CN113780636A - Solar radiation prediction method based on EMD-GRU-Attention - Google Patents
Solar radiation prediction method based on EMD-GRU-Attention Download PDFInfo
- Publication number
- CN113780636A CN113780636A CN202110988946.3A CN202110988946A CN113780636A CN 113780636 A CN113780636 A CN 113780636A CN 202110988946 A CN202110988946 A CN 202110988946A CN 113780636 A CN113780636 A CN 113780636A
- Authority
- CN
- China
- Prior art keywords
- gru
- solar radiation
- time
- emd
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005855 radiation Effects 0.000 title claims abstract description 68
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000007246 mechanism Effects 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 27
- 238000013135 deep learning Methods 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 17
- 238000005259 measurement Methods 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 11
- 230000014509 gene expression Effects 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000007704 transition Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000010248 power generation 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention relates to a solar radiation prediction method based on EMD-GRU-Attention, which comprises the steps of S1, inputting a layer; step S2, an EMD layer; step S3, a GRU layer; step S4, an Attention mechanism layer; and step S5, outputting the layer. According to the invention, signals with different frequencies are gradually decomposed according to the characteristics of data per se through an EMD algorithm, prediction models are respectively established for the decomposed subsequences, then the output of each prediction model is linearly weighted to obtain a final prediction result, when the input time sequence in solar radiation prediction is too long, the GRU network is easy to have the problems of information loss and difficult modeling, which can affect the prediction precision of the models, and an Attention mechanism distributes weight for the output of a hidden layer of the GRU network, so that the hidden layer of the GRU can highlight key information in prediction, a solar radiation value of one to six hours in the future is finally obtained, and the prediction precision and efficiency of the solar radiation multi-step prediction model are improved.
Description
Technical Field
The invention relates to the technical field of solar radiation prediction, in particular to a solar radiation prediction method based on EMD-GRU-Attention.
Background
Solar energy is widely applied to a heat energy system and an electric power system, the solar energy is directly from the sun in a radiation mode, but the solar radiation degree is easily influenced by atmospheric changes, so that the problems of mismatch with the heat load of a building and the like exist in a heat supply system, the problems of intermittence, uncertainty and the like exist in photovoltaic power generation power, and therefore, a high-precision radiation prediction model is researched, a reasonable optimization regulation and control plan is made for the heat supply system and the electric power system, and the solar energy utilization and development significance is achieved.
Predictive models for solar radiation can be generally divided into three categories: the first type is a physical prediction model, the second type is a statistical prediction model, and the third type is a hybrid prediction model; the deep learning models such as LSTM and GRU are concerned with the capability of deeply mining complex data, the GRU and the LSTM are similar in structure, but the GRU reduces the number of internal hidden states and thresholds, the problem of gradient explosion or disappearance is solved, the nonlinear time sequence data features are effectively extracted, and the training time of the models is shortened.
The invention provides a high-precision hybrid prediction model of GRU (generalized regression decomposition) based on EMD (empirical mode decomposition), which can realize solar radiation prediction for 6h in the future, calculate the weight of features by introducing an Attention mechanism and allocate more Attention to key data, thereby improving the prediction precision.
Disclosure of Invention
Therefore, the invention provides a solar radiation prediction method based on EMD-GRU-Attention, which is used for overcoming the problem of insufficient solar radiation prediction precision in the prior art.
To achieve the above objects, the present invention provides a solar radiation prediction method based on EMD-GRU-Attention, comprising,
step S1, inputting a layer, acquiring solar radiation data and related meteorological data, and preprocessing the solar radiation data and the related meteorological data;
step S2, an EMD layer decomposes the preprocessed solar radiation time-series data into a plurality of IMF components and a residual RES by utilizing EMD;
step S3, the GRU layer respectively establishes corresponding GRU deep learning prediction models aiming at each IMF component and the residual error RES, and the input of the GRU deep learning prediction models is the relevant meteorological data and the corresponding historical IMFs or residual error RES;
step S4, an Attention mechanism layer, wherein the input of the Attention mechanism layer is the output vector of the GRU layer, the probabilities corresponding to different eigenvectors are calculated according to the weight distribution principle, and a weight parameter matrix is updated and iterated to obtain a new vector;
and step S5, outputting the layer, and inputting the new vector into the full-connection layer to obtain a final prediction result.
