CN109447843B - Photovoltaic output prediction method and device - Google Patents
Photovoltaic output prediction method and device Download PDFInfo
- Publication number
- CN109447843B CN109447843B CN201811229686.6A CN201811229686A CN109447843B CN 109447843 B CN109447843 B CN 109447843B CN 201811229686 A CN201811229686 A CN 201811229686A CN 109447843 B CN109447843 B CN 109447843B
- Authority
- CN
- China
- Prior art keywords
- predicted
- index
- photovoltaic output
- moment
- output value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 239000011159 matrix material Substances 0.000 claims description 32
- 230000007704 transition Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 7
- 238000011161 development Methods 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 abstract description 4
- 210000002569 neuron Anatomy 0.000 description 16
- 230000008569 process Effects 0.000 description 16
- 238000012549 training Methods 0.000 description 16
- 238000013528 artificial neural network Methods 0.000 description 9
- 239000013598 vector Substances 0.000 description 9
- 230000006870 function Effects 0.000 description 8
- 238000010248 power generation Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 6
- 238000005070 sampling Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 241000529895 Stercorarius Species 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000004891 communication 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
- 238000012417 linear regression Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
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 photovoltaic output prediction method and a photovoltaic output prediction device, wherein the method comprises the following steps: establishing an ESN prediction model at a time to be predicted; acquiring the highest air temperature and the average air temperature in the latest preset time period, and acquiring the photovoltaic output value at the historical moment, wherein the time value at the historical moment is the same as the time value at the moment to be predicted; determining a point index of the time to be predicted and a day index of a prediction day of the time to be predicted; and taking the maximum air temperature, the average air temperature, the photovoltaic output value at the historical moment, the point index and the daily index as the input of the ESN prediction model to obtain the photovoltaic output value at the moment to be predicted. Compared with the prior art, the photovoltaic output value at the prediction time in one day is predicted by establishing an ESN prediction model and taking the air temperature, the type index and the historical photovoltaic output value as the input of an ESN network, and the problem of inaccurate prediction caused by less sample data when a neural network model is adopted is solved.
Description
Technical Field
The invention relates to the technical field of photovoltaics, in particular to a photovoltaic output method and a photovoltaic output device.
Background
Under the influence of economic globalization, the environmental pollution problem and the global energy crisis are two major problems which are increasingly prominent, and the development and utilization of new energy are important measures for keeping the long-term supply of energy. As one of the more mature new energy sources, solar energy is widely concerned by society due to its advantages of cleanness, no pollution, wide sources and the like, and large-scale photovoltaic grid-connected power generation becomes the development trend of solar power generation. Photovoltaic power generation is influenced by factors such as external weather and self performance, photovoltaic output is fluctuated and intermittent due to the fluctuation of the factors, the electric energy quality is influenced by large-scale photovoltaic grid-connected operation, and the safe, stable and economic operation of a power system is influenced, so that the photovoltaic output is very necessary to be predicted.
The conventional photovoltaic output prediction method is an artificial neural network prediction method. The artificial neural network can simulate the intelligent processing process of human brain, has strong autonomous learning and self-adaptive capacity, and is popular with researchers in photovoltaic power prediction. However, the neural network does not work well when the sample data is insufficient, the sample data amount accumulation of certain weather types (such as rainstorm, snowstorm, heavy rain and the like) is less, and the model fitting effect is poor.
Disclosure of Invention
In order to solve the technical problem, the embodiment of the invention provides a photovoltaic output prediction method and a photovoltaic output prediction device, and the technical scheme is as follows:
a photovoltaic output prediction method, comprising:
establishing an ESN prediction model at a time to be predicted;
acquiring the highest air temperature and the average air temperature in the latest preset time period, and acquiring the photovoltaic output value at the historical moment, wherein the time value at the historical moment is the same as the time value at the moment to be predicted;
determining a point index of the time to be predicted and a day index of a prediction day of the time to be predicted;
and taking the maximum air temperature, the average air temperature, the photovoltaic output value at the historical moment, the point index and the daily index as the input of the ESN prediction model to obtain the photovoltaic output value at the moment to be predicted.
