CN114611414A - Solar radiation prediction method combining EMD and TCN - Google Patents

Solar radiation prediction method combining EMD and TCN Download PDF

Info

Publication number
CN114611414A
CN114611414A CN202210500606.6A CN202210500606A CN114611414A CN 114611414 A CN114611414 A CN 114611414A CN 202210500606 A CN202210500606 A CN 202210500606A CN 114611414 A CN114611414 A CN 114611414A
Authority
CN
China
Prior art keywords
solar radiation
data
historical data
tcn
processed
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210500606.6A
Other languages
Chinese (zh)
Inventor
薛桂香
徐志杰
闫文杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hebei University of Technology
Original Assignee
Hebei University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hebei University of Technology filed Critical Hebei University of Technology
Priority to CN202210500606.6A priority Critical patent/CN114611414A/en
Publication of CN114611414A publication Critical patent/CN114611414A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/12—Timing analysis or timing optimisation

Landscapes

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

Abstract

The invention discloses a solar radiation prediction method combining EMD and TCN, which comprises the steps of obtaining solar radiation and meteorological historical data; performing correlation analysis on the historical data to remove irrelevant weather historical data; performing filling missing value and abnormal value correction to obtain historical data to be processed; decomposing the solar radiation historical data to be processed by adopting EMD (empirical mode decomposition) to obtain a plurality of components; combining the meteorological historical data to be processed with each component respectively to obtain data to be input of each component; respectively carrying out normalization processing on each data to be input, and then respectively inputting the data to be input into respective TCN models to obtain predicted data; then, tensor accumulation is carried out on all the prediction data to obtain a prediction value of the solar radiation; and adjusting parameters of the TCN model, and performing inverse normalization to obtain a final prediction result value. The invention combines EMD and TCN, has the advantages of parallel computation, low memory consumption and the like, and improves the solar radiation prediction performance.