Further, in step S1, performing filtering and denoising on the correlated meteorological data by using kalman filtering to correct measurement errors and human errors, where the kalman filtering estimates the current state by using a previous estimated value and a latest observed value of the current state under a linear unbiased minimum variance estimation criterion, determines an updated estimate of a new state according to the previous estimated value and the observed value of the current state, and adds interference of an external environment, where the kalman filtering is used to describe a linear random difference equation between the previous state and the current state and an observation equation of the current state are respectively,
xk=Axk-1+Buk-1+ωk
zk=Hxk-1+vk
wherein x is a state value, k is a current time state, k-1 is a previous time state, A is a transition matrix that linearly maps the previous state to the current state, B is a transition matrix that controls inputs to the current state, u is a control input, ω is a control valuekFor the system input random noise during state updating, z is the measured value, H is the conversion matrix from the current state variable to the observation variable, vkTo observe the noise vector.
Further, the kalman filter algorithm is performed recursively through a time update equation and a measurement update equation, the time update equation projects the current state variable as a priori estimate to the measurement update equation in time forward, the time update equation extrapolates the state variable and the error covariance from the time k-1 forward to the time k, the specific expression is,
wherein the content of the first and second substances,is an a posteriori estimate of the time k,is a priori estimate of the time of k,covariance is estimated a priori at time k, and Q is process noise covariance.
Further, the measurement update equation requires the calculation of the kalman gain, and then the a priori estimate and the new measured variable are combined to construct an improved a posteriori estimate, specifically expressed as,
wherein, PkEstimating covariance for a posteriori at time k, ZkAnd (4) for the posterior state estimation of the k step, after the calculation of the time updating equation and the measurement updating equation is completed, iterating the whole process again to obtain the denoising result of the related meteorological data.
Further, in the step S2, including,
s2.1, calculating all local maximum value points and minimum value points for the original time sequence x (t) of the solar radiation data, and obtaining an upper envelope e by adopting cubic spline function interpolation fittingmax(t) and lower envelope emin(t) sequence, and calculating the mean value sequence m of the upper and lower envelope sequences1(t),
S2.2, subtracting the mean value sequence m from the original time sequence1(t) obtaining a new data sequence h1(t),h1(t)=x(t)-m1(t);
S2.3, when the new sequence h1(t) if there are also a negative local maximum and a positive local minimum, it is determined that the IMF component is not present, and x (t) is set to h1(t) repeating the processes from S2.1 to S2.2 until the IMF condition is satisfied, to obtain a first IMF component;
s2.4, a sequence of remainders r from which the high-frequency components have been removedk(t) is obtained by the following formula,
rk(t)=x(t)-hk(t)
s2.5, corresponding the obtained residual items to x (t), repeating the steps from S2.1 to S2.4 to obtain the rest n IMF components IMF1(t),IMF2(t),…,IMFn(t) up to sequence rn(t) is a monotonic function, the original sequence is represented as,
the EMD decomposition is now complete.
Further, in said step S3, the input of the model is selected using the pearson correlation coefficient, which is defined as the quotient of the covariance and the standard deviation between two variables, reflecting the degree of correlation between the two variables, and the specific expression is,
correlation coefficient rhoX,YThe absolute value of which is less than 0.2, and finally 4 meteorological parameters are selected as the input of the predicted irradiance amplitude, namely the temperature, the relative humidity, the solar altitude and the wind speed.