Preferably, the method further comprises the following steps:
predetermining a weather type of the predicted day;
and determining the daily index and the point index according to the photovoltaic output curve and the weather type in the latest preset time period.
Preferably, in the latest preset time period, the day index corresponding to the weather type with the minimum photovoltaic output value is set to 1, and the point index corresponding to the weather type with the maximum photovoltaic output value is set to 1.
Preferably, the method further comprises the following steps:
respectively establishing 14 ESN prediction models for the prediction days from 6:00 to 19: 00;
and taking any integral point time of 6: 00-19: 00 as the preset time.
Preferably, the method further comprises the following steps:
and correcting the photovoltaic output value at the moment to be predicted through a Markov chain prediction model.
A photovoltaic output prediction apparatus, comprising:
the first establishing unit is used for establishing an ESN prediction model at the moment to be predicted;
the acquiring unit is used for acquiring the highest air temperature and the average air temperature in the latest preset time period and acquiring the photovoltaic output value at the historical moment, wherein the time value at the historical moment is the same as the time value at the moment to be predicted;
the first determining unit is used for determining the point index of the time to be predicted and the day index of the prediction day of the time to be predicted;
and the calculation unit is used for taking the highest air temperature, the average air temperature, the photovoltaic output value at the historical moment, the point index and the daily index as the input of the ESN prediction model so as to obtain the photovoltaic output value at the moment to be predicted.
Preferably, the method further comprises the following steps:
a second determination unit configured to determine a weather type of the predicted day in advance;
and the third determining unit is used for determining the daily index and the point index according to the photovoltaic output curve in the latest preset time period and the weather type.
Preferably, in the latest preset time period, the day index corresponding to the weather type with the minimum photovoltaic output value is set to 1, and the point index corresponding to the weather type with the maximum photovoltaic output value is set to 1.
Preferably, the method further comprises the following steps:
the second establishing unit respectively establishes 14 ESN prediction models for the prediction days from 6:00 to 19: 00;
correspondingly, the first establishing unit is specifically configured to:
and taking any integral point time of 6: 00-19: 00 as the preset time.
Preferably, the method further comprises the following steps:
and the correction unit is used for correcting the photovoltaic output value at the moment to be predicted through a Markov chain prediction model.
According to the technical scheme provided by the embodiment of the invention, the photovoltaic output value at the prediction time in one day is predicted by establishing the ESN prediction model and taking the air temperature, the type index and the historical photovoltaic output value as the input of the ESN, so that the problem of inaccurate prediction caused by less sample data when a neural network model is adopted is solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a schematic flow chart of a photovoltaic output prediction method according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a photovoltaic output prediction apparatus according to an embodiment of the present invention;
fig. 3 is a network structure diagram of an echo state network according to an embodiment of 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.
Referring to fig. 1, fig. 1 is a flowchart illustrating an implementation of a photovoltaic output prediction method according to an embodiment of the present invention, where the method includes:
and S101, establishing an ESN prediction model at the moment to be predicted.
And S102, acquiring the highest air temperature and the average air temperature in the latest preset time period, and acquiring the photovoltaic output value at the historical time, wherein the time value at the historical time is the same as the time value at the time to be predicted.
In practical application, meteorological data of a place where the photovoltaic power station is located within the last year can be collected, wherein the meteorological data comprise the highest air temperature, the lowest air temperature and the weather type, and historical output values of the photovoltaic power station within the last year within 24 hours each day are collected. And (5) intercepting the output value of 6: 00-19: 00 as a photovoltaic output curve. The main factors influencing the output power of the power generation system comprise illumination intensity, ambient temperature, the characteristics of a photovoltaic power station and the like, and because the data of the factors are not easy to collect, the weather type, the highest temperature, the average temperature and the historical photovoltaic output value are indirectly adopted as the factors influencing photovoltaic power generation.
The weather factors such as solar radiation, cloud cover, wind speed, temperature and the like in different types of weather are different, and the output power of the photovoltaic cell is also different. For example, the output power varies greatly between sunny and rainy days. The weather type has great influence on the output result of the photovoltaic power station, and the influence is directly reflected in the magnitude of the daily generated power. Therefore, the weather type is mapped into a numerical type index through processing photovoltaic output data, and the numerical type index is used as an input of the echo state network training and prediction.
