CN105956614A - Time series semantization predicting method and time series semantization predicting system - Google Patents
Time series semantization predicting method and time series semantization predicting system Download PDFInfo
- Publication number
- CN105956614A CN105956614A CN201610261255.2A CN201610261255A CN105956614A CN 105956614 A CN105956614 A CN 105956614A CN 201610261255 A CN201610261255 A CN 201610261255A CN 105956614 A CN105956614 A CN 105956614A
- Authority
- CN
- China
- Prior art keywords
- semantization
- time series
- state
- matrix
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a time series semantization predicting method and a time series semantization predicting system. The time series semantization predicting method is characterized in that M time series of a system acquired, and every time series comprises N values, and then an M*N observation state matrix is formed, and by adopting a preset strategy, the values are divided into L categories, and every category is corresponding to a semantization label state; an M*N hidden state matrix corresponding to the observation state matrix is formed according to the corresponding relations between the values and the semantization label states; a corresponding hidden Markov model is acquired by calculating according to the observation state matrix and the hidden state matrix; the value of the current time series of the system is acquired, and is used as the current observation state value, and the semantization label state value probability distribution of the system at the next moment is calculated according to the hidden Markov model. The predicting result is acquired quickly and obviously.
Description
Technical field
The present invention relates to a kind of forecast analysis field, particularly relate to a kind of time series semantization Forecasting Methodology and system.
Background technology
Along with the fast development of Internet of Things Sensor Network, more and more industrial control systems are linked in Internet of Things, by miscellaneous
Sensor, timing, thus can be to the state of user's real-time exhibition controlled system by the state transfer of system to service end or high in the clouds
Index.If these data being combined corresponding timestamp store, then these historical datas are the formation of a time sequence
Row.People are by the pattern rule of sequence analysis time, it is possible to relatively accurately find out inherent statistical property and the development of corresponding system
Rule, by these statistical properties and Rule Model, may be used for forecasting system future behaviour.Facilitate user timely
Make a policy, it is possible to faster controlled system is adjusted, thus avoid controlled system to work under abnormal condition and
Cause damage.In time series forecasting currently used relatively broad be autoregressive moving-average model (Auto Regressive
Moving Average Model, ARMA), one of model parameter method high-resolution analysis of spectrum method.This method is that research is steady
The typical method of stochastic process rational spectrum, it is adaptable to the biggest class practical problem, but its parameter estimation is comparatively laborious, particularly exists
The when that parameter estimation being non-linear, it is difficult to try to achieve the accurate valuation of arma modeling parameter.Although being proposed theoretically
The best estimate method of arma modeling parameter, but there is computationally intensive and it cannot be guaranteed that the shortcoming of convergence in them.
In consideration of it, how to find a kind of faster effective time series forecasting scheme to obtain the just one-tenth that predicts the outcome that user needs
Those skilled in the art's problem demanding prompt solution.
Summary of the invention
The shortcoming of prior art in view of the above, it is an object of the invention to provide a kind of time series semantization Forecasting Methodology and
System, for solve Time Series Forecasting Methods of the prior art computationally intensive and it cannot be guaranteed that convergence problem.
For achieving the above object and other relevant purposes, the present invention provides a kind of time series semantization Forecasting Methodology, described time
Sequence semantization Forecasting Methodology includes: obtaining M time series of a system, each time series includes N number of numerical value, structure
Becoming M × N observer state matrix, M Yu N is natural number;Use preset strategy that described numerical value is divided into L classification, L
For natural number, each classification correspond to a semantization tag state respectively;Close according to numerical value is corresponding with semantization tag state
System builds a M × N hidden state matrix corresponding with described observer state matrix;According to described observer state matrix
And hidden state matrix calculus obtains corresponding HMM parameter, described HMM parameter includes: with
Described L classification observer state transition probability matrix one to one, L × L hidden state transition probability matrix;Obtain
The numerical value of the current time sequence of system, as Current observation state value, calculates described according to described HMM parameter
The semantization label-like state value probability distribution of system subsequent time.
Alternatively, described preset strategy includes: use K means clustering algorithm to carry out cluster analysis.
Alternatively, implementing of calculating observation state transition probability matrix includes: described seasonal effect in time series numerical value is continuous real number,
Respectively the value region that each classification is corresponding is carried out Density Estimator, draw the probability-distribution function of each class label, and then obtain
Observer state transition probability matrix to each classification.