Further, in step S3, the EMD layer output data is divided into a training set and a test set, the training set is input into the GRU deep learning prediction model, the GRU deep learning prediction model learns the extracted features to capture the internal change rule thereof, the GRU deep learning prediction model calculates to obtain an output vector, the specific expression is,
rt=σ(W(r)xt+U(r)ht-1)#
zt=σ(W(z)xt+U(z)ht-1)#
wherein x istFor input at time t, htHidden state at time t, ht-1Hidden state at time t-1, ztAnd rtRespectively an update gate and a reset gate,is input xtAnd the previous moment ht-1A collection of hidden states, σ and tanh being activation functions, W(r)And U(r)Training parameter matrix for resetting gates, W(z)And U(z)For updating the training parameter matrix of the door, U and W areAnd (3) obtaining a training parameter matrix when the matrix is obtained, wherein x is the number multiplication of the matrix.
Further, in the step S5, the predicted value of the solar radiation output by the output layer isComprehensively evaluating the performance of the prediction method, and using RMSE and NSE with specific expressions,
wherein, yiIs the true value of the solar radiation,for the predicted value of solar radiation, n is the total number of samples, the performance of the prediction method increases with decreasing RMSE and with increasing NSE.
Compared with the prior art, the method has the advantages that the EMD algorithm can gradually decompose signals with different frequencies according to the characteristics of data, the prediction models are respectively established for the decomposed subsequences, then the output linear weighting of each prediction model is carried out to obtain the final prediction result, when the input time sequence in the solar radiation prediction is too long, the GRU network easily has the problems of information loss and difficulty in modeling, the prediction accuracy of the model can be influenced, and the Attention mechanism distributes weight for the GRU neural network hidden layer output, so that the GRU hidden layer can highlight the key information in prediction;
furthermore, the forecasting method utilizes a gated recursion unit to capture dynamic changes of solar radiation data and a hidden nonlinear relation between meteorological factors, models the time dependence, decomposes an original solar radiation sequence, establishes a model for each subsequence, and utilizes a linear weighting method to remarkably improve the forecasting precision and stability of the mixed model;
further, the prediction method predicts the solar radiation 1-6 hours in advance, and the EMD-AGRU model is in a stable state under different prediction levels, which shows that the EMD-AGRU model not only can realize short-term prediction, but also can be used for long-term solar radiation prediction tasks.
Drawings
FIG. 1 is a flow chart of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention;
FIG. 2 is a Kalman filtering denoising result diagram of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention;
FIG. 3 is a diagram illustrating the decomposition result of the EMD layer of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention;
FIG. 4 is a correlation thermodynamic diagram of the EMD-GRU-Attention based solar radiation prediction method of the present invention;
FIG. 5 is a network structure diagram of a GRU deep learning prediction model of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention;
FIG. 6 is an RMSE evaluation index chart of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention;
FIG. 7 is an NSE evaluation index chart of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention.
Detailed Description
In order that the objects and advantages of the invention will be more clearly understood, the invention is further described below with reference to examples; it should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only for explaining the technical principle of the present invention, and do not limit the scope of the present invention.
It should be noted that in the description of the present invention, the terms of direction or positional relationship indicated by the terms "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings, which are only for convenience of description, and do not indicate or imply that the device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and thus, should not be construed as limiting the present invention.
Furthermore, it should be noted that, in the description of the present invention, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
Referring to fig. 1, which is a flowchart illustrating a solar radiation prediction method based on EMD-GRU-Attention according to the present invention, the present invention provides a solar radiation prediction method based on EMD-GRU-Attention, including,
step S1, inputting a layer, acquiring solar radiation data and related meteorological data, and preprocessing the solar radiation data and the related meteorological data;
step S2, an EMD layer decomposes the preprocessed solar radiation time-series data into a plurality of IMF components and a residual RES by utilizing EMD;
step S3, the GRU layer respectively establishes corresponding GRU deep learning prediction models aiming at each IMF component and the residual error RES, and the input of the GRU deep learning prediction models is the relevant meteorological data and the corresponding historical IMFs or residual error RES;
step S4, an Attention mechanism layer, wherein the input of the Attention mechanism layer is the output vector of the GRU layer, the probabilities corresponding to different eigenvectors are calculated according to the weight distribution principle, and a weight parameter matrix is updated and iterated to obtain a new vector;
and step S5, outputting the layer, and inputting the new vector into the full-connection layer to obtain a final prediction result.