In order to improve the prediction accuracy, 14 ESN prediction models are built according to point-by-point prediction, namely, 6:00 to 19:00 of a corresponding day. Therefore, the type index is used as the input of the model, not only reflecting the influence of the weather type on the output all day, but also reflecting the contribution to the output at a certain time point. In view of this, the type index is defined as a two-dimensional array including two parts, i.e., (daily index, dot index), of daily index and dot index.
Step S103, determining the point index of the time to be predicted and the day index of the prediction day of the time to be predicted.
And collecting meteorological data of each day in the last year of the location of the photovoltaic power station, wherein the meteorological data comprise the highest air temperature, the lowest air temperature and the weather type, and collecting historical output values of the photovoltaic power station for 24 hours each day in the last year. And (5) intercepting the output value of 6: 00-19: 00 as a photovoltaic output curve. The main factors influencing the output power of the power generation system comprise illumination intensity, ambient temperature, the characteristics of a photovoltaic power station and the like, and because the data of the factors are not easy to collect, the weather type, the highest temperature, the average temperature and the historical photovoltaic output value are indirectly adopted as the factors influencing photovoltaic power generation.
The weather factors such as solar radiation, cloud cover, wind speed, temperature and the like in different types of weather are different, and the output power of the photovoltaic cell is also different. For example, the output power varies greatly between sunny and rainy days. The weather type has great influence on the output result of the photovoltaic power station, and the influence is directly reflected in the magnitude of the daily generated power. Therefore, the weather type is mapped into a numerical type index through processing photovoltaic output data, and the numerical type index is used as an input of the echo state network training and prediction.
In order to improve the prediction accuracy, 14 ESN prediction models are built according to point-by-point prediction, namely, 6:00 to 19:00 of a corresponding day. Therefore, the type index is used as the input of the model, not only reflecting the influence of the weather type on the output all day, but also reflecting the contribution to the output at a certain time point. In view of this, the type index is defined as a two-dimensional array including two parts, i.e., (daily index, dot index), of daily index and dot index.
And step S104, taking the maximum air temperature, the average air temperature, the photovoltaic output value at the historical moment, the point index and the daily index as the input of the ESN prediction model to obtain the photovoltaic output value at the moment to be predicted.
The prediction model based on the echo state network established by the embodiment highlights the influence of the weather type on the photovoltaic power generation power, the factor is considered in the modeling process, the weight of the daily index is increased, the model does not need to be decomposed into the submodels according to the weather factors in the prediction process, the prediction model is suitable for various weather types, and the prediction model has strong applicability and good prediction capability.
In order to improve the prediction accuracy, 14 ESN prediction models are built according to point-by-point prediction, namely, 6:00 to 19:00 of a corresponding day. The parameter setting, training and predicting processes of each network model are as follows:
input layer determination: and predicting the output value, the type index, the highest air temperature and the average air temperature at the same time three days before the day, and predicting the highest air temperature, the average air temperature and the type index of the day.
Determining an output layer: predicting the output value at a certain time from 6:00 to 19:00 of a day.
Preferably, after the photovoltaic output value at the time to be predicted is obtained, the photovoltaic output value at the time to be predicted can be corrected through a Markov chain prediction model.
The Markov chain prediction is used for predicting the possible states of some variables in a certain period in the future according to the current states and the change trends of the variables, and is suitable for expressing the problem with large random fluctuation. The method for correcting the prediction result of the echo state network by the Markov residual error predicts the output of the photovoltaic power station, accords with the characteristics of the photovoltaic power station, and complements the advantages of the photovoltaic power station and the Markov residual error, so that a more accurate prediction conclusion is obtained.
And taking different upper and lower threshold values of the relative error between the predicted value and the actual value of the ESN as a state division value domain. In markov prediction, the most critical step is to solve a state transition probability matrix. In the solving process of the state probability, the classification of the state is important, and common methods include a mean-square-difference classification method, a cluster analysis method and a left-right segmentation method. The invention adopts a mean-square error grading method. The basic idea is as follows: and analyzing the fluctuation amplitude and the fluctuation development trend of the error of the result obtained by the ESN echo state network prediction by using a Markov model to obtain a state transition probability matrix of the error, and correcting the neural network prediction result according to the matrix.
