CN112232565A - Two-stage time sequence prediction method, prediction system, terminal and medium - Google Patents
Two-stage time sequence prediction method, prediction system, terminal and medium Download PDFInfo
- Publication number
- CN112232565A CN112232565A CN202011103314.6A CN202011103314A CN112232565A CN 112232565 A CN112232565 A CN 112232565A CN 202011103314 A CN202011103314 A CN 202011103314A CN 112232565 A CN112232565 A CN 112232565A
- Authority
- CN
- China
- Prior art keywords
- stage
- elm
- prediction
- pso
- ceemdan
- 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
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 19
- 230000006641 stabilisation Effects 0.000 claims abstract description 6
- 238000011105 stabilization Methods 0.000 claims abstract description 6
- 238000010606 normalization Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 5
- 238000002156 mixing Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 abstract description 14
- 238000000354 decomposition reaction Methods 0.000 abstract description 12
- 230000003044 adaptive effect Effects 0.000 abstract description 3
- 230000002776 aggregation Effects 0.000 abstract description 2
- 238000004220 aggregation Methods 0.000 abstract description 2
- 239000002245 particle Substances 0.000 description 11
- 239000011159 matrix material Substances 0.000 description 10
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000012549 training Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 238000009499 grossing Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 229910052710 silicon Inorganic materials 0.000 description 1
- 239000010703 silicon Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012360 testing method 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
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (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)
Abstract
The invention belongs to the technical field of time sequence prediction, and discloses a two-stage time sequence prediction method, a prediction system, a terminal and a medium, wherein CEEMDAN is used for the stabilization treatment of a solar black number-of-months mean sequence; respectively establishing an ELM model for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm, and overlapping prediction results of components to obtain a first-stage prediction result; performing CEEMDAN-PSO-ELM modeling on the residual error obtained in the first stage to obtain a prediction result in the second stage; and summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result. The method combines the adaptive noise complete aggregation empirical mode decomposition in the multi-scale decomposition method with the extreme learning machine in the neural network algorithm, and further improves the prediction precision of a single ELM model on the basis of the traditional prediction model.
Description
Technical Field
The invention belongs to the technical field of time series prediction, and particularly relates to a two-stage time series prediction method, a prediction system, a terminal and a medium.
Background
In recent years, a nonlinear prediction model represented by a neural network is widely applied to prediction of a nonlinear system due to wide adaptive capacity and learning capacity, and the nonlinear and non-stationarity of a time sequence can be weakened by utilizing a multi-scale decomposition method, so that the prediction accuracy is effectively improved. However, in the existing prediction algorithm based on the neural network, two factors are not considered, one is how to optimize the model and parameters of the network in the prediction model, and the other is that the influence of the residual error on the prediction precision is not considered. In a prediction model based on a neural network, the selection of network parameters is crucial, and if the selection of the coefficients is not proper, the convergence speed of the network is slow, and the prediction accuracy is affected.
Through the above analysis, the problems and defects of the prior art are as follows: two factors have not been considered when using neural networks to predict time series: one is that it is difficult to ensure that the parameters of the network are better, so the obtained time series prediction result is not necessarily better; the second is that the effective information in the post-prediction residual is not considered.
The difficulty in solving the above problems and defects is: the input layer dimension and the hidden layer dimension of the neural network are small integers and are well determined, but the input weight and the hidden layer deviation are small fractions, and usually, the performance of the neural network is greatly influenced by small numerical value changes. How to find the optimal combination among thousands of parameter combinations is a difficult point to be solved. For residual errors, which contain less useful information, prediction of the residual error may not increase or decrease the accuracy of the final model if the selected model is incorrect.
The significance of solving the problems and the defects is as follows: if the network parameters can be determined by a better method, model precision fluctuation caused by randomness of the network parameters can be effectively reduced. If the model residual error can be effectively predicted, the information in the residual error can be fully utilized, and the accuracy of the model can be improved to a greater extent.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a two-stage time series prediction method, a prediction system, a terminal and a medium.