Please refer to fig. 2, which is a graph of kalman filtering denoising results of the EMD-GRU-Attention-based solar radiation prediction method of the present invention,
specifically, in step S1, a kalman filter is used to filter and denoise the correlated meteorological data, correct measurement errors and human errors, the kalman filter uses a previous estimated value and a latest observed value of a current state to estimate the current state under a linear unbiased minimum variance estimation criterion, and determines an updated estimate of a new state according to the previous estimated value and the observed value of the current state, and adds interference of an external environment, wherein the kalman filter is used to describe a linear random difference equation between a previous state and a next state and an observation equation of the current state are respectively,
xk=Axk-1+Buk-1+ωk
zk=Hxk-1+vk
wherein x is a state value, k is a current time state, k-1 is a previous time state, A is a transition matrix that linearly maps the previous state to the current state, B is a transition matrix that controls inputs to the current state, u is a control input, ω is a control valuekFor the system input random noise during state updating, z is the measured value, H is the conversion matrix from the current state variable to the observation variable, vkTo observe the noise vector.
Specifically, the kalman filtering algorithm is performed recursively through a time update equation and a measurement update equation, the time update equation projects the current state variable as a priori estimate to the measurement update equation in time and forward, the time update equation calculates the state variable and the error covariance from the time k-1 forward to the time k, and the specific expression is,
wherein the content of the first and second substances,is an a posteriori estimate of the time k,is a priori estimate of the time of k,covariance is estimated a priori at time k, and Q is process noise covariance.
Specifically, the measurement update equation requires the calculation of the kalman gain, and then the a priori estimate and the new measured variable are combined to construct an improved a posteriori estimate, specifically expressed as,
wherein, PkEstimating covariance for a posteriori at time k, ZkAnd (4) for the posterior state estimation of the k step, after the calculation of the time updating equation and the measurement updating equation is completed, iterating the whole process again to obtain the denoising result of the related meteorological data.
Please refer to fig. 3, which is a diagram illustrating the decomposition result of the EMD layer of the solar radiation prediction method based on EMD-GRU-Attention according to the present invention,
specifically, in the step S2, including,
s2.1, calculating all local maximum value points and minimum value points for the original time sequence x (t) of the solar radiation data, and obtaining an upper envelope e by adopting cubic spline function interpolation fittingmax(t) and lower envelope emin(t) sequence, and calculating the mean value sequence m of the upper and lower envelope sequences1(t),
S2.2, subtracting the mean value sequence m from the original time sequence1(t) obtaining a new data sequence h1(t),h1(t)=x(t)-m1(t);
S2.3, when the new sequence h1(t) if there are also a negative local maximum and a positive local minimum, it is determined that the IMF component is not present, and x (t) is set to h1(t) repeating the processes from S2.1 to S2.2 until the IMF condition is satisfied, to obtain a first IMF component;
s2.4, a sequence of remainders r from which the high-frequency components have been removedk(t) is obtained by the following formula,
rk(t)=x(t)-hk(t)
s2.5, corresponding the obtained residual items to x (t), repeating the steps from S2.1 to S2.4 to obtain the rest n IMF components IMF1(t),IMF2(t),…,IMFn(t) up to sequence rn(t) is a monotonic function, the original sequence is represented as,
the EMD decomposition is now complete.
In step S2, the complex signal is decomposed by EMD and then composed of several independent Intrinsic Mode Function (IMF) components with different scale features and a remainder RES, and the solar radiation sequence is a time sequence with nonlinear and non-stationary characteristics, so that the solar radiation data is decomposed by empirical mode decomposition, thereby realizing individual prediction of different scale features, achieving an effect of non-interference, reducing the prediction difficulty, and improving the prediction accuracy.
