CN101877077A - Time series predicting model - Google Patents
Time series predicting model Download PDFInfo
- Publication number
- CN101877077A CN101877077A CN2009102287274A CN200910228727A CN101877077A CN 101877077 A CN101877077 A CN 101877077A CN 2009102287274 A CN2009102287274 A CN 2009102287274A CN 200910228727 A CN200910228727 A CN 200910228727A CN 101877077 A CN101877077 A CN 101877077A
- Authority
- CN
- China
- Prior art keywords
- time series
- chaos
- layer
- prediction
- parameter
- 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
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the field of the analysis of nonlinear time series, in particular to a time series predicting model. The network model comprises an input layer, a middle layer and an output layer, wherein the middle layer unit consists of a chaos operator with the chaos characteristic. The parameter of the chaos operator is trained and adjusted through a chaos optimization algorithm. By learning and training, the network predicts the value at a certain time in future with the previous known value of the time series, and gradually modifies the parameter of the chaos operator according a predicted error, thereby gradually having the information which is accordant with the regularity in the time series and completing the predicting function of the time series. Particularly, the model can effectively realize the multi-step prediction of the time series. The time series is mainly used for the field of the prediction and the analysis of the time series in the practical engineering.
Description
Technical field
The invention belongs to the Nonlinear Time Series Analysis field, relate to the network model that a class is applied to actual engineering time sequence prediction.The present invention provides a kind of new prediction network model---based on the prediction network model of chaos operator.
Background technology
The time series forecasting analytical technology has important use to be worth at numerous areas such as economy, meteorology, geology, the hydrology, military affairs, medical science.Science is correctly carried out forecast analysis to various real time sequences can produce huge economic benefit and and social benefit.Because real system has complicated nonlinear characteristic, the linear model and the nonlinear model that are used for time series analysis in early days all have certain limitation in theory analysis and practical application.In recent years, many artificial intelligence approaches are in being usually used in the seasonal effect in time series uncertainty analysis.Especially the research of chaology is predicted seasonal effect in time series and is played active promoting function.Chaotic motion be a kind of by deterministic system produce, to starting condition have sensitive dependence, never repeat, the recovery aperiodic motion, extensively be present in the various fields such as meteorology, the hydrology, medical science, electronics, information science.For example, equal alleged occurrence chaotic characteristic in the sequences such as the pseudo-random code in the electronic systems such as financial time series and radar, communication, extra large clutter.Because the non-linear nature of chaos utilizes the prediction of chaology search time sequence and analysis fundamentally to guarantee its feasibility.
Common forecasting method is normally carried out the chaotic dynamics analysis to time series at present, by means of neural network statement forecast model, utilizes the theoretical definite network parameter of phase space reconfiguration, thereby realizes seasonal effect in time series forecasting research.
The subject matter of this class Forecasting Methodology is that the parameter of phase space reconfiguration is difficult for choosing, simultaneously neural network exists in the forecasting process to actual engineering time sequence that training speed is slow, dynamics is abundant inadequately and is easy to be absorbed in problem such as local minimum, therefore such Forecasting Methodology has the better prediction result to gross data usually, and for practical project sequence prediction effect and not ideal enough.
Therefore design a kind of novel model that is used for time series forecasting and have important use value.
Summary of the invention
Technical matters to be solved by this invention is, designs a kind of prediction network model based on the chaos operator, can realize actual engineering time sequence is carried out analyses and prediction.
The technical solution adopted in the present invention is: a kind of prediction network model comprises three-decker: input layer, middle layer and output layer.Input layer is made up of m unit, and output layer is made of 1 unit.
The model middle layer is made of a plurality of chaos operators.The weights that are connected between each unit of input layer and middle layer all are made as 1, and the middle layer all is made as 1/m with the weights that are connected between the output layer, and the learning algorithm of network adopts chaos optimization method, are mainly used in to regulate each chaos operator parameter of middle layer.