The invention is realized in such a way, a time sequence prediction method based on two-stage CEEMDAN-PSO-ELM is based on Matlab software, the stronger the nonlinearity and the non-stationarity of the time sequence is, the better the advantages of the invention compared with the traditional model can be embodied, the solar-black number-of-months mean sequence has very strong chaos characteristics, the following Ethernet-black time sequence is taken as an example to verify the superiority of the invention, and the time sequence prediction method based on two-stage CEEMDAN-PSO-ELM comprises the following steps:
the first stage, using CEEMDAN for the stabilization treatment of the sun black number-month average value sequence; and respectively establishing an ELM model for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm, and superposing each component prediction result to obtain a first-stage prediction result.
And in the second stage, CEEMDAN-PSO-ELM modeling is carried out on the residual error obtained in the first stage to obtain a prediction result in the second stage. And summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result.
Further, the time series prediction method based on the two-stage CEEMDAN-PSO-ELM comprises the following steps:
step one, decomposing the original sequence by using a CEEMDAN method to obtain IMF1,IMF2,…,IMFnAnd Res, decomposing the original sequence with stronger nonlinearity and non-stationarity into a plurality of subsequences with weaker nonlinearity and non-stationarity.
Step two, IMF1,IMF2,…,IMFnAnd Res, carrying out normalization processing to weaken the difference of each subsequence value range.
And step three, substituting each normalized subsequence into a PSO-ELM model for prediction, wherein a PSO algorithm optimizes the input layer dimension, the hidden layer dimension, the input weight and the hidden deviation of the ELM to obtain final network parameters, the output layer dimension of the network is set to be 1, the output result is subjected to inverse normalization to obtain a prediction result Y1,Y2,…Yn,Yn+1。
Step four, mixing Y1,Y2,…Yn,Yn+1Summing to obtain the first stage prediction result YsumSubtracting Y from the actual valuesumAn error sequence E is obtained.
Step five, decomposing the error sequence E by using a CEEMDAN method to obtain imf1,imf2,…,imfnAnd Res.
Step six, imf1,imf2,…,imfnAnd Res for normalization.
Step seven, each subsequence after normalization is brought into a PSO-ELM model for prediction to obtain a prediction result E1,E2,…Em,Em+1。
Step eight, mixing E1,E2,…Em,Em+1Summing to obtain the predicted result E of the second stagesum。
Step nine, predicting the result Y of the first stagesumAnd second stage prediction result EsumAnd summing to obtain a final predicted value.
Further, in step two, the IMF is performed1,IMF2,…,IMFnAnd Res, the method for carrying out normalization processing comprises the following steps:
wherein x ismaxAnd xminRespectively a maximum value and a minimum value in the input data; and y is the normalized input value.
Another object of the present invention is to provide a two-stage CEEMDAN-PSO-ELM-based time series prediction system for implementing the two-stage CEEMDAN-PSO-ELM-based time series prediction method, the two-stage CEEMDAN-PSO-ELM-based time series prediction system comprising:
the first-stage prediction result acquisition module is used for applying CEEMDAN to the stabilization treatment of the sun blackson number-of-months mean sequence; respectively establishing an ELM model for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm, and overlapping prediction results of components to obtain a first-stage prediction result;
the second-stage prediction result acquisition module is used for carrying out CEEMDAN-PSO-ELM modeling on the obtained residual error to obtain a second-stage prediction result;
and the prediction result summing module is used for summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result.
It is a further object of the invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the two-stage CEEMDAN-PSO-ELM based time series prediction method.
It is another object of the present invention to provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the two-stage CEEMDAN-PSO-ELM-based time series prediction method.
It is another object of the present invention to provide a computer program product stored on a computer readable medium, comprising a computer readable program for providing a user input interface to implement the two-stage CEEMDAN-PSO-ELM based time series prediction method when executed on an electronic device.
Another object of the present invention is to provide an information data processing terminal for implementing the two-stage CEEMDAN-PSO-ELM-based time series prediction method.
The invention also aims to provide an application of the two-stage CEEMDAN-PSO-ELM-based time series prediction method in solar black number-of-months mean prediction.
By combining all the technical schemes, the invention has the advantages and positive effects that: the time sequence prediction method based on the two-stage CEEMDAN-PSO-ELM combines the adaptive noise complete aggregation empirical mode decomposition (CEEMDAN) in the multi-scale decomposition method with the Extreme Learning Machine (ELM) in the neural network algorithm, and improves the precision of a single ELM model.