Please continue to refer to fig. 4, which is a related thermodynamic diagram of the EMD-GRU-Attention based solar radiation prediction method of the present invention,
specifically, in the step S3, the input of the model is selected using the pearson correlation coefficient, which is defined as the quotient of the covariance and the standard deviation between two variables, reflecting the degree of correlation between the two variables, and the specific expression is,
correlation coefficient rhoX,YAnd removing parameters with the absolute value less than 0.2, and finally selecting 4 meteorological parameters as the input of the predicted irradiance amplitude, wherein the parameters are temperature, relative humidity, solar altitude and wind speed, and finally establishing an optimal solar total radiation prediction model with high accuracy and strong reliability.
Please refer to fig. 5, which is a diagram of a network structure of a GRU deep learning prediction model according to the solar radiation prediction method based on EMD-GRU-Attention of the present invention;
specifically, in step S3, the EMD layer output data is divided into a training set and a test set, the training set is input into the GRU deep learning prediction model, the GRU deep learning prediction model learns the extracted features to capture the internal change rule thereof, the GRU deep learning prediction model calculates to obtain an output vector, specifically, the expression is,
rt=σ(W(r)xt+U(r)ht-1)#
zt=σ(W(z)xt+U(z)ht-1)#
wherein x istFor input at time t, htHidden state at time t, ht-1Hidden state at time t-1, ztAnd rtRespectively an update gate and a reset gate,is input xtAnd the previous moment ht-1A collection of hidden states, σ and tanh being activation functions, W(t)And U(t)Training parameter matrix for resetting gates, W(z)And U(z)For updating the training parameter matrix of the door, U and W areAnd (3) obtaining a training parameter matrix when the matrix is obtained, wherein x is the number multiplication of the matrix.
In the step S3, the GRU and the LSTM have similar structures, but the GRU reduces the number of internal hidden states and thresholds, solves the problem of gradient explosion or disappearance caused by inherent memory in the conventional RNN, can effectively extract the nonlinear time series data feature, and shortens the training time of the model.
Please refer to fig. 6 and 7, wherein fig. 6 is an RMSE evaluation index chart of the EMD-GRU-Attention based solar radiation prediction method of the present invention, fig. 7 is an NSE evaluation index chart of the EMD-GRU-Attention based solar radiation prediction method of the present invention,
in step S5, the predicted value of the solar radiation output by the output layer isComprehensively evaluating the performance of the prediction method, and using RMSE and NSE with specific expressions,
wherein, yiIs the true value of the solar radiation,the predicted value of the solar radiation is shown, and n is the total number of samples;
please refer to table 1 and table 2, wherein table 1 shows the evaluation and comparison results of the solar radiation prediction method based on EMD-GRU-Attention and other single model prediction methods by NSE, table 2 shows the evaluation and comparison results of the solar radiation prediction method based on EMD-GRU-Attention and other single model prediction methods by RMSE,
the performance of the prediction method increases with decreasing RMSE and with increasing NSE.
TABLE 1
TABLE 2
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention; various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. A solar radiation prediction method based on EMD-GRU-Attention is characterized by comprising the following steps,
step S1, inputting a layer, acquiring solar radiation data and related meteorological data, and preprocessing the solar radiation data and the related meteorological data;
step S2, an EMD layer decomposes the preprocessed solar radiation time-series data into a plurality of IMF components and a residual RES by utilizing EMD;
step S3, the GRU layer respectively establishes corresponding GRU deep learning prediction models aiming at each IMF component and the residual error RES, and the input of the GRU deep learning prediction models is the relevant meteorological data and the corresponding historical IMFs or residual error RES;
step S4, an Attention mechanism layer, wherein the input of the Attention mechanism layer is the output vector of the GRU layer, the probabilities corresponding to different eigenvectors are calculated according to the weight distribution principle, and a weight parameter matrix is updated and iterated to obtain a new vector;
and step S5, outputting the layer, and inputting the new vector into the full-connection layer to obtain a final prediction result.