The method comprises the following specific steps:
a. calculating a relative error xi of a predicted value of the ESN network test sample;
b. taking different upper and lower thresholds of a relative error xi of a predicted value of a test sample as a state division value domain, and establishing a state division standard;
c. determining a state transition matrix P from the relative error states(k)；
d. Determining an initial state vector V (0);
e. according to the state transition formula V (k) ═ V (0) P(k)Solving the state transition result of the k step;
f. the predicted value is corrected by P ═ P0(1-ξ*) In which P is0Is the prediction result of the network in the echo state, ξ*The average value of the upper threshold and the lower threshold of the error state interval is obtained.
According to the technical scheme provided by the embodiment of the invention, the photovoltaic output value at the prediction time in one day is predicted by establishing the ESN prediction model and taking the air temperature, the type index and the historical photovoltaic output value as the input of the ESN, so that the problem of inaccurate prediction caused by less sample data when a neural network model is adopted is solved.
The algorithm model involved in the present invention is explained below:
(1) echo state network model
Jaeger and Haas proposed an Echo State Network (ESN) and a corresponding learning algorithm in 2001, which opened a new way for the research of the recurrent neural network. This method, also called the so-called pool computation mode, introduces an internal network called pool, in which a complex and diverse nonlinear state space is excited when external input sequences enter this internal network, and then the network output is obtained through a simple readout network. The biggest difference from the recurrent neural network is that in the training process, the connection weight inside the reserve pool is fixed and unchanged, and the adjustment is only carried out aiming at the read-out network, so that the calculation amount of the training is greatly reduced, the local minimum phenomenon which is difficult to avoid by most of learning algorithms based on gradient descent is avoided, and the good modeling precision can be obtained.
The ESN constitutes a random network structure by randomly arranging large-scale sparsely connected analog neurons, this large-scale recursive network for processing random sparsely connected time-series input signals, is called "reserve pool" (Reservoirs), and its structure is shown in fig. 3. The linear combination of the neuron (reservoir internal unit) outputs x (n) forms the output signal y (n) of the system. Connection weight W between neurons in the reservoir, and connection weight W of input signal to neurons in the reservoirinAll are randomly generated and are kept unchanged after being generated, and training is not needed. The connection right to be trained only has the connection right W from the reserve pool to the system outputoutThe training process generally requires only a linear regression problem to be solved.
The structure of the ESN was further analyzed, as well as the mathematical model. Consider the network structure of an ESN as shown in fig. 1 [4 ]. Assume that the system has M input units, N internal Processing units (PEs), i.e., N internal neurons, and L output units. The values of the input unit, the internal state and the output unit at the moment n are 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。
structurally, the ESN is a special type of recurrent neural network, and the basic idea is to use a large-scale randomly connected recurrent network to replace the middle layer in the classical neural network, thereby simplifying the training process of the network. The state equation and the output equation of the echo state network can be given by equation (1):
w, W thereinin、WbackA connection weight matrix respectively representing the state variable, the input and the output to the state variable; woutA matrix of connection weights representing pool, input and output to output,
the bias term representing the output may alternatively represent noise. f ═ f1，f2，...，fN]Representing an internal neuron activation function, in general f1Take (i ═ 1, 2.., N) as the hyperbolic tangent function.
Representing output functions, the output layers being generally linear, i.e.
Taking an identity function. Among the above various connection weight matrices, the connection weight matrix W connected to the reserve poolin、W、WbackRandom generation, once generated, is fixed. And each connection weight matrix W connected to the outputoutThe connection weights are obtained by training according to input and output data of the system, and because the state variables and the output are in a linear relation, the connection weights only need to be solved by solving a linear loopThe problem is solved. W is a sparse connection matrix, and the sparsity of the sparse connection matrix is generally 1% -5%. It is important that the ESN network has the characteristics of an echo state, which means that the input vector before the network and the initial state of the reserve pool have less and less influence on the future state until fading. To guarantee the echo effect of the ESN network, the spectral radius of W must be guaranteed to be less than 1.
