CN110084359A - Dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and system - Google Patents
Dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and system Download PDFInfo
- Publication number
- CN110084359A CN110084359A CN201910213353.2A CN201910213353A CN110084359A CN 110084359 A CN110084359 A CN 110084359A CN 201910213353 A CN201910213353 A CN 201910213353A CN 110084359 A CN110084359 A CN 110084359A
- Authority
- CN
- China
- Prior art keywords
- cognitive map
- intuitionistic fuzzy
- fuzzy cognitive
- dynamic
- intuitionistic
- 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
- 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
- G06N3/043—Architecture, e.g. interconnection topology based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Mathematical Optimization (AREA)
- Molecular Biology (AREA)
- Pure & Applied Mathematics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Mathematical Analysis (AREA)
- Fuzzy Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Mathematics (AREA)
- Mathematical Physics (AREA)
- Automation & Control Theory (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Present disclose provides dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and systems.Wherein, dynamic intuitionistic fuzzy Cognitive Map construction method, comprising: normalized temporal sequence data is clustered into using fuzzy C-means clustering the node of intuitionistic fuzzy Cognitive Map；The weight matrix of training intuitionistic fuzzy Cognitive Map, obtains initial intuition Fuzzy Cognitive Map；It receives new normalized temporal sequence data and is input to initial intuition Fuzzy Cognitive Map；Using the position of dynamic fuzzy C mean cluster adjustment intuitionistic fuzzy Cognitive Map cluster centre, dynamic intuitionistic fuzzy Cognitive Map is obtained；During adjusting the position of intuitionistic fuzzy Cognitive Map cluster centre, if current time data successfully fall into an existing cluster and are greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value, judge whether current time data concept drift occur using drift detection, if so, the weight matrix of adjustment intuitionistic fuzzy Cognitive Map；Otherwise, intuitionistic fuzzy Cognitive Map is not changed.
Description
Technical field
The disclosure belong to time series forecasting field more particularly to a kind of dynamic intuitionistic fuzzy Cognitive Map construction method, when
Between sequence prediction method and system.
Background technique
Only there is provided background technical informations relevant to the disclosure for the statement of this part, it is not necessary to so constitute first skill
Art.
Fuzzy Cognitive Map (FCM) is indicated using graph structure, is made of node and arc, and node can be concept, entity
The causality between concept or entity is indicated Deng, arc.It is essentially different with traditional intelligence computation method, can be with
Easily with matrix knowledge is indicated and reasoning.Since it is the product that fuzzy logic and neural network combine, from structure
On see the neural network that can be considered as the single belt feedback of object-oriented, so the method for neural network can be used for reference herein,
But it has stronger semanteme than neural network again, has stronger interpretation, can more intuitively and easily study complexity
Adaptation system.Currently, it has become the direction of artificial intelligence study.It is readily observed in Cognitive Map in system each
How to interact between concept, each concept and which concept have causality.The most significant feature of Cognitive Map is exactly
There is simple additive property between system, and the dynamic causal system with feedback can be represented.Fuzzy Cognitive Map (FCM) is made
It is applied to every field for a kind of important model, is directed to antidiastole, Aeronautical Service management, urban design, strategy point
The modeling of the challenge in other fields such as analysis, stock analysis, control system.Fuzzy Cognitive Map (FCM) is as a kind of important
Prediction model also has good application in terms of the prediction for time series.Such as TV is received based on Fuzzy Cognitive Map
Prediction etc. depending on the prediction of rate, to room temperature.In the prediction for time series, the information of construction is from time sequence
It is abstracted in column, and constitutes Fuzzy Cognitive graph structure as node (concept).In the process of Information Granulating building model
In, using linguistic labels and fuzzy set, keep model easier to understand.
Inventors have found that the node of Fuzzy Cognitive Map is constructed by clustering, and the acquisition of the weight matrix of model is deposited
In certain difficulty.Weight matrix between the node (concept) of usual Fuzzy Cognitive Map (FCM) often relies on the field special
What the knowledge and experience of family provided.If the data of numeric type this for time series model, because different field
The variation tendency of different data and it is characterized in being difficult artificial capture.
A kind of method of the intuitionistic Fuzzy Sets as processing fuzzy message, expression and place of the intuitionistic Fuzzy Sets theory in ambiguity
The advantage for managing aspect, makes it be taken seriously, and as theoretical constantly improve, intuitionistic Fuzzy Sets obtain in terms of decision and reasoning
It is widely applied.Intuitionistic Fuzzy Sets (IFS) is being introduced to the Fuzzy Cognitive Map of time series forecasting, to propose intuition
Fuzzy Cognitive Map.