2. The method for solar radiation prediction based on EMD-GRU-Attention of claim 1, wherein in step S1, the relevant meteorological data is filtered and denoised by Kalman filtering to correct measurement errors and human errors, the Kalman filtering uses the previous estimation value and the latest observation value of the current state to estimate the current state under the linear unbiased minimum variance estimation criterion, the updated estimation of the new state is determined according to the previous estimation value and the observation value of the current state, and the interference of the external environment is added, wherein, the Kalman filtering is used to describe the linear random difference equation between the previous state and the current state and the observation equation of the current state are respectively,
xk=Axk-1+Buk-1+ωk
zk=Hxk-1+vk
wherein x is a state value, k is a current time state, k-1 is a previous time state, A is a transition matrix that linearly maps the previous state to the current state, B is a transition matrix that controls inputs to the current state, u is a control input, ω is a control valuekFor the system input random noise during state updating, z is the measured value, H is the conversion matrix from the current state variable to the observation variable, vkTo observe the noise vector.
3. The EMD-GRU-Attention-based solar radiation prediction method of claim 2, characterized in that the kalman filtering algorithm is performed recursively through a time update equation and a measurement update equation, the time update equation projecting the current state variable as a priori estimate and forward in time to the measurement update equation, the time update equation extrapolating forward from the time k-1 to the state variable and error covariance of the time k, the specific expression being,
4. The method of claim 3, wherein the measurement update equation requires Kalman gain calculations, and then the a priori estimates are combined with new measurement variables to construct an improved a posteriori estimate, as expressed in,
wherein, PkEstimating covariance for a posteriori at time k, ZkAnd (4) for the posterior state estimation of the k step, after the calculation of the time updating equation and the measurement updating equation is completed, iterating the whole process again to obtain the denoising result of the related meteorological data.
5. The method for solar radiation prediction based on EMD-GRU-Attention of claim 1, wherein in the step S2, comprising,
s2.1, calculating all local maximum value points and minimum value points for the original time sequence x (t) of the solar radiation data, and obtaining an upper envelope e by adopting cubic spline function interpolation fittingmax(t) and lower envelope emin(t) sequence, and calculating the mean value sequence m of the upper and lower envelope sequences1(t),
S2.2, subtracting the mean value sequence m from the original time sequence1(t) obtaining a new data sequence h1(t),h1(t)=x(t)-m1(t);
S2.3, when the new sequence h1(t) if there are also a negative local maximum and a positive local minimum, it is determined that the IMF component is not present, and x (t) is set to h1(t) repeating the processes from S2.1 to S2.2 until the IMF condition is satisfied, to obtain a first IMF component;
s2.4, a sequence of remainders r from which the high-frequency components have been removedk(t) is obtained by the following formula,
rk(t)=x(t)-hk(t)
s2.5, corresponding the obtained residual items to x (t), repeating the steps from S2.1 to S2.4 to obtain the rest n IMF components IMF1(t),IMF2(t),…,IMFn(t) up to sequence rn(t) is a monotonic function, the original sequence is represented as,
the EMD decomposition is now complete.
6. The EMD-GRU-Attention-based solar radiation prediction method of claim 1, wherein in the step S3, an input of a model is selected using a pearson correlation coefficient, which is defined as a quotient of a covariance and a standard deviation between two variables reflecting a degree of correlation between the two variables, and the specific expression is,
correlation coefficient rhoX,YThe absolute value of less than 0.2, and finally 4 meteorological parameters are selected as the input of the predicted irradiance amplitude, namely the temperature, the relative humidity and the solar altitudeAngle and wind speed.
7. The method of claim 6, wherein in step S3, the EMD layer output data is divided into a training set and a testing set, the training set is inputted into the GRU deep learning prediction model, the GRU deep learning prediction model learns the extracted features to capture the internal variation rule thereof, the GRU deep learning prediction model calculates an output vector, specifically expressed as,
rt=σ(W(r)xt+U(r)ht-1)#
zt=σ(W(z)xt+U(z)ht-1)#
wherein x istFor input at time t, htHidden state at time t, ht-1Hidden state at time t-1, ztAnd rtRespectively an update gate and a reset gate,is input xtAnd the previous moment ht-1A collection of hidden states, σ and tanh being activation functions, W(r)And U(r)Training parameter matrix for resetting gates, W(z)And U(z)For updating the training parameter matrix of the door, U and W areAnd (3) obtaining a training parameter matrix when the matrix is obtained, wherein x is the number multiplication of the matrix.