Alternatively, the semantization label-like state value of described system subsequent time is calculated according to described HMM parameter
Implement and include: described Current observation state value is converted into semantization label-like state value, in conjunction with described HMM
Hidden state transition probability matrix in parameter calculates the probability distribution of the semantization label-like state value of described system subsequent time.
Alternatively, described time series semantization Forecasting Methodology also includes: calculates according to described HMM parameter and is
The probability distribution of each numerical value in system next moment.
Alternatively, described HMM parameter also includes 1 × L initial state probabilities matrix.
The present invention provides a kind of time series semantization prognoses system, and described time series semantization prognoses system includes: modeling number
According to acquisition module, for obtaining M time series of a system, each time series includes N number of numerical value, constitutes a M
× N observer state matrix, M Yu N is natural number;Using preset strategy that described numerical value is divided into L classification, L is natural number,
Each classification correspond to a semantization tag state respectively;Corresponding relation according to numerical value with semantization tag state builds one
A M × N hidden state matrix corresponding with described observer state matrix;Model parameter calculation module, for according to described
Observer state matrix and hidden state matrix calculus obtain corresponding HMM parameter, described HMM
Parameter includes: with described L classification observer state transition probability matrix one to one, L × L hidden state transfer generally
Rate matrix;Predict the outcome computing module, for obtaining the numerical value of the current time sequence of system as Current observation state value, root
The semantization label-like state value probability distribution of described system subsequent time is calculated according to described HMM parameter.
Alternatively, described preset strategy includes: use K means clustering algorithm to carry out cluster analysis.
Alternatively, implementing of calculating observation state transition probability matrix includes: described seasonal effect in time series numerical value is continuous real number,
Respectively the value region that each classification is corresponding is carried out Density Estimator, draw the probability-distribution function of each class label, and then obtain
Observer state transition probability matrix to each classification.
Alternatively, the semantization label-like state value of described system subsequent time is calculated according to described HMM parameter
Implement and include: described Current observation state value is converted into semantization label-like state value, in conjunction with described HMM
Hidden state transition probability matrix in parameter calculates the probability distribution of the semantization label-like state value of described system subsequent time.
Alternatively, described time series semantization Forecasting Methodology also includes: calculates according to described HMM parameter and is
The probability distribution of each numerical value in system next moment.
Alternatively, described HMM parameter also includes 1 × L initial state probabilities matrix.
As it has been described above, a kind of time series semantization Forecasting Methodology of the present invention and system, have the advantages that data
Use the mode of semantization label after sequence segment symbolization, be then predicted time series analyzing, due to many times,
What user really needed predicts the outcome is semantization label value, and the present invention can predicting the outcome with quick obtaining semantization label value,
Thus obtain quickly and predict the outcome intuitively.Instant invention overcomes traditional Time Series Forecasting Methods due to directly to original number
Value Data is predicted analyzing that the dimension not only processing data caused is higher causes the bigger forecast cost of computing cost high but also result
The problem lacking semantization.
Accompanying drawing explanation
Fig. 1 is shown as the schematic flow sheet of an embodiment of the time series semantization Forecasting Methodology of the present invention.
Fig. 2 is shown as the meter of the hidden state transition probability matrix of an embodiment of the time series semantization Forecasting Methodology of the present invention
Calculate principle schematic.
Fig. 3 is shown as the module diagram of an embodiment of the time series semantization prognoses system of the present invention.
Fig. 4 is shown as the application scenarios schematic diagram of an embodiment of the time series semantization prognoses system of the present invention.
Element numbers explanation
1 time series semantization prognoses system
11 modeling data acquisition modules
12 model parameter calculation modules
13 predict the outcome computing module
S1~S3 step
Detailed description of the invention
Below by way of specific instantiation, embodiments of the present invention being described, those skilled in the art can be by disclosed by this specification
Content understand other advantages and effect of the present invention easily.The present invention can also be added by the most different detailed description of the invention
To implement or application, the every details in this specification can also be based on different viewpoints and application, in the essence without departing from the present invention
Various modification or change is carried out under god.