The invention optimizes the input layer dimension, the hidden layer dimension, the input weight and the hidden layer deviation of the ELM by using a Particle Swarm Optimization (PSO), thereby further improving the precision of the CEEMDAN-ELM model of the combined model; the two-stage prediction method fully utilizes error information to carry out error self-correction, and further improves the prediction precision on the basis of the traditional prediction model.
Technical effect or experimental effect of comparison. By comparing the predicted values with the true values of the sampled points 290-305, it can be seen that the model of the present invention is closer to the true values. As shown in fig. 9.
Drawings
FIG. 1 is a flow chart of a two-stage CEEMDAN-PSO-ELM-based time series prediction method according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of a two-stage CEEMDAN-PSO-ELM-based time series prediction method according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of the application of CEEMDAN to the smoothing process of the solar blackness mean value sequence according to the embodiment of the present invention.
Fig. 4 is a schematic diagram of optimizing ELM parameters of each sub-model by using a PSO algorithm according to an embodiment of the present invention.
Fig. 5 is a schematic diagram 1 of an intermediate process of CEEMDAN-PSO-ELM modeling on a residual obtained in the first stage according to an embodiment of the present invention.
Fig. 6 is a schematic diagram 2 of an intermediate process of CEEMDAN-PSO-ELM modeling of the residual error obtained in the first stage according to the embodiment of the present invention.
FIG. 7 is a diagram illustrating predicted results of models and other models according to the present invention.
Fig. 8 is a schematic diagram of comparing the error sequence of the model of the present invention with the error sequences of other models according to an embodiment of the present invention.
Fig. 9 is a comparison between the predicted value and the real value of the intercepted sampling points 290 to 305 provided in the embodiment of the present invention, and it can be seen that the model of the present invention is closer to the real value.
Fig. 10 is a network structure diagram of the extreme learning machine according to the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In view of the problems in the prior art, the present invention provides a two-stage time series prediction method, a prediction system, a terminal and a medium, which are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the two-stage CEEMDAN-PSO-ELM-based time series prediction method provided by the embodiment of the present invention includes the following steps:
s101, applying CEEMDAN to the smoothing treatment of the sun black number-month average sequence.
And S102, respectively establishing an ELM model for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm, and superposing each component prediction result to obtain a first-stage prediction result.
S103, performing CEEMDAN-PSO-ELM modeling on the residual error obtained in the first stage to obtain a prediction result in the second stage.
And S104, summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result.
Another object of the present invention is to provide a two-stage CEEMDAN-PSO-ELM-based time series prediction system for implementing the two-stage CEEMDAN-PSO-ELM-based time series prediction method, the two-stage CEEMDAN-PSO-ELM-based time series prediction system comprising:
the first-stage prediction result acquisition module is used for applying CEEMDAN to the stabilization treatment of the sun blackson number-of-months mean sequence; respectively establishing an ELM model for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm, and overlapping prediction results of components to obtain a first-stage prediction result;
the second-stage prediction result acquisition module is used for carrying out CEEMDAN-PSO-ELM modeling on the obtained residual error to obtain a second-stage prediction result;
and the prediction result summing module is used for summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result.
The present invention will be further described with reference to the following examples.
Example 1
The invention aims to realize the improvement of prediction precision on the basis of a traditional prediction model by using a two-stage CEEMDAN-PSO-ELM prediction method, which comprises the following specific steps:
step 2: blending IMF1,IMF2,…,IMFnAnd Res for normalization:
in the formula: x is the number ofmaxAnd xminRespectively a maximum value and a minimum value in the input data; and y is the normalized input value.
Step 4: adding Y1,Y2,…Yn,Yn+1Summing to obtain the first stage prediction result YsumSubtracting Y from the actual valuesumAn error sequence E is obtained.
step 6: imf1,imf2,…,imfnAnd Res are subjected to normalization processing;
Step 8: adding E1,E2,…Em,Em+1Summing to obtain the predicted result E of the second stagesum;
Step 9, predicting the result Y of the first stagesumAnd second stage prediction result EsumAnd summing to obtain a final predicted value.