Description

Solar radiation prediction method combining EMD and TCN
Technical Field
The invention relates to the field of solar radiation prediction, in particular to a solar radiation prediction method combining EMD and TCN.
Background
Due to a series of problems such as exhaustion of fossil energy and air pollution, people pay more and more attention to development of new energy. Solar energy, one of new energy sources, is affected by factors such as regions and climate, and energy reserves of the solar energy are required to be reasonably planned and controlled no matter solar heating or photovoltaic power generation is required. In order to improve the utilization rate of solar energy, it is becoming more and more important to predict solar radiation reliably and accurately, and thus research and development of solar radiation prediction are being promoted continuously. The existing solar irradiance prediction methods are mainly divided into traditional methods (physical models, statistical models and the like) and artificial intelligence methods (machine learning and deep learning).
Although a physical model for predicting solar radiation based on historical time-series data can achieve a certain prediction effect, the physical model is susceptible to severe weather. The statistical model is limited by the complex non-linearity of the time series data itself of the solar radiation. Machine learning and deep learning can well reveal the nonlinear relation between input and output data to improve the prediction accuracy of the model, and meanwhile, some limitations still exist. Jiang et al propose a solar radiation prediction method based on a Support Vector Machine (SVM) of similar data, the SVM model can better approximate the solar radiation characteristics of low and high frequency bands, but for large-scale training samples (m-order matrix calculation), the training speed of the SVM is slow, and the SVM is sensitive to conditions such as missing data, kernel functions and the like. Yadav et al proposed a prediction model for RNN based on adaptive learning rate, but RNN has a problem of gradient explosion and cannot perform parallel computations well.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a solar radiation prediction method combining EMD and TCN.
The technical scheme for solving the technical problem is to provide a solar radiation prediction method combining EMD and TCN, and is characterized by comprising the following steps:
step 1, acquiring solar radiation historical data and meteorological historical data;
step 2, performing correlation analysis between the solar radiation historical data and the meteorological historical data, and removing irrelevant meteorological historical data to obtain meteorological historical data after correlation analysis;
step 3, filling missing values in the solar radiation historical data and the weather historical data after correlation analysis, and then correcting abnormal values to obtain solar radiation historical data to be processed and weather historical data to be processed;
step 4, decomposing the historical solar radiation data to be processed by adopting EMD to obtain a plurality of components; the component comprises a plurality of subsequences to be processed IMF and a residual error;
and 5, respectively combining the to-be-processed meteorological historical data for each component to obtain to-be-input data of each component through a formula (6)
Figure 717459DEST_PATH_IMAGE001
；
Figure 262841DEST_PATH_IMAGE002
（6）
In the formula (6), Z is t and the continuous time step before t isjA certain component of (a);
Figure 377428DEST_PATH_IMAGE003
successive time steps at and before time t arejThe ith weather historical data to be processed; k represents the number of types of meteorological historical data to be processed;
step 6, respectively carrying out normalization processing on each data to be input to obtain the data to be input after respective normalization processing;
step 7, inputting each normalized data to be input into each TCN model respectively to obtain the predicted data of each normalized data to be input
Figure 282936DEST_PATH_IMAGE004
As shown in formula (7); then, carrying out tensor accumulation on all the prediction data to obtain a prediction value of the solar radiation;
Figure 533788DEST_PATH_IMAGE005
（7）
in the formula (7), the reaction mixture is,
Figure 374705DEST_PATH_IMAGE006
representing a TCN model;
Figure 851954DEST_PATH_IMAGE004
prediction data for a component at time t + 1;
step 8, adjusting parameters of the TCN model to minimize the error between the predicted value of the solar radiation and the true value of the solar radiation, so as to obtain the adjusted predicted value of the solar radiation; and then performing inverse normalization on the adjusted predicted value of the solar radiation to obtain a final predicted result value.
Compared with the prior art, the invention has the beneficial effects that:
(1) according to the method, the EMD (empirical mode decomposition) and the TCN are combined, the EMD-TCN model is constructed to predict the solar radiation, the problems of large-scale calculation, gradient disappearance or gradient explosion can be effectively avoided, the advantages of parallel calculation, low memory consumption and the like are achieved, and the prediction performance of the solar radiation is improved to the maximum extent.
(2) According to the method, through correlation analysis, the correlation between solar radiation prediction and various environmental factors is utilized, and in the prediction process, the input data is reconstructed by integrating meteorological historical data with certain correlation, so that the EMD-TCN model can effectively utilize key environmental factors, the model prediction accuracy is improved, and the convolution network is facilitated to improve the expression capacity of complex characteristics.
(3) The TCN can accurately model and predict complex multi-feature, multi-scale, and nonlinear time series data, has a larger receptive field and a more stable gradient, achieves an excellent balance between performance accuracy and time to train the model for time series modeling, and can reduce the use of resources.