It should be noted that the diagram provided in the present embodiment illustrates the basic conception of the present invention the most in a schematic way, the most graphic
In component count, shape and size time only display with relevant assembly in the present invention rather than is implemented according to reality draw, its reality
During enforcement, the kenel of each assembly, quantity and ratio can be a kind of random change, and its assembly layout kenel is likely to increasingly complex.
Invention applies hidden Markov model (Hidden Markov Model, HMM), to realize seasonal effect in time series semantization pre-
Survey.Hidden Markov model (Hidden Markov Model, HMM) is a kind of statistical model, and it is used for describing one containing hidden
Markov process containing unknown parameter.Its difficult point is to determine the implicit parameter of this process from observable parameter, then utilizes
These parameters are come for further analysis.The state of hidden Markov model can not observe directly, but can pass through observation vector sequence
Row are observed, each observation vector is to show as various state by some probability density distribution, each observation vector be by
One status switch with corresponding probability density distribution produces.So, hidden Markov model is a dual random process,
There is HMM and the display random function collection of certain status number.
The present invention provides a kind of time series semantization Forecasting Methodology.In one embodiment, as it is shown in figure 1, described time sequence
Row semantization Forecasting Methodology includes:
Step S1, obtain a system M time series, each time series includes N number of numerical value, constitute a M ×
N observer state matrix, M Yu N is natural number;Using preset strategy that described numerical value is divided into L classification, L is natural number,
Each classification correspond to a semantization tag state respectively;Corresponding relation according to numerical value with semantization tag state builds one
A M × N hidden state matrix corresponding with described observer state matrix.Described HMM parameter also includes one
Individual 1 × L initial state probabilities matrix.In one embodiment, described preset strategy includes: use K means clustering algorithm to enter
Row cluster analysis.To multiple seasonal effect in time series numerical value, according to the fine degree needed for prediction, K means clustering algorithm is used to carry out
Cluster analysis.Split data into multiple classification, and to each class declaration semantization label, such as: high, higher, normal, inclined
Low, low etc..Cluster is a process that data member the most similar in data set carries out taxonomic organization, and cluster is just
Being a kind of technology finding this immanent structure, clustering technique is commonly referred to as unsupervised learning.K mean cluster is foremost
Partition clustering algorithm, owing to succinct and efficiency make him become most widely used in all clustering algorithms.A given data point
Set and clusters number K needed, K is specified by user, and K mean algorithm is divided into K data repeatedly according to certain distance function
In individual cluster.
Step S2, obtains corresponding HMM ginseng according to described observer state matrix and hidden state matrix calculus
Number, described HMM parameter includes: with described L classification observer state transition probability matrix one to one, one
Individual L × L hidden state transition probability matrix.In one embodiment, calculating observation state transition probability matrix (is also referred to as seen
Survey state-transition matrix) implement and include: described seasonal effect in time series numerical value is continuous real number, corresponding to each classification respectively
Value region carry out Density Estimator, draw the probability-distribution function of each class label, and then obtain the observation shape of each classification
State transition probability matrix.In semantization sequence, state is actual implicit state in HMM, it is impossible to directly obtained by observation.
And time series itself can be obtained by directly observation.The method used due to the present invention, is directly to be mapped by some class
To the symbol that certain is set, such is collectively constituted by the sample point of apoplexy due to endogenous wind, and observation typically belongs to continuously in actual applications
Real number.Therefore, after hidden state determines, the numerical value of actual observation should obey such Density Distribution.It is not belonging to such
Numerical value, its probability density should be 0.Sometimes observer state transition probability matrix is also referred to as State-output matrix.
In one embodiment, when the time series that native system obtains is the observable numerical value in set of real numbers, this observer state
Probability matrix can not represent with discrete matrix, needs to use probability density function to represent.Density Estimator can be used
Mode, estimates to obtain the probability density function of each class, is the Observable state transition probability of each semantization symbol.At HMM
Middle hidden state transition probability matrix A describes the transition probability in HMM model between each hidden state, wherein
Aij=P (sj|si), 1≤i, j≤N, N are hidden state number, represent in t, state to be siUnder conditions of, when t+1
Quarter, state was sjProbability.Due to the observable numerical value o ∈ R in the data set of use in embodiment, therefore this observer state
The value of probability matrix can not calculate by discrete method, needs to use probability density function values to replace.The present invention uses core close
The mode that degree is estimated, estimates to obtain the probability density function of each class, is the Observable state transition probability of each semantization symbol.