Example 2
The embodiment of the invention adopts sun black number-month average Data, the Data is from an official website (http:// SIDC. oma. be/sil/datafiles) of a Solar effect Data Analysis Center (silicon) of a royal astronomical platform in Belgium, and the prediction flow is shown in FIG. 2. The first stage is as follows: the use of CEEMDAN for the smoothing of the mean of the number of months of the sun black is shown in FIG. 3; respectively establishing ELM models for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm as shown in table 1 and figure 4, and overlapping prediction results of each component to obtain a first-stage prediction result. And a second stage: and carrying out CEEMDAN-PSO-ELM modeling on the residual error obtained in the first stage, wherein the intermediate process is shown in FIG. 5, Table 2 and FIG. 6, and the prediction result of the second stage is obtained. And summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result. To verify the superiority of the present invention, the present invention was compared with Wavelet Neural Network (WNN), BP neural network (BPNN), ELM, CEEMDAN-ELM and CEEMDAN-PSO-ELM, respectively, and the predicted results of the proposed model and other models are shown in FIG. 7. As can be seen from fig. 7, there is a total of 9 peaks and troughs in the test set sequence (Part 1, Part 2, …, Part 9). When the number of solar black seeds is in a continuous rising or continuous falling stage, the fitting effect of each model is good. At the peaks and valleys, the single model fits to the discontinuities significantly off, while the combined model fits to the discontinuities more closely to the original data. The combined models are good in prediction effect, the quality of the combined models is difficult to see from the fitting graph, and calculation, analysis and comparison are needed to be further carried out through model evaluation indexes.
TABLE 1 optimization of ELM parameters for each sub-model using PSO algorithm
TABLE 2 CEEMDAN-PSO-ELM modelling of the residuals obtained in the first stage
The coefficient R is determined using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and coefficient R2The model was evaluated and the results are shown in table 3. As for a single model, the prediction result of the ELM on the sun black number-month-mean sequence is superior to WNN and BPNN, the CEEMDAN decomposition effectively improves the prediction precision of the ELM, the PSO further improves the prediction precision by optimizing on the basis of CEEMDAN-ELM, and the introduction of the two-stage method enables the precision of the model to be higher. The three error indexes of the model adopted by the invention are all smaller than those of other prediction models, and the decision coefficient is higher than those of other models, wherein the MAE is 0.2141, the RMSE2 is 0.3260, the MAPE is 0.0039, and R is2Is 1 (compared with ELM, MAE is improved by 79.80%, RMSE is improved by 77.13%, MAPE is improved by 80.79%. compared with CEEMDAN-ELM, MAE is improved by 43.66%, RMSE is improved by 37.87%, MAPE is improved by 45.83%. compared with CEEMDAN-PSO-ELM, MAE is improved by 37.07%, RMSE is improved by 34.30%, MAPE is improved by 32.76%).
Table 3 results of evaluation of model by determination coefficient R2
And (3) respectively subtracting the predicted values of the models from the real values of the sun-black number-of-months mean values to obtain error sequences of the models, comparing the error sequences of the models with the error sequences of other models as shown in FIG. 8, and obtaining that the models provided by the invention have smaller error sequence amplitude and are closer to 0 compared with other models.
Fig. 9 is a comparison between the predicted value and the real value of the intercepted sampling points 290 to 305 provided in the embodiment of the present invention, and it can be seen that the model of the present invention is closer to the real value.
The present invention is further described below with reference to specific examples.