(4) The EMD has obvious advantages in nonlinear and non-stable time sequence analysis, solar radiation data are decomposed through the EMD, independent prediction of local features is achieved, the effect of mutual noninterference is achieved, the prediction difficulty is reduced, and the solar radiation prediction performance is improved.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is a graph showing the components obtained by EMD decomposition of Tianjin as a measurement site in example 1 of the present invention;
FIG. 3 is a graph showing the predicted solar radiation values of the present invention and other existing models at Tianjin measuring place and 72 hours measuring time in example 1 of the present invention;
FIG. 4 is a graph showing the predicted solar radiation values of the present invention and other prior models at the measurement site of the building door for 72 hours in example 1 of the present invention;
FIG. 5 is a graph of a solar radiation value prediction curve of the present invention and other existing models measured in Hangzhou at 72 hours in example 1 of the present invention;
FIG. 6 is a graph showing the predicted solar radiation values for the model of the present invention and other prior models, measured at Cheng-du and 72 hours in example 1 of the present invention;
FIG. 7 is a RMSE histogram of the present invention and other prior art models in example 1 of the present invention;
FIG. 8 is a MAE histogram of the present invention and other prior art models in example 1 of the present invention;
FIG. 9 is a MSE histogram of the present invention in example 1 and other prior art models;
FIG. 10 is a NSE histogram of the present invention in example 1 of the present invention and other prior art models.
Detailed Description
Specific examples of the present invention are given below. The specific examples are only for illustrating the present invention in further detail and do not limit the scope of the claims of the present invention.
The invention provides a solar radiation prediction method (short method) combining EMD and TCN, which is characterized by comprising the following steps:
step 1, acquiring solar radiation historical data and meteorological historical data;
preferably, in step 1, the Solar Radiation historical data and the meteorological historical data are obtained from the National Solar Radiation Database (NSRDB); the solar radiation historical data is global level irradiance; meteorological historical data includes temperature, cloud type, dew point, ozone, relative humidity, sun zenith angle, surface albedo, pressure, degradable water content, wind direction, and wind speed.
Step 2, performing correlation analysis between the solar radiation historical data and the meteorological historical data, and removing irrelevant meteorological historical data to obtain meteorological historical data after correlation analysis;
preferably, in step 2, the weather historical data after the correlation analysis includes temperature, relative humidity, solar zenith angle and wind speed.
Preferably, in step 2, the Correlation analysis uses a Pearson Correlation Coefficient (Pearson Correlation Coefficient), which is expressed as:
Figure 436519DEST_PATH_IMAGE007
（1）
in the formula (1), the reaction mixture is,
Figure 167977DEST_PATH_IMAGE008
covariance as X and Y;
Figure 179796DEST_PATH_IMAGE009
is the variance of X;
Figure 878761DEST_PATH_IMAGE010
is the variance of Y.
Preferably, in step 2, the pearson correlation coefficient is a measure of the degree of linear correlation, and the larger the absolute value of the correlation coefficient is, the stronger the correlation is, i.e. the correlation isThe stronger the correlation, the closer the coefficient is to 1 or-1; conversely, the closer the correlation coefficient is to 0, the weaker the correlation coefficient is.
Figure 267017DEST_PATH_IMAGE011
Then, they are completely correlated;
Figure 226883DEST_PATH_IMAGE012
highly correlated;
Figure 534237DEST_PATH_IMAGE013
moderate correlation;
Figure 110711DEST_PATH_IMAGE014
low degree of correlation;
Figure 646866DEST_PATH_IMAGE015
and are not relevant.
Step 3, filling missing values in the solar radiation historical data and the weather historical data after correlation analysis, and then correcting abnormal values to obtain solar radiation historical data to be processed and weather historical data to be processed;
preferably, in step 3, the specific operations of filling missing values are: in order to ensure the continuity of time series data (namely solar radiation historical data and meteorological historical data after correlation analysis), taking the mean value and the variance of adjacent values of missing values in the data; and randomly generating data obeying Gaussian distribution according to the mean value and the variance of the adjacent values, and filling the missing value by using the data.
The specific operations of outlier correction are: the outliers are treated as missing values and are then processed according to the specific operation that filled the missing values.
Step 4, decomposing the solar radiation historical data to be processed by adopting EMD to obtain a plurality of components; the component comprises a plurality of subsequences to be processed IMF (Intrinsic Mode Function) and a Residual error;
preferably, the specific steps of step 4 are as follows:
step 4.1, according to the treatmentSolar radiation historical data
Figure 726818DEST_PATH_IMAGE016
Finding all its maximum points to form the upper envelope sequence
Figure 80438DEST_PATH_IMAGE017
Finding all its minimum points to form the lower envelope sequence
Figure 501799DEST_PATH_IMAGE018
Then, the mean value of the upper envelope sequence and the lower envelope sequence is calculated to obtain a mean value envelope sequence
Figure 231858DEST_PATH_IMAGE019
As shown in formula (2):
Figure 776103DEST_PATH_IMAGE020
（2）
step 4.2, using the historical data of solar radiation to be processed
Figure 566204DEST_PATH_IMAGE016
Subtracted mean envelope sequence
Figure 976326DEST_PATH_IMAGE019
Obtaining an intermediate signal
Figure 244496DEST_PATH_IMAGE021
As shown in formula (3):
Figure 33461DEST_PATH_IMAGE022
（3）
step 4.3, judging the intermediate signal
Figure 869830DEST_PATH_IMAGE021
Whether the IMF condition is satisfied; if not, repeating the step 4.1 and the step 4.2 until the IMF condition is met, and generally repeating for a plurality of times; if satisfied, get the 1 st