1) Selection of reserve pool parameters
The construction method of the ESN is simple, but some key parameters in the network must be empirically selected and adjusted during a specific use.
a) Size N of reserve pool
The reserve pool scale refers to the number of neurons in the reserve pool, and the selection of the reserve pool scale is related to the number of training samples, so that the network performance is greatly influenced. In general, the larger the size of the reserve pool, the more complex the dynamic system that the ESN can represent may be. The larger the pool size, the more accurate the ESN describes it for a given dynamic system. However, the pool size cannot be arbitrarily increased because overfitting may be caused if the pool size is too large. Overfitting will result in a reduction in the generalization ability of the model to the test data. The general selection principle is to increase the pool size step by step until the processing capacity of the network for the test samples (such as classification error rate, prediction error, etc.) deteriorates.
b) Internal connection weight matrix spectrum radius SR of reserve pool
The radius of the internal connection weight spectrum of the reserve tank is a characteristic value of the maximum absolute value of the internal connection weight matrix W of the reserve tank, and is expressed as λmax. SR is a key parameter of the reserve pool, usually when lambdamaxThe ESN can have echo status attributes < 1 to ensure that the network is not affected by the status and inputs over a long enough period. For a specific time series prediction problem, the selection of the parameter SR has a great influence on the ESN performance, and in the research on the adaptability of the reserve pool, the SR can be optimized as a key parameter.
c) Input unit scale IS of reserve pool
The reservoir input unit scale parameter IS a scale factor to be multiplied before the input signal of the reservoir IS connected to the neuron inside the reservoir. Because of the different choice of neuron activation functions and the different characteristics of the sample data in the reservoir, the input signal IS usually not directly added to the reservoir, but rather IS scaled by a scaling factor IS, i.e. the input signal IS scaled first.
d) Reserve pool sparsity degree SD
The reservoir sparsity degree SD represents the connection situation between neurons in the reservoir. The connection relationship between not all neurons in the reservoir, but only some neurons, is present, and the parameter SD represents the percentage of the total number (N) of neurons connected to each other in the reservoir. The richness of the vectors in the reserve pool is influenced by the richness of the vectors in the reserve pool, and the richer the vectors of the network are, the stronger the nonlinear approximation capability of the network is.
In addition, it is noted that the selection of the internal neuron excitation function is also included. Jaeger originally demonstrated an echogenic state using linear neurons. Generally, for a linear neuron network, it is possible to provide stronger short-term memory while possessing an echo state. However, most systems requiring neural networks for modeling have strong nonlinearity, and finally, the more common sigmoid excitation function is often adopted in practical application.
2) Training of echo state networks
The training process of the echo state network is to determine an output connection weight matrix W in the system according to a given training sample (u (n), y (n), n being 1, 2outThe process of (1). For simplicity, W is assumed herebackThe simultaneous input/output connection right is also assumed to be O, and the sample data (u (n), y (n)), where n is 1, 2. The training process of the echo state network can be divided into two stages: a Sampling (Sampling) stage and a Weight calculation (Weight calculation) stage.
a) Sampling
The sampling phase first arbitrarily selects the initial state of the network, but normally selects the initial state of the network as O, i.e., x (O) ═ 0. Training samples (u (n), n 1, 2.., M) are connected via input of the connection weight WinSample data y (n) is fed back via feedback connection WbackRespectively added to the reserve pool, and according to the system (1), the calculation and corresponding output of the system state are sequentially completed
And (4) calculating and collecting. Note that the calculation of the system state x (n) at each time needs to write the sample data y (n) into the output unit. In order to calculate the output connection weight matrix, it is necessary to collect (sample) internal state variables from a certain time. It is assumed here that the system states are collected starting from time m and are represented by a vector (x)1(i)，x2(i)，...，xN(i) (i ═ M, M + 1.., M) forms a matrix B (M-M +1, N) for the rows, while corresponding sample data y (N) is also collected and forms a column vector T (M-M +1, 1). Two points need to be explained here:
if the system comprises input-to-output and output-to-output connection rights, when collecting the state matrix R of the system, corresponding input and output parts also need to be collected;
in order to eliminate the influence of any initial state on the dynamic characteristics of the system, the state of the system is always collected from a certain time later. From this point in time, the system can be considered to reflect the mapping relationship between the input and output sample data.