In one embodiment, to primordial time series data collection, data are reduced according to the symbol label of first step definition with this
The possible state value of collection.And using new sequence as the hidden state collection of HMM, try to achieve the implicit shape of this time series data collection
State shift-matrix A.Each class is carried out Density Estimator (Kernel Density Estimation, KDE), show that it is distributed letter
Number, i.e. the probability-distribution function of certain symbol, in this, as the state transition probability matrix B of HMM.And try to achieve each sequence
Initial state probabilities distribution π, so far the HMM of data set can be expressed as λ=(π, A, B).In one embodiment, adopt
Numerical value synthesis being controlled time series data collection with K-means clustering algorithm clusters, and obtains 5 classifications, respectively correspondence
{ " low ", " on the low side ", " normally ", " higher ", " high " } 5 semantization symbols set.Density Estimator is used to obtain each
The observer state transitional provavility density of semantization character.And statistical computation goes out the hidden state transfer matrix of 5*5, with at the beginning of 1*5
Beginning probability matrix, is built into the HMM of data set.
Step S3, the numerical value of the current time sequence of acquisition system is as Current observation state value, according to described Hidden Markov mould
Shape parameter calculates the semantization label-like state value probability distribution of described system subsequent time.In one embodiment, according to described
HMM parameter calculates the implementing of semantization label-like state value of described system subsequent time and includes: by described
Current observation state value is converted into semantization label-like state value, shifts in conjunction with the hidden state in described HMM parameter
Probability matrix calculates the probability distribution of the semantization label-like state value of described system subsequent time.
In one embodiment, described time series semantization Forecasting Methodology also includes: according to described HMM parameter
Calculate the probability distribution of each numerical value in system next one moment.Specifically, described Current observation state value is changed Chinese idiom
Justiceization label-like state value, calculates described system in conjunction with the hidden state transition probability matrix in described HMM parameter
The probability distribution of the semantization label-like state value of subsequent time;Then, further combine in described HMM parameter
Observer state transition probability matrix calculate the probability distribution of each numerical value in next moment.
In one embodiment, described time series semantization Forecasting Methodology includes: 1, obtains 5 time serieses of a system,
Each time series includes 6 numerical value, constitutes 5 × 6 observer state matrixes.Described 5 × 6 observer state matrixes(each of which row represents a time series).
2, use preset strategy that described numerical value is divided into 3 classifications, each classification correspond to a semantization tag state respectively;
According to numerical value and the corresponding relation of semantization tag state build one corresponding with described observer state matrix one 5 × 6 hidden
Containing state matrix.Specifically, all numerical value carry out K-average (K-Means) cluster, and numerical value is divided into multiple class bunch.And according to
Practical situation uses different semantization symbols to different classes, is identified.Semantization identification procedure uses structure data set
Process is as follows, and all elements of matrix adds the vectorial V of a 1*30 (5*6=30)tIn, use KMEANS (Vt, 3) in,
Wherein 3 represent that this data set is divided into 3 classes by needs.3 class bunch C can be obtained1(1,2,3), C2(7,8), C3(14,15,16)。
Assume to use " low ", " in ", if " high " represent these three class, then just Ts can be also mapped onto hidden state matrix:
3, obtain corresponding HMM parameter according to described observer state matrix and hidden state matrix calculus, described
HMM parameter includes: (can be referred to as with described 3 classifications observer state transition probability matrix one to one
Observer state transfer matrix), 3 × 3 hidden state transition probability matrix (can be also simply referred to as hidden state transfer matrix).