Figure 377034DEST_PATH_IMAGE023
；
Step 4.4, using the historical data of solar radiation to be processed
Figure 809415DEST_PATH_IMAGE016
Obtained by subtracting step 4.3
Figure 984044DEST_PATH_IMAGE023
To obtain a new sequence
Figure 725735DEST_PATH_IMAGE024
As shown in formula (4):
Figure 720236DEST_PATH_IMAGE025
（4）
step 4.5, mixing
Figure 330209DEST_PATH_IMAGE024
Taken as the historical data of the solar radiation to be processed in step 4.1
Figure 483979DEST_PATH_IMAGE016
And then repeating the steps 4.1-4.4 until a sequence which can not be decomposed any more is obtained after iteration for m times
Figure 255626DEST_PATH_IMAGE026
And m
Figure 878368DEST_PATH_IMAGE027
Then the solar radiation historical data is processed
Figure 292032DEST_PATH_IMAGE016
Expressed as:
Figure 910095DEST_PATH_IMAGE028
（5）
in the formula (5), the reaction mixture is,mfor the number of IMFs obtained after decomposition，
Figure 6971DEST_PATH_IMAGE027
Represents the ith IMF;
Figure 976064DEST_PATH_IMAGE026
representing the residual error.
And 5, respectively combining the to-be-processed meteorological historical data for each component to obtain corresponding to-be-input data of each component through a formula (6)
Figure 537626DEST_PATH_IMAGE029
；
Figure 541354DEST_PATH_IMAGE002
（6）
In the formula (6), Z is t and the continuous time step before t isjA certain component of (a);
Figure 45017DEST_PATH_IMAGE003
successive time steps at and before time t arejThe ith meteorological historical data to be processed; k represents the number of types of meteorological historical data to be processed;
preferably, the specific steps of step 5 are as follows:
step 5.1, for the 1 st weather historical data M to be processed1Obtaining the kind M1The successive time steps at and before time t arejThe 1 st weather historical data to be processed of the type M is obtained11-dimensional tensor at time t
Figure 501406DEST_PATH_IMAGE030
；
Step 5.2, repeating the step 5.1 for the rest 2 nd to kth weather historical data to be processed to obtain k 1-dimensional tensors at the time t, namely the k 1-dimensional tensors
Figure 991293DEST_PATH_IMAGE031
；
Step 5.3, repeating the step 5.1-5.2 for a certain component obtained in the step 4 to obtain Z;
step 5.4, passing Z through the Concatenate layer of Keras deep learning framework and
Figure 724894DEST_PATH_IMAGE031
combining tensors to obtain the data to be input of the component at the time t
Figure 540403DEST_PATH_IMAGE029
；
And 5.5, repeating the steps 5.1-5.4 to obtain data to be input of all components at the time t, namely obtaining the data to be input of each subsequence to be processed IMF and a residual error at the time t.
Step 6, respectively carrying out normalization processing on each data to be input to obtain the data to be input after respective normalization processing;
preferably, in step 6, the normalization process adopts a Min-Max normalization method, so that the data with different characteristics satisfy the standard normal distribution of N (0, 1). After the original data are normalized, all indexes are in the same order of magnitude, and the method is suitable for comprehensive comparison and evaluation.
Step 7, inputting each normalized data to be input into each TCN model respectively to obtain the output of each normalized data to be input, namely the predicted data
Figure 844608DEST_PATH_IMAGE004
As shown in formula (7); then, tensor accumulation is carried out on all the prediction data through an Add layer of a Keras deep learning framework, and a prediction value of solar radiation is obtained;
Figure 138186DEST_PATH_IMAGE005
（7）
in the formula (7), the reaction mixture is,
Figure 726293DEST_PATH_IMAGE006
representing a TCN model;
Figure 447124DEST_PATH_IMAGE004
prediction data for a component at time t + 1;
preferably, in step 7, the TCN model performs convolution operations in strict time sequence, and the convolution operations only occur at time t and earlier elements of the previous layer, so as to obtain the output of the current layer at time t; on the basis of one-dimensional full convolution network and causal convolution, the filter is subjected to
Figure 878106DEST_PATH_IMAGE032
And sequences in hidden or input layers
Figure 100008DEST_PATH_IMAGE033
The operation F of the dilation convolution or dilation-causal convolution of an element s in a sequence x is defined as:
Figure 401677DEST_PATH_IMAGE034
（8）
in the formula (8) "
Figure 434355DEST_PATH_IMAGE035
"represents a convolution operation with an expansion coefficient of d; k is the size of the filter;
Figure 352632DEST_PATH_IMAGE036
for the elements in the sequence x to be,
Figure 722434DEST_PATH_IMAGE037
indicating past orientations, i.e.
Figure 32936DEST_PATH_IMAGE038
Not directly subtracting
Figure 361149DEST_PATH_IMAGE039
But rather represents the first number of forward numbers starting from the current s element
Figure 110931DEST_PATH_IMAGE039
A plurality of; when in use
Figure 815581DEST_PATH_IMAGE040
When the temperature of the water is higher than the set temperature,
Figure 747634DEST_PATH_IMAGE036
representing the current element s.
Step 8, adjusting parameters of the TCN model (namely optimizing the TCN model), so that the error between the predicted value of the solar radiation and the true value of the solar radiation (namely the actual measured value at the same time as the predicted value of the solar radiation) is minimized, and the adjusted predicted value of the solar radiation is obtained; then, the regulated predicted value of the solar radiation is subjected to inverse normalization to obtain a final result predicted value
Figure 981169DEST_PATH_IMAGE041
。
Preferably, in step 8, the result is restored by using a Min-Max standardization method in the reverse normalization.
Preferably, in step 8, the parameters include the number of layers, the number of nodes in the hidden layer, a time step, the number of training times, the batch size, the size of a convolution kernel, a filter size, an expansion coefficient, a padding type (padding used in convolution of the TCN model, and an optional padding type is causal or non-causal), an activation function, an optimizer, and a loss function; the optimizer adopts an Adam algorithm; the loss function adopts a mean square error, and the formula is as follows:
Figure 483826DEST_PATH_IMAGE042
（9）