Inventor also found, construct intuitionistic fuzzy Cognitive Map by introducing intuitionistic Fuzzy Sets, in the past for medical decision making, pre-
The intuitionistic fuzzy Cognitive Map of survey be constructed according to the experience of expert and known knowledge, once model construction well would not
Change again, but due to the complexity and variability of time series, it is next can not to determine whether the model built meets
Data variation trend.
Summary of the invention
To solve the above-mentioned problems, the first aspect of the disclosure provides a kind of dynamic intuitionistic fuzzy Cognitive Map building side
Method, the influence according to current time data to model are made adjustment, and can capture the variation of data information in real time, from
And keep prediction more accurate.
To achieve the goals above, the disclosure adopts the following technical scheme that
A kind of dynamic intuitionistic fuzzy Cognitive Map construction method characterized by comprising
Normalized temporal sequence data is clustered into using fuzzy C-means clustering the node of intuitionistic fuzzy Cognitive Map；
The weight matrix of training intuitionistic fuzzy Cognitive Map, obtains initial intuition Fuzzy Cognitive Map；
It receives new normalized temporal sequence data and is input to initial intuition Fuzzy Cognitive Map；
Using the position of dynamic fuzzy C mean cluster adjustment intuitionistic fuzzy Cognitive Map cluster centre, dynamic intuition mould is obtained
Paste Cognitive Map；
During adjusting the position of intuitionistic fuzzy Cognitive Map cluster centre, if current time data successfully fall into one
Existing cluster and it is greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value
When value, judge whether current time data concept drift occur using drift detection, if so, adjustment intuitionistic fuzzy Cognitive Map
Weight matrix；Otherwise, intuitionistic fuzzy Cognitive Map is not changed.
To solve the above-mentioned problems, the second aspect of the disclosure provides a kind of dynamic intuitionistic fuzzy Cognitive Map building system
System, the influence according to current time data to model are made adjustment, and can capture the variation of data information in real time, from
And keep prediction more accurate.
To achieve the goals above, the disclosure adopts the following technical scheme that
A kind of dynamic intuitionistic fuzzy Cognitive Map building system, comprising:
Cluster module is used to that normalized temporal sequence data to be clustered into intuitionistic fuzzy using fuzzy C-means clustering and recognizes
Know the node of figure；
Weight matrix training module is used to train the weight matrix of intuitionistic fuzzy Cognitive Map, and it is fuzzy to obtain initial intuition
Cognitive Map；
Sequence data receiving module, is used to receive new normalized temporal sequence data and to be input to initial intuition fuzzy
Cognitive Map；
Module is adjusted, the position using dynamic fuzzy C mean cluster adjustment intuitionistic fuzzy Cognitive Map cluster centre is used for,
Obtain dynamic intuitionistic fuzzy Cognitive Map；
During adjusting the position of intuitionistic fuzzy Cognitive Map cluster centre, if current time data successfully fall into one
Existing cluster and it is greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value
When value, judge whether current time data concept drift occur using drift detection, if so, adjustment intuitionistic fuzzy Cognitive Map
Weight matrix；Otherwise, intuitionistic fuzzy Cognitive Map is not changed.
To solve the above-mentioned problems, a kind of Time Series Forecasting Methods are provided in terms of the third of the disclosure, by dynamic
State intuitionistic fuzzy Cognitive Map realizes the prediction of time series, can establish from historical data, the subjectivity for avoiding expert from providing
It influences, improves the accuracy of prediction.
To achieve the goals above, the disclosure adopts the following technical scheme that
A kind of Time Series Forecasting Methods, comprising:
Time series data to be predicted is input in dynamic intuitionistic fuzzy Cognitive Map；The dynamic intuitionistic fuzzy cognition
Figure is realized using dynamic intuitionistic fuzzy Cognitive Map construction method described above；
The degree of membership that should be predicted by dynamic intuitionistic fuzzy Cognitive Map output phase；
Anti fuzzy method, the truthful data accordingly predicted are carried out to the degree of membership of prediction.
To solve the above-mentioned problems, the 4th aspect of the disclosure provides a kind of time series forecasting system, by dynamic
State intuitionistic fuzzy Cognitive Map realizes the prediction of time series, can establish from historical data, the subjectivity for avoiding expert from providing
It influences, improves the accuracy of prediction.