Wherein it is possible to know " low " by 1,2,3}, " in " by { 7,8}, " high " is by { observer states such as 14,15,16} are mapped to.Right
Statistical method can be used to try to achieve each actual observed value in discrete series, in the accounting of his place class.In such as classification " low "
The number of times that " 1 " occurs is 1, " 2 " occurrence number is 2, and the number of times that " 3 " occur is 6.It is hereby achieved that " low "
Observer state transition probability matrix be 1/9,2/9,6/9}, be followed successively by when hidden state is low, observer state is 1,2,3
Probability.Other classifications obtain by that analogy " in " and the observer state transition probability matrix of " high ".Obtain " in " sight
Survey state transition probability matrix be 1/2,1/2}, be followed successively by when hidden state is middle, observer state is the probability of 7,8;Obtain
The observer state transition probability matrix of " high " be 4/9,4/9,1/9}, be followed successively by when hidden state is high, observer state is 14,
The probability of 15,16.Model exists three state, then it represents that 3 × 3 hidden state of the transformational relation between three kinds of states turn
Move probability matrix can with statistical calculation method as in figure 2 it is shown, thus correspondence obtains 3 × 3 following hidden state transition probability matrixs:
4, initial state probabilities matrix can be calculated further, each seasonal effect in time series original state is added up, permissible
Trying to achieve the sequence number that original state (value of the first row i.e. often gone in observer state matrix) is " low " is 3, original state
For " in " sequence number be 2, original state be the sequence number of " high " be 0.Initial state probabilities matrix the most now
For { 3/5,2/5,0/5}.
5, the numerical value of the current time sequence of acquisition system is as Current observation state value, according to described HMM parameter
Calculate the semantization label-like state value probability distribution of described system subsequent time.Such as, current time sequence is acquired
[7,3,3,7,8,14], use semantization rule can be mapped to [in, low, low, in, in, high], then predict lower a moment
Hidden state, i.e. hidden state matrix are transformed into the probability distribution of other states from height.Last state is high, then next
State is the probability 1/8 of " low ", for " in " probability 3/8, for the probability 4/8 of " high ".Lower two states are segmented into
3*2 kind situation, the probability for " low, low " is 1/8*2/6, and the probability for " low, in " is " 1/8*2/6 ", for " in,
Low " probability be 3/8*3/11, by that analogy.The probability distribution of each numerical value in the most next all right moment, then
Next numeral (observer state value) is the observer state transition probability matrix of hidden state probability and this state predicted
Product.Such as next status predication be " in " probability be 3/8, then probable value be the probability of 7 be 3/8*1/2, be 8
Probability is 3/8*1/2, by that analogy.
The present invention provides a kind of time series semantization prognoses system, and described time series semantization prognoses system can be applied as above
Described time series semantization Forecasting Methodology.In one embodiment, as it is shown on figure 3, described time series semantization is predicted
System includes modeling data acquisition module 11, model parameter calculation module 12 and the computing module 13 that predicts the outcome.Wherein:
Modeling data acquisition module 11 is for obtaining M time series of a system, and each time series includes N number of numerical value,
Constituting M × N observer state matrix, M Yu N is natural number;Use preset strategy that described numerical value is divided into L classification,
L is natural number, and each classification correspond to a semantization tag state respectively;Corresponding with semantization tag state according to numerical value
Relation builds a M × N hidden state matrix corresponding with described observer state matrix.Described HMM
Parameter also includes 1 × L initial state probabilities matrix.In one embodiment, described preset strategy includes: use K equal
Value clustering algorithm carries out cluster analysis.
Model parameter calculation module 12 is connected with modeling data acquisition module 11, for according to described observer state matrix and hidden
Being calculated corresponding HMM parameter containing state matrix, described HMM parameter includes: with described L
Individual classification observer state transition probability matrix one to one, L × L hidden state transition probability matrix.An enforcement
In example, implementing of calculating observation state transition probability matrix includes: described seasonal effect in time series numerical value is continuous real number, respectively
The value region that each classification is corresponding is carried out Density Estimator, draws the probability-distribution function of each class label, and then obtain each
The observer state transition probability matrix of individual classification.
The computing module 13 that predicts the outcome is connected with model parameter calculation module 12, for obtaining the number of the current time sequence of system
Value, as Current observation state value, calculates the semantization mark of described system subsequent time according to described HMM parameter
Sign state value probability distribution.In one embodiment, a period of time under described system is calculated according to described HMM parameter
The implementing of semantization label-like state value carved includes: described Current observation state value is converted into semantization label-like state value,
The semantization of described system subsequent time is calculated in conjunction with the hidden state transition probability matrix in described HMM parameter
The probability distribution of label-like state value.
In one embodiment, the computing module 13 that predicts the outcome described in is additionally operable to: according to described HMM parameter meter
The probability distribution of each numerical value in calculation system next one moment.Specifically, described Current observation state value is converted into semanteme
Change label-like state value, calculate under described system in conjunction with the hidden state transition probability matrix in described HMM parameter
The probability distribution of the semantization label-like state value in one moment;Then, further combine in described HMM parameter
Observer state transition probability matrix calculates the probability distribution of each numerical value in next moment.