The objective of the invention is to propose a kind of prediction network model based on the chaos operator, by learning training, network model can short-term have the dynamics similar to the time series of being predicted, finishes the seasonal effect in time series multi-step prediction.
Description of drawings
Fig. 1 is the structural representation of prediction network.
Fig. 2 is the program flow diagram of network training and prediction.
Fig. 3 carries out 1 step prediction result curve map for model of the present invention to loading by certain county's electric system moon, wherein solid line is a measured value, and dotted line is a predicted value.
Fig. 4 carries out 3 step prediction result curve maps for model of the present invention to loading by certain county's electric system moon, wherein solid line is a measured value, and dotted line is a predicted value.
Fig. 5 carries out 5 step prediction result curve maps for model of the present invention to loading by certain county's electric system moon, wherein solid line is a measured value, and dotted line is a predicted value.
Fig. 6 carries out 15 step prediction result curve maps for model of the present invention to loading by certain county's electric system moon, wherein solid line is a measured value, and dotted line is a predicted value.
Embodiment
Below in conjunction with embodiment and accompanying drawing the present invention is described in further detail.
The present invention is three layers of prediction network as shown in Figure 1.
If known time series x to be predicted
1, x
2..., x
n, prediction step is p, and input layer unit number m represents the information content that utilizes, and quantity is too many, causes information redundancy easily, and quantity causes information dropout very little easily, therefore is difficult for selecting.Select according to the phase space reconfiguration theory, embedding dimension should be greater than 2d+1, and d is a fractal dimension.Considering in the practical project sequence that the time interval between two elements is longer, in some cases, is 1 even select time delay, also may cause losing of information.Therefore choosing of parameter m should not be too small.For guaranteeing the possibility of prediction, the quantity of input layer unit should be greater than the step-length p of prediction at least.Therefore, the quantity of input layer unit is chosen according to following formula.
m=max{[2d+1],p} (1)
P is a prediction step in the following formula, and " [.] " is for rounding operation.
Choosing of delay time T can be chosen according to the auto correlation function method, and the auto correlation function of promptly selecting to make the time series collection is pairing time delay during zero passage first.
The excitation function in network middle layer is made of the chaos operator of following formula.
z
n+1=sinαz
n (2)
Iteration function z
N+1=sin α zx
nThe Lyapunov index increase gradually along with the increase of α, when α>2.74, the Lyapunov index of this mapping be on the occasion of, because positive Lyapunov index is the principal character of chaos, therefore, this iteration map o'clock just shows chaotic characteristic in α>2.74, and along with the increase of α, index of chaotic degree also constantly increases.Simultaneously, the Lyapunov index greater than 0 interval in, exist numerous period windows again, the subtle change of parameter alpha will produce very big influence to the state of mapping, show complicated dynamics.
The weights that are connected between each unit of input layer and middle layer all are made as 1, and the middle layer all is made as 1/m, i.e. w1 with the weights that are connected between the output layer
Ij=1, w2
Ij=1/m because the dynamic behavior of chaos operator is very complicated, utilize traditional learning algorithm to be difficult to the effective adjusting of realization to parameter, so network is selected the training study algorithm of chaos optimization algorithm as this network for use.The learning algorithm of network is mainly used in regulates each chaos operator parameter alpha of middle layer
i, make network short-term to have the dynamics similar to the time series of being predicted, finish seasonal effect in time series is effectively predicted.
The Logistic of learning algorithm selecting type (3) mapping is as the chaos mechanism of optimized Algorithm, and when μ=4, the Logistic mapping is in chaos state.y
0Select the random number between (0,1).
y
t+1=μy
t(1-y
t) (3)
The program step of network training and prediction is as follows:
A, initialization network parameter.
According to network structure, between (0,1), set chaos operator parameter alpha at random
i, sample sequence number k=1, i=(m-1) τ+k, parameter regulation amplitude λ=0.001.
B, computational grid output reach the error with expectation value.