in the formula (9), the reaction mixture is,
Figure 461009DEST_PATH_IMAGE043
the true value for the ith solar radiation,
Figure 749033DEST_PATH_IMAGE044
for the predicted value of the ith solar radiation, n represents the number of samples.
Example 1
Step 1, acquiring solar radiation historical data and meteorological historical data from a national solar radiation database, wherein the database contains enough hourly solar irradiation data on time and space scales. In order to better evaluate the prediction performance and generalization capability of the EMD-TCN model, the data set used in this embodiment is time-by-time data of 4 measurement locations (tianjin, xiamen, hangzhou, and chengdu) with large latitude and longitude spans and different environmental and climate conditions, and the time period is from 1 month and 1 day in 2019 to 12 months and 12 days in 2019 (i.e., sample data of 8760 hours);
step 2, taking Tianjin as an example, as shown in Table 1, performing correlation analysis on the solar radiation historical data and meteorological historical data, and keeping meteorological historical data with the absolute value of the Pearson correlation coefficient being more than or equal to 0.2, namely keeping temperature, relative humidity, solar zenith angle and wind speed;
Figure 153470DEST_PATH_IMAGE045
in table 1, T represents temperature, CT represents cloud type, DP represents dew point, O represents ozone, RH represents relative humidity, SZA represents solar zenith angle, SA represents surface albedo, P represents pressure, PW represents reducible water content, WD represents wind direction, and WS represents wind speed.
Step 3, filling missing values in the solar radiation historical data and the weather historical data after correlation analysis, and then correcting abnormal values to obtain solar radiation historical data to be processed and weather historical data to be processed;
step 4, decomposing the solar radiation historical data to be processed by adopting EMD (empirical mode decomposition), wherein the decomposition result is shown in figure 2, and obtaining a plurality of components, namely 9 subsequences to be processed IMF (intrinsic mode function) and 1 residual error;
step 5, for each component, combining the meteorological historical data to be processed respectively to obtain the data to be input of each component;
step 6, respectively carrying out normalization processing on each data to be input to obtain the data to be input after respective normalization processing;
step 7, inputting each normalized data to be input into each TCN model to obtain the predicted data of each normalized data to be input; then, tensor accumulation is carried out on all the prediction data through an Add layer of a Keras deep learning framework, and a prediction value of solar radiation is obtained;
step 8, adjusting parameters of the TCN model to minimize the error between the predicted value of the solar radiation and the true value of the solar radiation, so as to obtain the adjusted predicted value of the solar radiation; and then performing inverse normalization on the adjusted predicted value of the solar radiation to obtain a final predicted result value.
Through tests, the prediction results are compared by adopting four existing models, namely an EMD-TCN (empirical mode decomposition-convolutional network), a TCN (time convolutional network), an LSTM (long short term memory), an RNN (recurrent neural network) and a GRU (gated cyclic unit), wherein the parameters of the five models comprise: the number of layers is 1, the number of nodes in the hidden layer is 16, the time step is 3, the batch size is 64, the ReLU is adopted as an activation function, the mean square error is adopted as a loss function, Adam is adopted as an optimizer, and the training times are 50. The prediction results are shown in fig. 3-6, and fig. 3-6 respectively show the solar radiation value prediction curves of five prediction models in 4 measurement places for 72 hours, which shows that the prediction algorithm of the invention has better fitting effect and generalization capability.
In order to comprehensively evaluate the performance of each model, the Root Mean Square Error (RMSE), the Mean Absolute Error (MAE), the Mean Square Error (MSE), and the nash efficiency coefficient (NSE) were used to evaluate the performance index of each model, and the results are shown in fig. 7 to 10 and table 2.
Figure 268056DEST_PATH_IMAGE046
The data in table 2 are the average of 4 measurement sites. As can be seen from table 2, the present invention is superior to the other four existing models in terms of RMSE, MAE, MSE and NSE at four different measurement sites. In particular, the average NSE of the present invention is 0.935 and the average RMSE is 44W/m when compared to the TCN model2Average MAEIs 29.4W/m2While the mean NSE of the TCN model was 0.911 and the mean RMSE was 51.4W/m2The average MAE was 31.9W/m2Therefore, the combination of EMD on the basis of TCN is beneficial to improving the prediction accuracy.
The evaluation index of each model in 4 measurement points can be displayed more intuitively by performing analysis using the histogram (fig. 7 to 10). For RMSE, as shown in fig. 7, the height of each column of the present invention is substantially lower than that of the other four prior models, i.e., the deviation between the predicted value and the true value of the present invention is minimal. Similarly, for MSE and MAE, as shown in fig. 8 and 9, the column heights of the present invention are substantially lower than those of the other four prior models, indicating that the prediction effect of the present invention is good. For NSE, as shown in fig. 10, the corresponding values of all pillars are less than 1, and the heights of the pillars are higher than those of the other four existing models, i.e., the fitting effect of the present invention is better.
In conclusion, the EMD-TCN model is superior to other four existing models in terms of RMSE, MAE, MSE and NSE, shows good prediction effects in four measurement places with large latitude and longitude spans and different environmental and climatic conditions, and verifies that the EMD-TCN model has good generalization capability, prediction accuracy and applicability.
Nothing in this specification is said to apply to the prior art.