In one embodiment, an application scenarios of described time series semantization prognoses system as shown in Figure 4, the described time
Sequence semantization prognoses system includes: obtain the time series of a system from cloud database, and by described time series number
Value carries out cluster and semantization label, obtains an observer state matrix and corresponding hidden state matrix (modeling data acquisition mould
Block 11 realizes);Then, described Hidden Markov (HMM) is built according to observer state matrix and corresponding hidden state matrix
Model parameter (model parameter calculation module 12 realizes);Then according to constructed described Hidden Markov (HMM) mould
Shape parameter is predicted analyzing, and predicts the outcome user's output, can include the semantization tag state of described system subsequent time
The probability distribution of value is or/and the probability distribution (computing module 13 that predicts the outcome realizes) of each numerical value in next moment.Will
The result of semantization exports user interface, is used for controlling by numeric type result return controlled system.And this result and model can
With multiplexing in the controlled system of same characteristic features.System can also continue to apply the time series obtained to carry out learning training, to structure
Build described Hidden Markov (HMM) model parameter to be adjusted.Use the pattern of cloud computing, the learning training to certain example
As a result, it is possible to be quickly applied under identical applied environment.Simultaneously by cloud computing center, can than single instance system faster
Collected more sample data.
In sum, a kind of time series semantization Forecasting Methodology and the system of the present invention will use after data sequence break sign
The mode of semantization label, is then predicted time series analyzing, thus obtains quickly and predict the outcome intuitively.So,
The present invention effectively overcomes various shortcoming of the prior art and has high industrial utilization.
The principle of above-described embodiment only illustrative present invention and effect thereof, not for limiting the present invention.Any it is familiar with this skill
Above-described embodiment all can be modified under the spirit and the scope of the present invention or change by the personage of art.Therefore, such as
All that in art, tool usually intellectual is completed under without departing from disclosed spirit and technological thought etc.
Effect is modified or changes, and must be contained by the claim of the present invention.
Claims (10)
1. a time series semantization Forecasting Methodology, it is characterised in that described time series semantization Forecasting Methodology includes:
Obtaining M time series of a system, each time series includes N number of numerical value, constitutes M × N observation
State matrix, M Yu N is natural number;Using preset strategy that described numerical value is divided into L classification, L is natural number, each
Classification correspond to a semantization tag state respectively;Corresponding relation according to numerical value and semantization tag state build one with
M × N hidden state matrix that described observer state matrix is corresponding;
Corresponding HMM parameter is obtained according to described observer state matrix and hidden state matrix calculus, described
HMM parameter includes: with described L classification observer state transition probability matrix one to one, L ×
L hidden state transition probability matrix;
The numerical value of the current time sequence of acquisition system is as Current observation state value, according to described HMM parameter
Calculate the semantization label-like state value probability distribution of described system subsequent time.
Time series semantization Forecasting Methodology the most according to claim 1, it is characterised in that: described preset strategy includes: use
K means clustering algorithm carries out cluster analysis.
Time series semantization Forecasting Methodology the most according to claim 1, it is characterised in that: calculating observation state transition probability square
Implementing of battle array includes: described seasonal effect in time series numerical value is continuous real number, enters the value region that each classification is corresponding respectively
Row Density Estimator, draws the probability-distribution function of each class label, and then obtains the observer state transition probability of each classification
Matrix.
Time series semantization Forecasting Methodology the most according to claim 1, it is characterised in that: according to described HMM
Parameter calculates the implementing of semantization label-like state value of described system subsequent time and includes: by described Current observation state
Value is converted into semantization label-like state value, in conjunction with the hidden state transition probability matrix meter in described HMM parameter
Calculate the probability distribution of the semantization label-like state value of described system subsequent time.
Time series semantization Forecasting Methodology the most according to claim 1, it is characterised in that: described time series semantization is predicted
Method also includes: calculate the probability of each numerical value in system next one moment according to described HMM parameter
Distribution.