With [x
i, x
I-τ..., x
I-(m-1) τ] in the fan-in network, computational grid output x '
I+pAnd with expectation value x
I+pBetween error E=| x '
I+p-x
I+p|.
C, initialization study parameter.Chaos operator sequence number l=1; E
Temp=E.
D, utilize formula (3) to calculate y
T+1
e、α
temp=α
l,α
l=α
l+λy
t+1,y
t=y
t+1。
F, with [x
i, x
I-τ..., x
I-(m-1) τ] in the fan-in network, computational grid output x "
I+p, and error E '=| x "
I+p-x
I+p|.
If g were E '<E
Temp, E then
Temp=E ' returns steps d.Otherwise to step h.
If h were E ' 〉=E
Temp, α then
l=α
Temp
I, utilize formula (3) to calculate y
T+1
j、α
temp=α
l,α
l=α
l-λy
t+1,y
t=y
t+1。
K, with [x
i, x
I-τ..., x
I-(m-1) τ] in the fan-in network, computational grid output x "
I+p, and error E '=| x "
I+p-x
I+p|.
If l were E '<E
Temp, E then
Temp=E ' returns step I.Otherwise to step m.
If m were E ' 〉=E
Temp, α then
l=α
Temp, l=l+1.
If n is l>m, then weights are regulated and are finished, and to step o, otherwise are back to steps d.
The prediction training of o, next given data.
K=k+1, i=(m-1) τ+k is if i+p≤n then returns step b; Otherwise go to step p.
The prediction of p, unknown data.
With [x
i, x
I-τ..., x
I-(m-1) τ] input in the network, computing time sequence next predicted value x ' constantly
I+p
Following certain value constantly of the given value prediction of network utilisation time series the last period, and according to predicated error roll-off network parameter gradually.Network is made up of a plurality of chaos operators, because chaos has the predictable characteristics of short-term, therefore, predicated error according to current known time sequence, dynamically regulate the parameter of chaos operator, improve precision of prediction gradually, prediction network short-term is had and the corresponding to dynamics of being predicted of time series, thus the forecast function of deadline sequence effectively.
Embodiment
The prediction of load value of the electric system moon has certain directive significance to the production practices activity.Model of the present invention to certain county's electric system moon load carry out multi-step prediction, be chosen as 3 time delay, gained the results are shown in Fig. 3 to-Fig. 6.The input layer unit number m=5 of network when wherein 1 step prediction, the prediction of 3 steps and 5 steps predictions, network input layer unit number m=15 during 15 steps predictions.
Because forecasting sequence is a month load sequence, so the time interval between two data is longer, be 1 month, and total data length is limited, data volume is less, therefore this class sequence is predicted to have great difficulty.Prediction network of the present invention can obtain predicting the outcome with real data is corresponding to, though certain error is arranged, the general trend that predicts the outcome is consistent with actual sequence, and works as prediction step more in short-term, and error all within the acceptable range.Simultaneously, along with the increase of prediction step, because legacy data has been difficult to the rule information in the following period of reflection, so prediction effect variation gradually.
This shows that model of the present invention can be realized seasonal effect in time series is effectively predicted in certain prediction step.
Claims (4)
1. a time series predicting model is characterized in that, comprising: 3-tier architecture;
Ground floor is an input layer, gets being chosen for of quantity m of input block:
m=max{[2d+1],p} (1)
D is the seasonal effect in time series fractal dimension in the following formula, and p is a prediction step, and " [.] " is for rounding operation;
The second layer is the middle layer, is made of a plurality of chaos operators, and chaos operator expression formula is:
z
n+1=sinαz
n (2)
The 3rd layer is output layer, is made of 1 unit.
2. time series predicting model according to claim 1 is characterized in that, input layer all is made as 1 with the weights that are connected between the middle layer, and the middle layer all is made as 1/m with the weights that are connected between the output layer.