b) Weight calculation
In order to realize the calculation of the weight, the output connection weight W needs to be calculated according to the system state matrix and the sample data collected in the sampling stageout. Because of the state variables x (n) and the system output
Are linear relationship, and the goal to be achieved is to utilize the actual output of the network
Approaches the desired output y (n), i.e.
That is, it is desirable to calculate the weight
(
Is a matrix WoutElement(s) to meet the minimum mean square error of the system, i.e. the following optimization problem needs to be solved:
from a mathematical point of view, this is a linear regression problem, which can be summarized as the inverse matrix problem of the matrix B, which may be ill-conditioned in practical applications, which can be further processed computationally as the pseudo-inverse problem of the matrix B, i.e. the problem of the inverse matrix B
(Wout)T＝B-1T
By this time, ESN network training has been completed.
(2) Markov chain
The Markov chain is called Andrew Markov (1856-. In this process, the past (i.e., historical state before the current date) is irrelevant to the predicted future (i.e., future state after the current date) given the current knowledge or information.
The Markov chain prediction is used for predicting the possible states of some variables in a certain period in the future according to the current states and the change trends of the variables, and is suitable for describing the problem with high random fluctuation.
In markov prediction, the most critical step is to solve a state transition probability matrix. During the solution of the state probabilities, the hierarchy of states appears to be offImportantly, a common method is mean-square error classification. For sequence x1,L,xnWith a mean value of
The mean-square error is s, and after using the mean-square error classification method, the sequence can be generally divided into 5 levels:
wherein a is1,a4Take on the value of [1.0,1.5]Middle value, a2,a3In the [0.3,0.6 ]]Taking the value in the step (1).
According to Markov theory, State EiAfter k steps becomes EjHas a probability of
In the formula:
for the sample from EtTo EfThe number of transfers of (c); n is a radical ofiFor the total number of occurrences of the state, the k-step state probability transition matrix is
The state probability transition matrix is an n-order square matrix and has two characteristics:
Calculating the state vector of the k step by using a state probability transition formula
V(k)＝,V(0)P(k) (5)
Where V (0) is the initial state vector.
Referring to fig. 2, fig. 2 is a schematic structural diagram of a photovoltaic output prediction apparatus according to an embodiment of the present invention, where the working process of each module in the schematic structural diagram refers to the execution process of the method in the corresponding embodiment of fig. 1, and the apparatus includes:
a first establishing unit 210, configured to establish an ESN prediction model at a time to be predicted;
the obtaining unit 220 is configured to obtain a highest air temperature and an average air temperature in a latest preset time period, and obtain a photovoltaic output value at a historical time, where a time value of the historical time is the same as a time value of the time to be predicted;
a first determining unit 230, configured to determine a point index of the time to be predicted and a day index of a prediction day where the time to be predicted is located;
and the calculating unit 240 is configured to use the highest air temperature, the average air temperature, the photovoltaic output value at the historical time, the point index and the daily index as inputs of the ESN prediction model to obtain the photovoltaic output value at the time to be predicted.
According to the technical scheme provided by the embodiment of the invention, the photovoltaic output value at the prediction time in one day is predicted by establishing the ESN prediction model and taking the air temperature, the type index and the historical photovoltaic output value as the input of the ESN, so that the problem of inaccurate prediction caused by less sample data when a neural network model is adopted is solved.
In another embodiment of the present invention, the method further comprises:
a second determination unit configured to determine a weather type of the predicted day in advance;
and the third determining unit is used for determining the daily index and the point index according to the photovoltaic output curve in the latest preset time period and the weather type.
In another embodiment of the present invention, in the latest preset time period, the day index corresponding to the weather type with the minimum photovoltaic output value is set to 1, and the point index corresponding to the weather type with the maximum photovoltaic output value is set to 1.