Example 3
CEEMDAN decomposition principle
Huang proposes an Empirical Mode Decomposition (EMD) method that can decompose an arbitrary signal into characteristic Mode Functions (IMFs). The CEEMDAN decomposition method is provided by M.A. Colorminas on the basis of the study of Huang and other people, the CEEMDAN utilizes the characteristic of Gaussian white noise zero mean value to enable the decomposition effect of signal data to be more complete, and the specific processing process is as follows:
xi(n)=x(n)+γwi(n) (i=1,…,I) (1)
The first stage residue signal is then expressed as:
step 3: definition EkFor the k-th IMF component after EMD operation on the signal data, r1(n)+γ1E1(wi(n)) decomposing to obtain the second stage IMF component:
the second remaining component is:
step 4: so on the k-th residual component is
The k +1 th IMF component is
And Step 5, repeating Step 4 until the residual components can not meet the EMD decomposition condition or the iteration is finished. Finally, the target data sequence is decomposed into
Extreme learning machine principle
Extreme Learning Machines (ELMs) are yellow-extensive and teach a new algorithm for Single-hidden Layer Feed Forward Networks (SLFNs) proposed in 2006. Compared with the traditional neural network based on gradient learning, the ELM has the unique advantage that the optimal solution can be obtained through calculation by only setting the number of hidden layer units and then randomly generating the input connection weight vector and the hidden layer offset, so that the traditional multiple iterations are avoided. The learning speed is high, and the training time is greatly reduced, so that the generalization of the neural network is improved, and the accuracy of the network operation result is improved. The network structure is shown in fig. 10.
For any N non-identical samples (x)i,yi) Which isIn xi=[xi1,xi2,…,xin]T∈Rn,yi=[yi1,yi2,…,yik]T∈RkThe output of a feedforward neural network with l hidden nodes and an excitation function of G (x) is expressed as
In the formula: d is 1,2, …, N, j is 1,2, …, l; w is ai=[wi1,wi2,…,win]TIs the input weight connecting the input layer to the jth hidden layer node; beta is aj=[βj1,βj2,…,βjk]TIs the output weight connecting the jth hidden node to the output node; bjIs the offset value of the jth hidden node; w is aj·xiRepresents the vector wjAnd xiInner product of (d); the excitation function G (x) may be chosen as "Sigmoid", "Tansig", "Sine", or "RBF", etc.
Converting equation (9) to a matrix form, one can obtain:
Y=Hβ (10)
in the formula: h is the hidden layer output matrix of the network.
In the extreme learning machine algorithm, an input weight and a hidden layer can be randomly given without adjustment in the training process, a hidden layer matrix H is a determined matrix before training, and the training of the feedforward neural network is actually converted into a least square solution for solving an output weight matrixTo a problem of (a). The output weight matrix β can be expressed as
In the formula: h+Is the generalized inverse of matrix H.
According to the equations (9) to (11), the output weight matrix is determined by the input weight matrix and the hidden layer bias, and since the initial input weight matrix and the hidden layer bias are randomly given by the ELM, there may exist some input weight matrices and hidden layer biases being 0, which may cause some hidden layer nodes to be invalid. Therefore, in some practical applications, the accuracy and time of ELM training are affected by randomness.
Principle of particle swarm optimization algorithm
Particle Swarm Optimization (PSO) was first proposed by Kennedy and Eberhart in 1995. The PSO algorithm stems from the study of bird predation behavior, where the simplest and most efficient way for each bird to find the food is to search the surrounding area of the bird that is currently closest to the food.
Suppose that in a D-dimensional search space, there is a population X ═ of n particles (X)1,X2,…,Xn) Wherein the ith particle represents a vector X of dimension Di=[Xi1,Xi2,…,XiD]TRepresenting the position of the ith particle in the D-dimensional search space, and also representing a potential solution to the problem. The position X of each particle can be calculated according to the objective functioniA corresponding fitness value. Velocity of the ith particle is Vi=[Vi1,Vi2,…,ViD]TWith an individual extremum of Pi=[Pi1,Pi2,…,PiD]TGlobal extremum of the population is Pg=[Pg1,Pg2,…,PgD]T。
In each iteration process, the speed and position updating formula of the particle updating itself through the individual extremum and the global extremum is as follows:
in the formulae (12) and (13), ω is an inertial weight; d ═ 1,2, …, D; 1,2, …, n; k is the current iteration number; vidIs the velocity of the particle; c. C1And c2A non-negative constant, called the acceleration factor; r is1And r2Is distributed in [0,1 ]]A random number in between. To prevent a blind search for particles, it is generally proposed to limit their position and velocity to a certain interval [ -X [ ]max,Xmax]、[-Vmax,Vmax]。
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
The above description is only for the purpose of illustrating the present invention and the appended claims are not to be construed as limiting the scope of the invention, which is intended to cover all modifications, equivalents and improvements that are within the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A time series prediction method based on two-stage CEEMDAN-PSO-ELM is characterized by comprising the following steps:
carrying out stabilization treatment on the sun black number-month average sequence; substituting each decomposed subsequence into an ELM model, optimizing ELM parameters of each subsequence by using a PSO algorithm, and overlapping component prediction results of each subsequence to obtain a first-stage prediction result;
performing CEEMDAN-PSO-ELM modeling on the obtained residual error to obtain a prediction result of the second stage; and summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result.