Claims (10)

1. A method for predicting solar radiation in conjunction with EMD and TCN, the method comprising the steps of:
step 1, acquiring solar radiation historical data and meteorological historical data;
step 2, performing correlation analysis between the solar radiation historical data and the meteorological historical data, and removing irrelevant meteorological historical data to obtain meteorological historical data after correlation analysis;
step 3, filling missing values in the solar radiation historical data and the weather historical data after correlation analysis, and then correcting abnormal values to obtain solar radiation historical data to be processed and weather historical data to be processed;
step 4, decomposing the solar radiation historical data to be processed by adopting EMD to obtain a plurality of components; the component comprises a plurality of subsequences to be processed IMF and a residual error;
and 5, respectively combining the to-be-processed meteorological historical data for each component to obtain to-be-input data of each component through a formula (6)
Figure 825848DEST_PATH_IMAGE001
；
Figure 274147DEST_PATH_IMAGE002
（6）
In the formula (6), Z is t and the continuous time step before t isjA certain component of (a);
Figure 460409DEST_PATH_IMAGE003
successive time steps at and before time t arejThe ith meteorological historical data to be processed; k represents the number of types of meteorological historical data to be processed;
step 6, respectively carrying out normalization processing on each data to be input to obtain the data to be input after respective normalization processing;
step 7, inputting each normalized data to be input into each TCN model respectively to obtain the predicted data of each normalized data to be input
Figure 121197DEST_PATH_IMAGE004
As shown in formula (7); then, carrying out tensor accumulation on all the prediction data to obtain a prediction value of the solar radiation;
Figure 466728DEST_PATH_IMAGE005
（7）
in the formula (7), the reaction mixture is,
Figure 180868DEST_PATH_IMAGE006
representing a TCN model;
Figure 979060DEST_PATH_IMAGE004
prediction data for a component at time t + 1;
step 8, adjusting parameters of the TCN model to minimize the error between the predicted value of the solar radiation and the true value of the solar radiation, so as to obtain the adjusted predicted value of the solar radiation; and then performing inverse normalization on the adjusted predicted value of the solar radiation to obtain a final predicted result value.
2. The method for predicting solar radiation by combining EMD and TCN according to claim 1, wherein in step 1, the solar radiation historical data and meteorological historical data are obtained from national solar radiation database; the solar radiation historical data is global level irradiance; meteorological historical data includes temperature, cloud type, dew point, ozone, relative humidity, sun zenith angle, surface albedo, pressure, degradable water content, wind direction, and wind speed.
3. The method for solar radiation prediction with EMD and TCN in combination as claimed in claim 1 wherein in step 2, the weather history data after correlation analysis includes temperature, relative humidity, sun zenith angle and wind speed.
4. The method for predicting solar radiation by combining EMD and TCN as claimed in claim 1, wherein in step 2, the correlation analysis uses Pearson's correlation coefficient, which is expressed as:
Figure 584485DEST_PATH_IMAGE007
（1）
in the formula (1), the reaction mixture is,
Figure 518943DEST_PATH_IMAGE008
is X andthe covariance of Y;
Figure 43465DEST_PATH_IMAGE009
is the variance of X;
Figure 453586DEST_PATH_IMAGE010
is the variance of Y.
5. The method for predicting solar radiation in combination with EMD and TCN according to claim 1, wherein the specific operations of filling missing values in step 3 are: for missing values in the data, taking the mean and variance of adjacent values of the missing values; randomly generating data which obeys Gaussian distribution according to the mean value and the variance of the adjacent values, and filling the missing value by adopting the data;
the specific operations of outlier correction are: the outliers are treated as missing values and are then processed according to the specific operation that filled the missing values.
6. The method for predicting solar radiation by combining EMD and TCN according to claim 1, wherein the specific steps of step 4 are as follows:
step 4.1, according to the historical data of the solar radiation to be processed
Figure 456177DEST_PATH_IMAGE011
Finding all its maximum points to form the upper envelope sequence
Figure 386087DEST_PATH_IMAGE012
Finding all its minimum points to form the lower envelope sequence
Figure 81511DEST_PATH_IMAGE013
Then, the mean value of the upper envelope sequence and the lower envelope sequence is calculated to obtain a mean value envelope sequence
Figure 854295DEST_PATH_IMAGE014
As shown in formula (2):
Figure 283746DEST_PATH_IMAGE015
（2）
step 4.2, using the historical data of solar radiation to be processed
Figure 458375DEST_PATH_IMAGE011
Mean-subtracted envelope sequence
Figure 934487DEST_PATH_IMAGE014
Obtaining an intermediate signal