6. a time series semantization prognoses system, it is characterised in that: described time series semantization prognoses system includes:
Modeling data acquisition module, for obtaining M time series of a system, each time series includes N number of number
Value, constitutes M × N observer state matrix, M Yu N is natural number;Use preset strategy that described numerical value is divided into L
Individual classification, L is natural number, and each classification correspond to a semantization tag state respectively;According to numerical value and semantization label
The corresponding relation of state builds a M × N hidden state matrix corresponding with described observer state matrix;
Model parameter calculation module, for obtaining the most hidden according to described observer state matrix and hidden state matrix calculus
Markov model parameter, described HMM parameter includes: with described L classification observer state one to one
Transition probability matrix, L × L hidden state transition probability matrix;
Predict the outcome computing module, for obtaining the numerical value of the current time sequence of system as Current observation state value, according to
Described HMM parameter calculates the semantization label-like state value probability distribution of described system subsequent time.
Time series semantization prognoses system the most according to claim 6, it is characterised in that: described preset strategy includes: use
K means clustering algorithm carries out cluster analysis.
Time series semantization prognoses system the most according to claim 6, it is characterised in that: calculating observation state transition probability square
Implementing of battle array includes: described seasonal effect in time series numerical value is continuous real number, enters the value region that each classification is corresponding respectively
Row Density Estimator, draws the probability-distribution function of each class label, and then obtains the observer state transition probability of each classification
Matrix.
Time series semantization prognoses system the most according to claim 6, it is characterised in that: according to described HMM
Parameter calculates the implementing of semantization label-like state value of described system subsequent time and includes: by described Current observation state
Value is converted into semantization label-like state value, in conjunction with the hidden state transition probability matrix meter in described HMM parameter
Calculate the probability distribution of the semantization label-like state value of described system subsequent time.
Time series semantization prognoses system the most according to claim 6, it is characterised in that: described time series semantization is predicted
Method also includes: calculate the probability of each numerical value in system next one moment according to described HMM parameter
Distribution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610261255.2A CN105956614A (en) | 2016-04-25 | 2016-04-25 | Time series semantization predicting method and time series semantization predicting system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610261255.2A CN105956614A (en) | 2016-04-25 | 2016-04-25 | Time series semantization predicting method and time series semantization predicting system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105956614A true CN105956614A (en) | 2016-09-21 |
Family
ID=56916821
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610261255.2A Pending CN105956614A (en) | 2016-04-25 | 2016-04-25 | Time series semantization predicting method and time series semantization predicting system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105956614A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106502398A (en) * | 2016-10-21 | 2017-03-15 | 浙江工业大学 | A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration |
CN107426033A (en) * | 2017-08-15 | 2017-12-01 | 深圳市盛路物联通讯技术有限公司 | A kind of method and apparatus that status predication is carried out to Internet of Things access terminal |
CN107886103A (en) * | 2016-09-29 | 2018-04-06 | 日本电气株式会社 | For identifying the method, apparatus and system of behavior pattern |
CN108229538A (en) * | 2016-12-13 | 2018-06-29 | 波音公司 | Communication tool system prediction meanss and method |
CN108304446A (en) * | 2017-12-07 | 2018-07-20 | 河南电力医院 | A kind of visable representation method of health examination physiological time sequence data, storage medium |
CN109993305A (en) * | 2018-01-03 | 2019-07-09 | 成都二十三魔方生物科技有限公司 | Ancestral source polymorphism prediction technique based on big data intelligent algorithm |
CN114152454A (en) * | 2020-09-08 | 2022-03-08 | 中国科学院上海高等研究院 | Mechanical equipment fault diagnosis method based on CEEMDAN-CSE model and establishment method of model |
CN115174130A (en) * | 2022-03-10 | 2022-10-11 | 中国科学院沈阳自动化研究所 | HMM-based AGV semantic attack detection method |