2. The two-stage CEEMDAN-PSO-ELM-based time series prediction method of claim 1, wherein the two-stage CEEMDAN-PSO-ELM-based time series prediction method specifically comprises the following steps:
step one, decomposing the original sequence by using a CEEMDAN method to obtain IMF1,IMF2,…,IMFnAnd Res;
step two, IMF1,IMF2,…,IMFnAnd Res are subjected to normalization processing;
step three, each subsequence after normalization is brought into a PSO-ELM model for prediction, the output result is reversely normalized, and a prediction result Y is obtained1,Y2,…Yn,Yn+1;
Step four, mixing Y1,Y2,…Yn,Yn+1Summing to obtain the first stage prediction result YsumSubtracting Y from the actual valuesumObtaining an error sequence E;
step five, decomposing the error sequence E by using a CEEMDAN method to obtain imf1,imf2,…,imfnAnd Res;
step six, imf1,imf2,…,imfnAnd Res are subjected to normalization processing;
step seven, each subsequence after normalization is brought into a PSO-ELM model for prediction to obtain a prediction result E1,E2,…Em,Em+1;
Step eight, mixing E1,E2,…Em,Em+1Summing to obtain the predicted result E of the second stagesum;
Step nine, predicting the result Y of the first stagesumAnd second stage prediction result EsumAnd summing to obtain a final predicted value.
3. The two-stage CEEMDAN-PSO-ELM-based time series prediction method of claim 2, wherein in the second step, IMF is applied1,IMF2,…,IMFnAnd Res, the method for carrying out normalization processing comprises the following steps:
wherein x ismaxAnd xminRespectively a maximum value and a minimum value in the input data; and y is the normalized input value.
4. The two-stage CEEMDAN-PSO-ELM-based time series prediction method of claim 2, wherein in the third step, the PSO algorithm optimizes the input layer dimension, the hidden layer dimension, the input weights and the hidden bias of the ELM to obtain the final network parameters, and the output layer dimension of the network is set to 1.
5. A two-stage CEEMDAN-PSO-ELM-based time series prediction system for implementing the two-stage CEEMDAN-PSO-ELM-based time series prediction method according to any one of claims 1 to 4, wherein the two-stage CEEMDAN-PSO-ELM-based time series prediction system comprises:
the first-stage prediction result acquisition module is used for applying CEEMDAN to the stabilization treatment of the sun blackson number-of-months mean sequence; respectively establishing an ELM model for each decomposed subsequence, optimizing ELM parameters of each subsequence by using a PSO algorithm, and overlapping prediction results of components to obtain a first-stage prediction result;
the second-stage prediction result acquisition module is used for carrying out CEEMDAN-PSO-ELM modeling on the obtained residual error to obtain a second-stage prediction result;
and the prediction result summing module is used for summing the first-stage prediction result and the second-stage prediction result to obtain a final prediction result.
6. A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the two-stage CEEMDAN-PSO-ELM based time series prediction method of any of claims 1 to 4.
7. A computer-readable storage medium storing instructions which, when executed on a computer, cause the computer to perform the two-stage CEEMDAN-PSO-ELM-based time series prediction method according to any one of claims 1 to 4.
8. A computer program product stored on a computer readable medium, comprising a computer readable program for providing a user input interface for implementing the two-stage CEEMDAN-PSO-ELM based time series prediction method according to any one of claims 1 to 4 when executed on an electronic device.
9. An information data processing terminal, characterized in that the information data processing terminal is used for implementing the two-stage CEEMDAN-PSO-ELM-based time series prediction method of any one of claims 1 to 4.