Figure 194567DEST_PATH_IMAGE016
As shown in formula (3):
Figure 538961DEST_PATH_IMAGE017
（3）
step 4.3, judging the intermediate signal
Figure 692730DEST_PATH_IMAGE016
Whether the IMF condition is satisfied; if not, repeating the step 4.1 and the step 4.2 until the IMF condition is met; if so, obtaining the 1 st
Figure 995536DEST_PATH_IMAGE018
；
Step 4.4, using the historical data of solar radiation to be processed
Figure 87120DEST_PATH_IMAGE011
Obtained by subtracting step 4.3
Figure 500783DEST_PATH_IMAGE018
To obtain a new sequence
Figure 10524DEST_PATH_IMAGE019
As shown in formula (4):
Figure 218652DEST_PATH_IMAGE020
（4）
step 4.5, mixing
Figure 187745DEST_PATH_IMAGE019
Taken as the historical data of the solar radiation to be processed in step 4.1
Figure 14887DEST_PATH_IMAGE011
And then repeating the steps 4.1-4.4 until a sequence which can not be decomposed any more is obtained
Figure 753035DEST_PATH_IMAGE021
And m are
Figure 522277DEST_PATH_IMAGE022
Then the solar radiation historical data is processed
Figure 978666DEST_PATH_IMAGE011
Expressed as:
Figure 468554DEST_PATH_IMAGE023
（5）
in the formula (5), the reaction mixture is,mto the number of IMFs obtained after decomposition,
Figure 936575DEST_PATH_IMAGE022
represents the ith IMF;
Figure 752084DEST_PATH_IMAGE021
representing the residual error.
7. The method for predicting solar radiation by combining EMD and TCN according to claim 1, wherein the specific steps of step 5 are as follows:
step 5.1, treat the 1 st kindHistorical data M of vital signs1Obtaining the kind M1The successive time steps at and before time t arejThe 1 st weather historical data to be processed of the type M is obtained11-dimensional tensor at time t
Figure 53359DEST_PATH_IMAGE024
Step 5.2, repeating the step 5.1 for the 2 nd to the kth weather historical data to be processed to obtain
Figure 612517DEST_PATH_IMAGE025
；
Step 5.3, repeating the step 5.1-5.2 for a certain component obtained in the step 4 to obtain Z;
step 5.4, passing Z through the Concatenate layer of Keras deep learning framework and
Figure 200624DEST_PATH_IMAGE025
combining tensors to obtain the data to be input of the component at the time t
Figure 655876DEST_PATH_IMAGE026
；
And 5.5, repeating the steps 5.1-5.4 to obtain the data to be input of all the components at the time t.
8. The method for predicting solar radiation by combining EMD and TCN as claimed in claim 1, wherein in step 6, the normalization process adopts Min-Max normalization method, so that the data with different characteristics satisfy the standard normal distribution of N (0, 1).
9. The method for predicting solar radiation by combining EMD and TCN according to claim 1, wherein in step 7, convolution operation is performed in the TCN model according to the strict time sequence, and the convolution operation only occurs on the time t and the earlier elements of the previous layer, so that the output of the current layer at the time t is obtained; at one endBased on dimensional full convolution network and causal convolution, for filter
Figure 86857DEST_PATH_IMAGE027
And sequences in hidden or input layers
Figure 308760DEST_PATH_IMAGE028
The operation F of the dilation convolution or the dilation-causal convolution of the element s in the sequence x is defined as:
Figure 876008DEST_PATH_IMAGE029
（8）
in formula (8): "
Figure 643107DEST_PATH_IMAGE030
"represents a convolution operation with an expansion coefficient of d; k is the size of the filter;
Figure 561384DEST_PATH_IMAGE031
for the elements in the sequence x to be,
Figure 88443DEST_PATH_IMAGE032
the orientation in the past is indicated and,
Figure 244617DEST_PATH_IMAGE033
not directly subtracting
Figure 572831DEST_PATH_IMAGE034
But rather represents the first number of forward numbers starting from the current s element
Figure 588191DEST_PATH_IMAGE034
A plurality of; when the temperature is higher than the set temperature
Figure 292842DEST_PATH_IMAGE034
When the value is not less than 0, the reaction time is not less than 0,
Figure 693736DEST_PATH_IMAGE031
representing the current element s.
10. The method for solar radiation prediction with EMD and TCN in combination as claimed in claim 1 wherein in step 8, the parameters include number of layers, number of nodes in hidden layer, time step, number of training times, batch size, size of convolution kernel, filter size, expansion coefficient, fill type, activation function, optimizer and loss function; the optimizer adopts an Adam algorithm; the loss function adopts a mean square error, and the formula is as follows:
Figure 192851DEST_PATH_IMAGE035
（9）
in the formula (9), the reaction mixture is,
Figure 695507DEST_PATH_IMAGE036
the true value for the ith solar radiation,
Figure 938270DEST_PATH_IMAGE037
for the predicted value of the ith solar radiation, n represents the number of samples.
CN202210500606.6A 2022-05-10 2022-05-10 Solar radiation prediction method combining EMD and TCN Pending CN114611414A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210500606.6A CN114611414A (en) 2022-05-10 2022-05-10 Solar radiation prediction method combining EMD and TCN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210500606.6A CN114611414A (en) 2022-05-10 2022-05-10 Solar radiation prediction method combining EMD and TCN