8. The method for predicting EMD-GRU-Attention-based solar radiation according to claim 1, wherein in the step S5, the predicted value of the solar radiation output by the output layer isComprehensively evaluating the performance of the prediction method, and using RMSE and NSE with specific expressions,
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110988946.3A CN113780636A (en) | 2021-08-26 | 2021-08-26 | Solar radiation prediction method based on EMD-GRU-Attention |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110988946.3A CN113780636A (en) | 2021-08-26 | 2021-08-26 | Solar radiation prediction method based on EMD-GRU-Attention |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113780636A true CN113780636A (en) | 2021-12-10 |
Family
ID=78839556
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110988946.3A Pending CN113780636A (en) | 2021-08-26 | 2021-08-26 | Solar radiation prediction method based on EMD-GRU-Attention |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113780636A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114611414A (en) * | 2022-05-10 | 2022-06-10 | 河北工业大学 | Solar radiation prediction method combining EMD and TCN |
CN114820038A (en) * | 2022-03-31 | 2022-07-29 | 中国农业银行股份有限公司 | User loss prediction method, device, equipment and medium |
CN116451598A (en) * | 2023-06-20 | 2023-07-18 | 杭州经纬信息技术股份有限公司 | Solar irradiance prediction method based on denoising diffusion probability model |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104376370A (en) * | 2014-10-28 | 2015-02-25 | 南京南瑞集团公司 | Dam horizontal displacement prediction method |
CN104700120A (en) * | 2015-03-23 | 2015-06-10 | 南京工业大学 | Method for extracting and classifying fMRI features based on adaptive entropy algorithm for projection clustering (APEC) |
CN104715261A (en) * | 2015-03-23 | 2015-06-17 | 南京工业大学 | FMRI dynamic brain function sub-network construction and parallel connection SVM weighted recognition method |
CN109886498A (en) * | 2019-03-01 | 2019-06-14 | 北京邮电大学 | A kind of EMD-GRU Short-Term Load Forecasting Method based on feature selecting |
CN110222714A (en) * | 2019-05-05 | 2019-09-10 | 河海大学 | A kind of sun based on ARMA and BP neural network always irradiates resource prediction method |
US20200041550A1 (en) * | 2016-10-05 | 2020-02-06 | Telecom Italia S.P.A. | Method and system for estimating energy generation based on solar irradiance forecasting |
CN112070311A (en) * | 2020-09-10 | 2020-12-11 | 天津大学 | Day-ahead light power prediction method based on similar day clustering and meteorological factor weighting |
CN112288193A (en) * | 2020-11-23 | 2021-01-29 | 国家海洋信息中心 | Ocean station surface salinity prediction method based on GRU deep learning of attention mechanism |
CN112766078A (en) * | 2020-12-31 | 2021-05-07 | 辽宁工程技术大学 | Power load level prediction method of GRU-NN based on EMD-SVR-MLR and attention mechanism |
CN113065526A (en) * | 2021-05-06 | 2021-07-02 | 吉林大学 | Electroencephalogram signal classification method based on improved depth residual error grouping convolution network |
-
2021
- 2021-08-26 CN CN202110988946.3A patent/CN113780636A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104376370A (en) * | 2014-10-28 | 2015-02-25 | 南京南瑞集团公司 | Dam horizontal displacement prediction method |
CN104700120A (en) * | 2015-03-23 | 2015-06-10 | 南京工业大学 | Method for extracting and classifying fMRI features based on adaptive entropy algorithm for projection clustering (APEC) |
CN104715261A (en) * | 2015-03-23 | 2015-06-17 | 南京工业大学 | FMRI dynamic brain function sub-network construction and parallel connection SVM weighted recognition method |
US20200041550A1 (en) * | 2016-10-05 | 2020-02-06 | Telecom Italia S.P.A. | Method and system for estimating energy generation based on solar irradiance forecasting |
CN109886498A (en) * | 2019-03-01 | 2019-06-14 | 北京邮电大学 | A kind of EMD-GRU Short-Term Load Forecasting Method based on feature selecting |