In another embodiment of the present invention, the method further comprises:
the second establishing unit respectively establishes 14 ESN prediction models for the prediction days from 6:00 to 19: 00;
correspondingly, the first establishing unit is specifically configured to:
and taking any integral point time of 6: 00-19: 00 as the preset time.
In another embodiment of the present invention, the method further comprises:
and the correction unit is used for correcting the photovoltaic output value at the moment to be predicted through a Markov chain prediction model.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
For device or system embodiments, as they correspond substantially to method embodiments, reference may be made to the method embodiments for some of their descriptions. The above-described embodiments of the apparatus or system are merely illustrative, and the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
In the several embodiments provided in the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other ways without departing from the spirit and scope of the present invention. The present embodiment is an exemplary embodiment only, and should not be taken as limiting, and the specific contents given should not limit the object of the present invention. For example, the division of the unit or the sub-unit is only one logical function division, and there may be another division manner in actual implementation, for example, a plurality of units or a plurality of sub-units are combined together. In addition, various elements or components may be combined or may be integrated into another system, or some features may be omitted, or not implemented.
Additionally, the systems, apparatus, and methods described, as well as the illustrations of various embodiments, may be combined or integrated with other systems, modules, techniques, or methods without departing from the scope of the invention. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The foregoing is directed to embodiments of the present invention, and it is understood that various modifications and improvements can be made by those skilled in the art without departing from the spirit of the invention.
Claims (8)
1. A method of photovoltaic power output prediction, comprising:
establishing an ESN prediction model at a time to be predicted;
acquiring the highest air temperature and the average air temperature in the latest preset time period, and acquiring the photovoltaic output value at the historical moment, wherein the time value at the historical moment is the same as the time value at the moment to be predicted;
mapping the weather type into a numerical type index, wherein the numerical type index is specifically a two-dimensional array and comprises a day index and a point index; the point index reflects the contribution to the output at a certain time point, and the day index reflects the influence of the weather type on the output all day;
determining a point index of the time to be predicted and a day index of a prediction day of the time to be predicted;
taking the maximum air temperature, the average air temperature, the photovoltaic output value at the historical moment, the point index and the day index as the input of the ESN prediction model to obtain the photovoltaic output value at the moment to be predicted;
correcting the photovoltaic output value at the moment to be predicted through a Markov chain prediction model, wherein the correction comprises the following steps: and analyzing the fluctuation range and the fluctuation development trend of the photovoltaic output value error at the moment to be predicted by applying a Markov model to the photovoltaic output value at the moment to be predicted to obtain a state transition probability matrix of the error, and correcting the photovoltaic output value at the moment to be predicted according to the state transition probability matrix.
2. The method of claim 1, further comprising:
predetermining a weather type of the predicted day;
and determining the daily index and the point index according to the photovoltaic output curve and the weather type in the latest preset time period.
3. The method according to claim 2, wherein in the latest preset time period, the day index corresponding to the weather type with the minimum photovoltaic output value is set to 1, and the point index corresponding to the weather type with the maximum photovoltaic output value is set to 1.
4. The method of claim 1, further comprising:
respectively establishing 14 ESN prediction models for the prediction days from 6:00 to 19: 00;
and taking any integral point time from 6:00 to 19:00 as the time to be predicted.
5. A photovoltaic output prediction device, comprising:
the first establishing unit is used for establishing an ESN prediction model at the moment to be predicted;
the acquiring unit is used for acquiring the highest air temperature and the average air temperature in the latest preset time period and acquiring the photovoltaic output value at the historical moment, wherein the time value at the historical moment is the same as the time value at the moment to be predicted;
the weather type mapping device comprises a first mapping unit, a second mapping unit and a control unit, wherein the first mapping unit is used for mapping a weather type into a numerical type index, and the type index is specifically a two-dimensional array and comprises a day index and a point index; the point index reflects the contribution to the output at a certain time point, and the day index reflects the influence of the weather type on the output all day;
the first determining unit is used for determining the point index of the time to be predicted and the day index of the prediction day of the time to be predicted;
the calculation unit is used for taking the highest air temperature, the average air temperature, the photovoltaic output value at the historical moment, the point index and the daily index as the input of the ESN prediction model so as to obtain the photovoltaic output value at the moment to be predicted;
the correction unit is used for correcting the photovoltaic output value at the moment to be predicted through a Markov chain prediction model, and comprises: and analyzing the fluctuation range and the fluctuation development trend of the photovoltaic output value error at the moment to be predicted by applying a Markov model to the photovoltaic output value at the moment to be predicted to obtain a state transition probability matrix of the error, and correcting the photovoltaic output value at the moment to be predicted according to the state transition probability matrix.