To achieve the goals above, the disclosure adopts the following technical scheme that
A kind of time series forecasting system, comprising:
Data reception module is used to receive time series data to be predicted and is input to dynamic intuitionistic fuzzy and recognizes
Know in figure；The dynamic intuitionistic fuzzy Cognitive Map is realized using dynamic intuitionistic fuzzy Cognitive Map construction method described above；
Degree of membership prediction module is used for the degree of membership that should be predicted by dynamic intuitionistic fuzzy Cognitive Map output phase；
Anti fuzzy method module is used to carry out anti fuzzy method, the truthful data accordingly predicted to the degree of membership of prediction.
To solve the above-mentioned problems, the 5th aspect of the disclosure provides a kind of computer readable storage medium, basis
Influence of the current time data to model is made adjustment, and can capture in real time the variation of data information, to make prediction more
Precisely.
To achieve the goals above, the disclosure adopts the following technical scheme that
A kind of computer readable storage medium, is stored thereon with computer program, realization when which is executed by processor
Step in dynamic intuitionistic fuzzy Cognitive Map construction method described above.
A kind of computer readable storage medium, is stored thereon with computer program, realization when which is executed by processor
Step in Time Series Forecasting Methods described above.
To solve the above-mentioned problems, the 6th aspect of the disclosure provides a kind of computer equipment, according to current time
Influence of the data to model is made adjustment, and can capture in real time the variation of data information, to keep prediction more accurate.
To achieve the goals above, the disclosure adopts the following technical scheme that
A kind of computer equipment can be run on a memory and on a processor including memory, processor and storage
Computer program, the processor realize dynamic intuitionistic fuzzy Cognitive Map construction method described above when executing described program
In step.
A kind of computer equipment can be run on a memory and on a processor including memory, processor and storage
Computer program, the processor realize the step in Time Series Forecasting Methods described above when executing described program.
The beneficial effect of the disclosure is:
(1) due to the hesitation degree that major advantage is to belong in view of element to set of IFS, and hesitation value is logical
Cross what its existing relationship between degree of membership, non-affiliated degree was calculated.Introduce the hesitation structure to the relationship between node
Initial intuition Fuzzy Cognitive Map is built, the disclosure introduces the building dynamic intuition mould that hesitates when determining concept value on this basis
Cognitive Map is pasted, hesitation effectiveness is traveled to by final decision by reasoning process.
(2) model constructed by the disclosure is one dynamic, and tune is made in the influence according to current time data to model
Whole, model can capture in real time the variation of data information, to keep prediction more accurate.
(3) reasonable weight matrix in order to obtain, makes model be more in line with actual conditions, and the weight study of model is made
With particle swarm optimization algorithm.
(4) when carrying out dynamic and changing the intuitionistic fuzzy Cognitive Map of building, the present disclosure introduces drifts to detect.The disclosure makes
It is performance method, according to prediction error to determine whether producing concept drift and then being changed to model.
(5) disclosure realizes the prediction of time series by dynamic intuitionistic fuzzy Cognitive Map, can be from historical data
It establishes, the subjective impact for avoiding expert from providing；By introducing the influence of hesitation degree, keep model prediction more accurate.It considers
The diversity of time series carries out dynamic adjustment to model structure and weight using dynamic fuzzy C mean cluster, makes model can
With the adaptive variation tendency for meeting time series, drift detection is introduced when for updating weight, when floating
Mean that changing occurs in the variation tendency of data when shifting, make adjustment to model this when, improves the prediction effect of model
Rate and accuracy.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is the Fuzzy Cognitive Map that the embodiment of the present disclosure provides.
Fig. 2 is the dynamic intuitionistic fuzzy Cognitive Map construction method flow chart that the embodiment of the present disclosure provides.
Fig. 3 is the initial intuition Fuzzy Cognitive Map that the embodiment of the present disclosure provides.
Fig. 4 be the embodiment of the present disclosure provide there are four node dynamic intuitionistic fuzzy Cognitive Map.
Fig. 5 is the position adjustment for the cluster centre that the embodiment of the present disclosure provides.
Fig. 6 (a) is 1 scalar time sequence of experiment that the embodiment of the present disclosure provides.
Fig. 6 (b) is the amplitude variation for 1 time series of experiment that the embodiment of the present disclosure provides.
