresponding vertex degree of the maximum weather pattern of photovoltaic power generating value sets not 1.
9. device according to claim 6, which is characterized in that further include:
Second establishes unit from 6:00 to 19:00, establishes 14 ESN prediction models respectively to the prediction day；
Correspondingly, the first establishing unit is specifically used for:
Using any integral point moment of the 6:00 into 19:00 as the predetermined time.
10. according to the described in any item devices of claim 6-9, which is characterized in that further include:
Amending unit, for being modified by photovoltaic power generating value of the Markov Chain prediction model to the moment to be predicted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811229686.6A CN109447843B (en) | 2018-10-22 | 2018-10-22 | Photovoltaic output prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811229686.6A CN109447843B (en) | 2018-10-22 | 2018-10-22 | Photovoltaic output prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109447843A true CN109447843A (en) | 2019-03-08 |
| CN109447843B CN109447843B (en) | 2022-01-28 |
Family
ID=65546918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811229686.6A Active CN109447843B (en) | 2018-10-22 | 2018-10-22 | Photovoltaic output prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447843B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110187280A (en) * | 2019-05-20 | 2019-08-30 | 天津大学 | A method of the lithium battery remaining life probabilistic forecasting based on gray model |
| CN111967652A (en) * | 2020-07-22 | 2020-11-20 | 国网浙江省电力有限公司电力科学研究院 | Double-layer cooperative real-time correction photovoltaic prediction method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103500365A (en) * | 2013-09-18 | 2014-01-08 | 广州供电局有限公司 | Photovoltaic power generation power prediction method and system |
| CN107766986A (en) * | 2017-10-31 | 2018-03-06 | 天津大学 | Leak integral form echo state network on-line study photovoltaic power Forecasting Methodology |
-
2018
- 2018-10-22 CN CN201811229686.6A patent/CN109447843B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103500365A (en) * | 2013-09-18 | 2014-01-08 | 广州供电局有限公司 | Photovoltaic power generation power prediction method and system |
| CN107766986A (en) * | 2017-10-31 | 2018-03-06 | 天津大学 | Leak integral form echo state network on-line study photovoltaic power Forecasting Methodology |
Non-Patent Citations (3)
| Title |
|---|
| 李乐等: "基于近邻传播聚类和回声状态网络的光伏预测", 《电力自动化设备》 * |
| 王大虎等: "基于改进MNN光伏发电功率预测模型", 《电子设计工程》 * |
| 王大虎等: "基于模块化回声状态神经网络光伏发电量预测", 《测控技术》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110187280A (en) * | 2019-05-20 | 2019-08-30 | 天津大学 | A method of the lithium battery remaining life probabilistic forecasting based on gray model |
| CN111967652A (en) * | 2020-07-22 | 2020-11-20 | 国网浙江省电力有限公司电力科学研究院 | Double-layer cooperative real-time correction photovoltaic prediction method |
| CN111967652B (en) * | 2020-07-22 | 2023-10-24 | 国网浙江省电力有限公司电力科学研究院 | Double-layer collaborative real-time correction photovoltaic prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109447843B (en) | 2022-01-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Random forest solar power forecast based on classification optimization | |
| Abuella et al. | Solar power forecasting using artificial neural networks | |
| Stefenon et al. | Photovoltaic power forecasting using wavelet neuro-fuzzy for active solar trackers | |
| Tao et al. | Forecasting power output for grid-connected photovoltaic power system without using solar radiation measurement | |
| Karabacak et al. | Artificial neural networks for controlling wind–PV power systems: A review | |
| CN113128793A (en) | Photovoltaic power combination prediction method and system based on multi-source data fusion | |
| CN102930358B (en) | A kind of neural net prediction method of photovoltaic power station power generation power | |
| CN101383023B (en) | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation | |
| CN108388956A (en) | Consider the photovoltaic power prediction technique of attenuation | |
| Monfared et al. | Fuzzy predictor with additive learning for very short-term PV power generation | |
| CN112101669B (en) | Photovoltaic power interval prediction method based on improved extreme learning machine and quantile regression | |
| CN115115125B (en) | Photovoltaic power interval probability prediction method based on deep learning fusion model | |
| CN111191854A (en) | Photovoltaic power generation prediction model and method based on linear regression and neural network | |
| Ciabattoni et al. | Solar irradiation forecasting using RBF networks for PV systems with storage | |
| Al-Rousan et al. | Assessment of predicting hourly global solar radiation in Jordan based on Rules, Trees, Meta, Lazy and Function prediction methods | |
| CN106446440A (en) | Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine | |
| Aliberti et al. | Forecasting Short-term Solar Radiation for Photovoltaic Energy Predictions. | |
| CN109447843A (en) | A kind of photovoltaic power generation output forecasting method and device | |
| Omar et al. | Seasonal clustering forecasting technique for intelligent hourly solar irradiance systems | |
| CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
| Mishra et al. | Global solar radiation forecast using an ensemble learning approach | |
| Basaran et al. | A short-term photovoltaic output power forecasting based on ensemble algorithms using hyperparameter optimization | |
| CN116167465A (en) | Solar irradiance prediction method based on multivariate time series ensemble learning | |
| Paulin et al. | SOLAR PHOTOVOLTAIC OUTPUT POWER FORECASTING USING BACK PROPAGATION NEURAL NETWORK. | |
| CN111797981B (en) | Single-solar photovoltaic power generation capacity prediction method based on two-way long-short-term memory 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 |