Publications (1)

Publication Number Publication Date
CN114611414A true CN114611414A (en) 2022-06-10

Family

ID=81869147

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210500606.6A Pending CN114611414A (en) 2022-05-10 2022-05-10 Solar radiation prediction method combining EMD and TCN

Country Status (1)

Country Link
CN (1) CN114611414A (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942194A (en) * 2019-11-27 2020-03-31 徐州上若科技有限公司 Wind power prediction error interval evaluation method based on TCN
CN111382906A (en) * 2020-03-06 2020-07-07 南京工程学院 Power load prediction method, system, equipment and computer readable storage medium
CN112785051A (en) * 2021-01-14 2021-05-11 武汉纺织大学 Cloud resource prediction method based on combination of EMD and TCN
CN113657012A (en) * 2021-07-21 2021-11-16 西安理工大学 TCN and particle filter-based method for predicting residual life of key equipment
CN113780636A (en) * 2021-08-26 2021-12-10 河北工业大学 Solar radiation prediction method based on EMD-GRU-Attention

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942194A (en) * 2019-11-27 2020-03-31 徐州上若科技有限公司 Wind power prediction error interval evaluation method based on TCN
CN111382906A (en) * 2020-03-06 2020-07-07 南京工程学院 Power load prediction method, system, equipment and computer readable storage medium
CN112785051A (en) * 2021-01-14 2021-05-11 武汉纺织大学 Cloud resource prediction method based on combination of EMD and TCN
CN113657012A (en) * 2021-07-21 2021-11-16 西安理工大学 TCN and particle filter-based method for predicting residual life of key equipment
CN113780636A (en) * 2021-08-26 2021-12-10 河北工业大学 Solar radiation prediction method based on EMD-GRU-Attention

Similar Documents

Publication Publication Date Title
Tian Modes decomposition forecasting approach for ultra-short-term wind speed
CN110942194A (en) Wind power prediction error interval evaluation method based on TCN
Zhang et al. Multi-head attention-based probabilistic CNN-BiLSTM for day-ahead wind speed forecasting
CN111814956B (en) Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction
CN108053048A (en) A kind of gradual photovoltaic plant ultra-short term power forecasting method of single step and system
CN103117546A (en) Ultrashort-term slide prediction method for wind power
CN111815027A (en) Photovoltaic station generated power prediction method and system
CN112668611B (en) Kmeans and CEEMD-PE-LSTM-based short-term photovoltaic power generation power prediction method
Zhang et al. Interval prediction of ultra-short-term photovoltaic power based on a hybrid model
CN112862274A (en) WRF-LES and Deepar combined wind power prediction method
CN116702831A (en) Hybrid short-term wind power prediction method considering massive loss of data
CN117909888B (en) Intelligent artificial intelligence climate prediction method
CN116799796A (en) Photovoltaic power generation power prediction method, device, equipment and medium
CN113984198B (en) Shortwave radiation prediction method and system based on convolutional neural network
Sulaiman et al. Long-term solar irradiance forecasting using multilinear predictors
Li et al. A deep learning framework for non-stationary time series prediction
CN113111592A (en) Short-term wind power prediction method based on EMD-LSTM
Phan et al. Application of a new Transformer-based model and XGBoost to improve one-day-ahead solar power forecasts
CN117060407A (en) Wind power cluster power prediction method and system based on similar day division
Liu et al. Monthly wind distribution prediction based on nonparametric estimation and modified differential evolution optimization algorithm
Fuselero et al. Daily Solar Radiation Forecasting based on a Hybrid NARXGRU Network in Dumaguete, Philippines.
CN114611414A (en) Solar radiation prediction method combining EMD and TCN
CN115081681B (en) Wind power prediction method based on propset algorithm
CN115456286A (en) Short-term photovoltaic power prediction method
CN112036672B (en) New energy power generation ultra-short term power prediction method and system based on iterative correction

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20220610