CN110222714A (en) * | 2019-05-05 | 2019-09-10 | 河海大学 | A kind of sun based on ARMA and BP neural network always irradiates resource prediction method |
CN112070311A (en) * | 2020-09-10 | 2020-12-11 | 天津大学 | Day-ahead light power prediction method based on similar day clustering and meteorological factor weighting |
CN112288193A (en) * | 2020-11-23 | 2021-01-29 | 国家海洋信息中心 | Ocean station surface salinity prediction method based on GRU deep learning of attention mechanism |
CN112766078A (en) * | 2020-12-31 | 2021-05-07 | 辽宁工程技术大学 | Power load level prediction method of GRU-NN based on EMD-SVR-MLR and attention mechanism |
CN113065526A (en) * | 2021-05-06 | 2021-07-02 | 吉林大学 | Electroencephalogram signal classification method based on improved depth residual error grouping convolution network |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114820038A (en) * | 2022-03-31 | 2022-07-29 | 中国农业银行股份有限公司 | User loss prediction method, device, equipment and medium |
CN114611414A (en) * | 2022-05-10 | 2022-06-10 | 河北工业大学 | Solar radiation prediction method combining EMD and TCN |
CN116451598A (en) * | 2023-06-20 | 2023-07-18 | 杭州经纬信息技术股份有限公司 | Solar irradiance prediction method based on denoising diffusion probability model |
CN116451598B (en) * | 2023-06-20 | 2023-09-05 | 杭州经纬信息技术股份有限公司 | Solar Irradiance Prediction Method Based on Denoising Diffusion Probability Model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113780636A (en) | Solar radiation prediction method based on EMD-GRU-Attention | |
EP3161527B1 (en) | Solar power forecasting using mixture of probabilistic principal component analyzers | |
CN108764539B (en) | Upstream and downstream water level prediction method for cascade power station | |
CN110888059B (en) | Charge state estimation algorithm based on improved random forest combined volume Kalman | |
Stephenson et al. | Bayesian inference for extremes: accounting for the three extremal types | |
CN112434848B (en) | Nonlinear weighted combination wind power prediction method based on deep belief network | |
CN111144644B (en) | Short-term wind speed prediction method based on variation variance Gaussian process regression | |
Mestav et al. | State estimation for unobservable distribution systems via deep neural networks | |
CN104021289B (en) | Non-Gaussian unsteady-state noise modeling method | |
CN113406521B (en) | Lithium battery health state online estimation method based on feature analysis | |
CN110991721A (en) | Short-term wind speed prediction method based on improved empirical mode decomposition and support vector machine | |
CN116861201A (en) | Power supply system based on artificial intelligence | |
CN110490366A (en) | Runoff forestry method based on variation mode decomposition and iteration decision tree | |
CN111931983A (en) | Precipitation prediction method and system | |
CN116106761A (en) | Real-time lithium ion battery electric quantity estimation method based on typical correlation analysis | |
Ramakrishna et al. | A Stochastic Model for Short-Term Probabilistic Forecast of Solar Photo-Voltaic Power | |
CN114091647A (en) | Solar 10.7 cm radio flow forecasting method based on BP neural network | |
CN113836823A (en) | Load combination prediction method based on load decomposition and optimized bidirectional long-short term memory network | |
CN113361782A (en) | Photovoltaic power generation power short-term rolling prediction method based on improved MKPLS | |
CN116722529A (en) | Short-term photovoltaic power prediction method and system | |
CN116167465A (en) | Solar irradiance prediction method based on multivariate time series ensemble learning | |
CN112417768B (en) | Wind power correlation condition sampling method based on vine structure Pair-Copula | |
Bantupalli et al. | Wind Speed forecasting using empirical mode decomposition with ANN and ARIMA models | |
CN114139777A (en) | Wind power prediction method and device | |
CN117875398B (en) | Nonlinear system ash bin identification method capable of learning pole allocation |
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 |