3. time series predicting model according to claim 1, it is characterized in that, the parameter of middle layer chaos operator utilizes the chaos optimization algorithm to regulate training, predicated error according to current known time sequence, regulate the parameter of chaos operator, improve precision of prediction gradually, can make prediction network short-term have the dynamics identical with the time series of being predicted, thus the forecast function of deadline sequence effectively.
4. time series predicting model according to claim 1 is characterized in that, utilizes given data to form training sample, and training sample only uses once in the parameter regulation process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009102287274A CN101877077A (en) | 2009-11-25 | 2009-11-25 | Time series predicting model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009102287274A CN101877077A (en) | 2009-11-25 | 2009-11-25 | Time series predicting model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN101877077A true CN101877077A (en) | 2010-11-03 |
Family
ID=43019629
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009102287274A Pending CN101877077A (en) | 2009-11-25 | 2009-11-25 | Time series predicting model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101877077A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102495944A (en) * | 2011-11-11 | 2012-06-13 | 苏州大学 | Time series forecasting method and equipment and system adopting same |
CN103209005A (en) * | 2013-04-18 | 2013-07-17 | 西安电子科技大学 | Hopping sequence prediction system based on graphical model |
CN104657749A (en) * | 2015-03-05 | 2015-05-27 | 苏州大学 | Method and device for classifying time series |
CN105976026A (en) * | 2016-04-20 | 2016-09-28 | 天津工业大学 | Wind speed sequence prediction method based on associative neural network |
CN107516114A (en) * | 2017-08-28 | 2017-12-26 | 湖南大学 | A kind of time Series Processing method and device |
CN108551286A (en) * | 2018-05-02 | 2018-09-18 | 湖南大学 | A kind of AC servo motor scene Efficiency testing method and system |
CN108584592A (en) * | 2018-05-11 | 2018-09-28 | 浙江工业大学 | A kind of shock of elevator car abnormity early warning method based on time series predicting model |
CN110222387A (en) * | 2019-05-24 | 2019-09-10 | 北京化工大学 | The polynary drilling time sequence prediction method of integral CRJ network is leaked based on mixing |
CN110596199A (en) * | 2019-09-02 | 2019-12-20 | 安徽康佳同创电器有限公司 | Electronic nose, smell identification method and storage medium |
CN111191833A (en) * | 2019-12-25 | 2020-05-22 | 湖北美和易思教育科技有限公司 | Intelligent experiment process recommendation method and system based on neural network |
CN112268350A (en) * | 2020-10-22 | 2021-01-26 | 天津大学 | Air conditioner side load prediction method based on system delay |
-
2009
- 2009-11-25 CN CN2009102287274A patent/CN101877077A/en active Pending
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102495944A (en) * | 2011-11-11 | 2012-06-13 | 苏州大学 | Time series forecasting method and equipment and system adopting same |
CN102495944B (en) * | 2011-11-11 | 2014-11-05 | 苏州大学 | Time series forecasting method and equipment and system adopting same |
CN103209005A (en) * | 2013-04-18 | 2013-07-17 | 西安电子科技大学 | Hopping sequence prediction system based on graphical model |
CN103209005B (en) * | 2013-04-18 | 2016-06-01 | 西安电子科技大学 | The pre-examining system of frequency hop sequences of a kind of graphic based model |
CN104657749A (en) * | 2015-03-05 | 2015-05-27 | 苏州大学 | Method and device for classifying time series |
CN105976026A (en) * | 2016-04-20 | 2016-09-28 | 天津工业大学 | Wind speed sequence prediction method based on associative neural network |
CN105976026B (en) * | 2016-04-20 | 2018-04-03 | 天津工业大学 | Wind series Forecasting Methodology based on associative neural network |
CN107516114A (en) * | 2017-08-28 | 2017-12-26 | 湖南大学 | A kind of time Series Processing method and device |
CN108551286A (en) * | 2018-05-02 | 2018-09-18 | 湖南大学 | A kind of AC servo motor scene Efficiency testing method and system |
CN108584592A (en) * | 2018-05-11 | 2018-09-28 | 浙江工业大学 | A kind of shock of elevator car abnormity early warning method based on time series predicting model |