10. Use of the two-stage CEEMDAN-PSO-ELM-based time series prediction method according to any one of claims 1 to 4 for solar black number-of-months mean prediction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011103314.6A CN112232565A (en) | 2020-10-15 | 2020-10-15 | Two-stage time sequence prediction method, prediction system, terminal and medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011103314.6A CN112232565A (en) | 2020-10-15 | 2020-10-15 | Two-stage time sequence prediction method, prediction system, terminal and medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112232565A true CN112232565A (en) | 2021-01-15 |
Family
ID=74113156
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011103314.6A Pending CN112232565A (en) | 2020-10-15 | 2020-10-15 | Two-stage time sequence prediction method, prediction system, terminal and medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112232565A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115946820A (en) * | 2023-03-15 | 2023-04-11 | 天津大学 | Training method of floating environment data prediction model, floating control method and equipment |
CN117669335A (en) * | 2024-02-01 | 2024-03-08 | 北京国星创图科技有限公司 | Space environment-based data mining method and system |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109299430A (en) * | 2018-09-30 | 2019-02-01 | 淮阴工学院 | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages |
-
2020
- 2020-10-15 CN CN202011103314.6A patent/CN112232565A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109299430A (en) * | 2018-09-30 | 2019-02-01 | 淮阴工学院 | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages |
Non-Patent Citations (3)
Title |
---|
崔金鑫;邹辉文;: "原油期货价格预测模型CEEMDAN-PSO-ELM", 计算机系统应用, no. 02, pages 1 - 10 * |
檀梦皎 等: "基于CEEMDAN理论和PSO-ELM模型的滑坡位移预测", 《地质科技情报》, vol. 38, no. 6, pages 165 - 175 * |
田中大;李树江;王艳红;王向东;沙毅;: "太阳黑子数平滑月均值的混合预测模型", 中国科学:物理学 力学 天文学, no. 11, pages 28 - 39 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115946820A (en) * | 2023-03-15 | 2023-04-11 | 天津大学 | Training method of floating environment data prediction model, floating control method and equipment |
CN117669335A (en) * | 2024-02-01 | 2024-03-08 | 北京国星创图科技有限公司 | Space environment-based data mining method and system |
CN117669335B (en) * | 2024-02-01 | 2024-05-07 | 北京国星创图科技有限公司 | Space environment data mining method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109948029B (en) | Neural network self-adaptive depth Hash image searching method | |
CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
Zhang et al. | Dynamic extreme learning machine and its approximation capability | |
CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
CN112884236B (en) | Short-term load prediction method and system based on VDM decomposition and LSTM improvement | |
CN113553755B (en) | Power system state estimation method, device and equipment | |
CN112733933B (en) | Data classification method and device based on unified optimization target frame graph neural network | |
CN115511177A (en) | Ultra-short-term wind speed prediction method based on INGO-SWGMN hybrid model | |
CN116316591A (en) | Short-term photovoltaic power prediction method and system based on hybrid bidirectional gating cycle | |
Peng et al. | Multi-step-ahead host load prediction with gru based encoder-decoder in cloud computing | |
CN112232565A (en) | Two-stage time sequence prediction method, prediction system, terminal and medium | |
CN112597610B (en) | Optimization method, device and equipment for lightweight design of mechanical arm structure | |
Wang et al. | The trend-fuzzy-granulation-based adaptive fuzzy cognitive map for long-term time series forecasting | |
CN113836823A (en) | Load combination prediction method based on load decomposition and optimized bidirectional long-short term memory network | |
CN114548532A (en) | VMD-based TGCN-GRU ultra-short-term load prediction method and device and electronic equipment | |
CN112231839A (en) | Intelligent exploration method for peak-tracking sampling of aircraft aiming at high time-consuming constraint | |
CN114676645B (en) | Non-stationary time sequence prediction method and system | |
CN109540089B (en) | Bridge deck elevation fitting method based on Bayes-Kriging model | |
CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
CN111695260B (en) | Material performance prediction method and system | |
CN116960978A (en) | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction | |
CN115619563A (en) | Stock price analysis method based on neural network | |
CN116415177A (en) | Classifier parameter identification method based on extreme learning machine | |
CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
CN115984025A (en) | Influence propagation estimation method and system based on deep learning graph network model |
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 |