CN116340796A (en) * | 2023-05-22 | 2023-06-27 | 平安科技(深圳)有限公司 | Time sequence data analysis method, device, equipment and storage medium |
-
2016
- 2016-04-25 CN CN201610261255.2A patent/CN105956614A/en active Pending
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107886103B (en) * | 2016-09-29 | 2023-12-08 | 日本电气株式会社 | Method, device and system for identifying behavior patterns |
CN107886103A (en) * | 2016-09-29 | 2018-04-06 | 日本电气株式会社 | For identifying the method, apparatus and system of behavior pattern |
CN106502398B (en) * | 2016-10-21 | 2019-01-29 | 浙江工业大学 | A kind of semantization activity recognition method based on Multi-view Integration study |
CN106502398A (en) * | 2016-10-21 | 2017-03-15 | 浙江工业大学 | A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration |
CN108229538A (en) * | 2016-12-13 | 2018-06-29 | 波音公司 | Communication tool system prediction meanss and method |
CN107426033A (en) * | 2017-08-15 | 2017-12-01 | 深圳市盛路物联通讯技术有限公司 | A kind of method and apparatus that status predication is carried out to Internet of Things access terminal |
CN108304446A (en) * | 2017-12-07 | 2018-07-20 | 河南电力医院 | A kind of visable representation method of health examination physiological time sequence data, storage medium |
CN109993305A (en) * | 2018-01-03 | 2019-07-09 | 成都二十三魔方生物科技有限公司 | Ancestral source polymorphism prediction technique based on big data intelligent algorithm |
CN109993305B (en) * | 2018-01-03 | 2023-01-03 | 成都二十三魔方生物科技有限公司 | Ancestral polymorphism prediction method based on big data artificial intelligence algorithm |
CN114152454A (en) * | 2020-09-08 | 2022-03-08 | 中国科学院上海高等研究院 | Mechanical equipment fault diagnosis method based on CEEMDAN-CSE model and establishment method of model |
CN114152454B (en) * | 2020-09-08 | 2024-03-22 | 中国科学院上海高等研究院 | Mechanical equipment fault diagnosis method based on CEEMDAN-CSE model and establishment method of model |
CN115174130A (en) * | 2022-03-10 | 2022-10-11 | 中国科学院沈阳自动化研究所 | HMM-based AGV semantic attack detection method |
CN115174130B (en) * | 2022-03-10 | 2023-06-20 | 中国科学院沈阳自动化研究所 | AGV semantic attack detection method based on HMM |
CN116340796A (en) * | 2023-05-22 | 2023-06-27 | 平安科技(深圳)有限公司 | Time sequence data analysis method, device, equipment and storage medium |
CN116340796B (en) * | 2023-05-22 | 2023-12-22 | 平安科技(深圳)有限公司 | Time sequence data analysis method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105956614A (en) | Time series semantization predicting method and time series semantization predicting system | |
Radiuk | Impact of training set batch size on the performance of convolutional neural networks for diverse datasets | |
EP3948437B1 (en) | Predictive classification of future operations | |
CN108062561A (en) | A kind of short time data stream Forecasting Methodology based on long memory network model in short-term | |
CN108647643B (en) | Packed tower flooding state online identification method based on deep learning | |
CN110334839A (en) | Flight is delayed prediction technique, device, equipment and storage medium | |
CN112180471B (en) | Weather forecasting method, device, equipment and storage medium | |
CN106778838A (en) | A kind of method for predicting air quality | |
CN113486578A (en) | Method for predicting residual life of equipment in industrial process | |
CN107423190A (en) | A kind of daily record data points to recognition methods and device extremely | |
CN113130014B (en) | Rare earth extraction simulation method and system based on multi-branch neural network | |
CN112185104A (en) | Traffic big data restoration method based on countermeasure autoencoder | |
CN110321940A (en) | The feature extraction of aircraft telemetry and classification method and device | |
CN109508788A (en) | A kind of SDN method for predicting based on arma modeling | |
Zougagh et al. | Prediction models of demand in supply chain | |
Tian et al. | Spatio-temporal position prediction model for mobile users based on LSTM | |
CN103208036B (en) | A kind of short-term load forecasting method of electrically-based user data | |
Manoj et al. | FWS-DL: forecasting wind speed based on deep learning algorithms | |
CN103106329B (en) | A kind of training sample constructed in groups method for SVR short-term load forecasting | |
Abid et al. | Multi-directional gated recurrent unit and convolutional neural network for load and energy forecasting: A novel hybridization | |
CN117196185A (en) | Military system demand generation method and system | |
Singh | Interpretable machine-learning approach in estimating FDI inflow: visualization of ML models with LIME and H2O | |
Gorripaty et al. | Decision support framework to assist air traffic management | |
Mahaweerawat et al. | Adaptive self-organizing map clustering for software fault prediction | |
CN113837220A (en) | Robot target identification method, system and equipment based on online continuous learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160921 |
|
RJ01 | Rejection of invention patent application after publication |