CN108584592B (en) * | 2018-05-11 | 2019-10-11 | 浙江工业大学 | A kind of shock of elevator car abnormity early warning method based on time series predicting model |
CN110222387A (en) * | 2019-05-24 | 2019-09-10 | 北京化工大学 | The polynary drilling time sequence prediction method of integral CRJ network is leaked based on mixing |
CN110222387B (en) * | 2019-05-24 | 2021-01-12 | 北京化工大学 | Multi-element drilling time sequence prediction method based on mixed leaky integration CRJ network |
CN110596199A (en) * | 2019-09-02 | 2019-12-20 | 安徽康佳同创电器有限公司 | Electronic nose, smell identification method and storage medium |
CN111191833A (en) * | 2019-12-25 | 2020-05-22 | 湖北美和易思教育科技有限公司 | Intelligent experiment process recommendation method and system based on neural network |
CN111191833B (en) * | 2019-12-25 | 2023-04-18 | 武汉美和易思数字科技有限公司 | Intelligent experiment process recommendation method and system based on neural network |
CN112268350A (en) * | 2020-10-22 | 2021-01-26 | 天津大学 | Air conditioner side load prediction method based on system delay |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101877077A (en) | Time series predicting model | |
Peng et al. | Effective long short-term memory with differential evolution algorithm for electricity price prediction | |
Marino et al. | Building energy load forecasting using deep neural networks | |
Li et al. | Prediction for tourism flow based on LSTM neural network | |
Li et al. | Urban traffic flow forecasting using Gauss–SVR with cat mapping, cloud model and PSO hybrid algorithm | |
Li et al. | T2F-LSTM method for long-term traffic volume prediction | |
Yang et al. | A New Strategy for Short‐Term Load Forecasting | |
KR102388215B1 (en) | Apparatus and method for predicting drug-target interaction using deep neural network model based on self-attention | |
Mostafavi et al. | A novel machine learning approach for estimation of electricity demand: An empirical evidence from Thailand | |
Li et al. | Multi-reservoir echo state computing for solar irradiance prediction: A fast yet efficient deep learning approach | |
Lin et al. | Temporal convolutional attention neural networks for time series forecasting | |
Rizwan et al. | Artificial intelligence based approach for short term load forecasting for selected feeders at madina saudi arabia | |
CN101763600A (en) | Land use supply and demand prediction method based on model cluster | |
CN114572229B (en) | Vehicle speed prediction method, device, medium and equipment based on graph neural network | |
Dariane et al. | Comparative analysis of evolving artificial neural network and reinforcement learning in stochastic optimization of multireservoir systems | |
CN107194460A (en) | The quantum telepotation recurrent neural network method of Financial Time Series Forecasting | |
Souhe et al. | A hybrid model for forecasting the consumption of electrical energy in a smart grid | |
Narvekar et al. | Weather forecasting using ANN with error backpropagation algorithm | |
Mehdizadeh Khorrami et al. | Forecasting heating and cooling loads in residential buildings using machine learning: A comparative study of techniques and influential indicators | |
Shu et al. | Multi-step-ahead monthly streamflow forecasting using convolutional neural networks | |
Liu et al. | Runoff prediction using a novel hybrid ANFIS model based on variable screening | |
Yan et al. | An improved grasshopper optimization algorithm for global optimization | |
Wang et al. | A multiple-parameter approach for short-term traffic flow prediction | |
CN102663493A (en) | Delaying nerve network used for time sequence prediction | |
Huang et al. | Deep learning model-transformer based wind power forecasting approach |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C53 | Correction of patent for invention or patent application | ||
CB02 | Change of applicant information |
Address after: 300387 Tianjin city Xiqing District West Binshui Road No. 399 Applicant after: Tianjin Polytechnic University Address before: 300160 Tianjin City Hedong District Forest Road No. 63 Applicant before: Tianjin Polytechnic University |
|
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20101103 |