6. The apparatus of claim 5, further comprising:
a second determination unit configured to determine a weather type of the predicted day in advance;
and the third determining unit is used for determining the daily index and the point index according to the photovoltaic output curve in the latest preset time period and the weather type.
7. The device according to claim 6, wherein in the latest preset time period, the day index corresponding to the weather type with the minimum photovoltaic output value is set to 1, and the point index corresponding to the weather type with the maximum photovoltaic output value is set to 1.
8. The apparatus of claim 5, further comprising:
the second establishing unit respectively establishes 14 ESN prediction models for the prediction days from 6:00 to 19: 00;
correspondingly, the first establishing unit is specifically configured to:
and taking any integral point time from 6:00 to 19:00 as the time 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 CN109447843A (en) | 2019-03-08 |
| CN109447843B true 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) |
Families Citing this family (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 |
| 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 |
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 |
|---|
| 基于改进MNN光伏发电功率预测模型;王大虎等;《电子设计工程》;20180620;第26卷(第12期);第1页至第5页 * |
| 基于模块化回声状态神经网络光伏发电量预测;王大虎等;《测控技术》;20180118(第01期);第59页至63页 * |
| 基于近邻传播聚类和回声状态网络的光伏预测;李乐等;《电力自动化设备》;20160710;第36卷(第07期);第41页至第46页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109447843A (en) | 2019-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | A review on global solar radiation prediction with machine learning models in a comprehensive perspective | |
| US20220373984A1 (en) | Hybrid photovoltaic power prediction method and system based on multi-source data fusion | |
| CN108280551B (en) | Photovoltaic power generation power prediction method utilizing long-term and short-term memory network | |
| CN110084367B (en) | Soil moisture content prediction method based on LSTM deep learning model | |
| CN108764539B (en) | Upstream and downstream water level prediction method for cascade power station | |
| CN110738344B (en) | Distributed reactive power optimization method and device for load prediction of power system | |
| CN112488415A (en) | Power load prediction method based on empirical mode decomposition and long-and-short-term memory network | |
| CN112561058B (en) | Short-term photovoltaic power prediction method based on Stacking-integrated learning | |
| CN111222674A (en) | Short-term photovoltaic power generation capacity prediction method based on long-term and short-term memory neural network | |
| CN109583565B (en) | Flood prediction method based on attention model long-time and short-time memory network | |
| WO2018161722A1 (en) | Power load forecasting method based on long short-term memory neural network | |
| CN107766986A (en) | Leak integral form echo state network on-line study photovoltaic power Forecasting Methodology | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN110232476A (en) | A kind of micro-grid load prediction technique based on deep learning | |
| CN114330935B (en) | New energy power prediction method and system based on multiple combination strategies integrated learning | |
| CN110956312A (en) | Photovoltaic power distribution network voltage prediction method based on EMD-CNN deep neural network | |
| CN114492922A (en) | Medium-and-long-term power generation capacity prediction method | |
| CN114065653A (en) | Construction method of power load prediction model and power load prediction method | |
| CN109447843B (en) | Photovoltaic output prediction method and device | |
| CN115409258A (en) | Hybrid deep learning short-term irradiance prediction method | |
| CN116384583A (en) | Photovoltaic power prediction method based on multiple neural networks | |
| CN115660887A (en) | Photovoltaic output prediction method and system based on limited weather forecast information | |
| CN114330934A (en) | Model parameter self-adaptive GRU new energy short-term power generation power prediction method | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| CN116632823A (en) | Short-term photovoltaic power prediction method based on power conversion model and multi-layer perceptron |
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 |