Fig. 6 (c) is that 1 pair of time series of experiment that the embodiment of the present disclosure provides is handled, and becomes it in amplitude and amplitude
It shows in the two-dimensional space of change.
Fig. 6 (d) is the clustering prototype that the experiment 1 that the embodiment of the present disclosure provides is obtained by fuzzy C-means clustering.
Fig. 7 (a) is variation of 1 fuzzy set of experiment (performance indicator) Q that provides of the embodiment of the present disclosure with cluster.
Fig. 7 (b) is the error amount in the experiment 1 that the embodiment of the present disclosure provides.
Fig. 8 (a) is the static fuzzy Cognitive Map for the experiment 1 that the embodiment of the present disclosure provides.
The data at current time are predicted in the experiment 1 that Fig. 8 (b) provides for the embodiment of the present disclosure.
Fig. 8 (c) is the drift detection for the experiment 1 that the embodiment of the present disclosure provides.
The comparison diagram of three kinds of methods in the experiment 1 that Fig. 8 (d) provides for the embodiment of the present disclosure.
Fig. 9 is 1 predicted value of experiment that provides of the embodiment of the present disclosure compared with true value.
2 scalar time sequence of experiment that Figure 10 (a) embodiment of the present disclosure provides.
Figure 10 (b) is the amplitude variation for 2 time series of experiment that the embodiment of the present disclosure provides.
Figure 10 (c) is that 2 pairs of time serieses of experiment that the embodiment of the present disclosure provides are handled, and becomes it in amplitude and amplitude
It shows in the two-dimensional space of change.
Figure 10 (d) is the clustering prototype that the experiment 2 that the embodiment of the present disclosure provides is obtained by fuzzy C-means clustering.
Figure 11 (a) is the phenomenon that experiment 2 that the embodiment of the present disclosure provides is drifted about, and changes the weight of model.
Figure 11 (b) is that the experiment 2 that the embodiment of the present disclosure provides introduces hesitation degree.
Figure 12 (a) is the static fuzzy Cognitive Map for the experiment 2 that the embodiment of the present disclosure provides.
The data at current time are predicted in the experiment 2 that Figure 12 (b) provides for the embodiment of the present disclosure.
Figure 12 (c) is the drift detection for the experiment 2 that the embodiment of the present disclosure provides.
The comparison diagram of three kinds of methods in the experiment 2 that Figure 12 (d) provides for the embodiment of the present disclosure.
Figure 13 is 2 predicted value of experiment that provides of the embodiment of the present disclosure compared with true value.
Figure 14 (a) is 3 scalar time sequence of experiment that the embodiment of the present disclosure provides.
Figure 14 (b) is the amplitude variation for 3 time series of experiment that the embodiment of the present disclosure provides.
Figure 14 (c) is that 3 pairs of time serieses of experiment that the embodiment of the present disclosure provides are handled, and becomes it in amplitude and amplitude
It shows in the two-dimensional space of change.
Figure 14 (d) is the clustering prototype that the experiment 3 that the embodiment of the present disclosure provides is obtained by fuzzy C-means clustering.
Figure 15 (a) is the 3 static fuzzy Cognitive Map of experiment that the embodiment of the present disclosure provides.
The data at current time are predicted in the experiment 3 that Figure 15 (b) provides for the embodiment of the present disclosure.
Figure 15 (c) is the drift detection for the experiment 3 that the embodiment of the present disclosure provides.
The comparison diagram of three kinds of methods in the experiment 3 that Figure 15 (d) provides for the embodiment of the present disclosure.
Figure 16 is 3 predicted value of experiment that provides of the embodiment of the present disclosure compared with true value.
4 scalar time sequence of experiment that Figure 17 (a) embodiment of the present disclosure provides.
Figure 17 (b) is the amplitude variation for 4 time series of experiment that the embodiment of the present disclosure provides.
Figure 17 (c) is that 4 pairs of time serieses of experiment that the embodiment of the present disclosure provides are handled, and becomes it in amplitude and amplitude
It shows in the two-dimensional space of change.
Figure 17 (d) is the clustering prototype that the experiment 4 that the embodiment of the present disclosure provides is obtained by fuzzy C-means clustering.
Figure 18 (a) is the error amount for the experiment 4 that the embodiment of the present disclosure provides.
Figure 18 (b) is that the experiment 4 that the embodiment of the present disclosure provides introduces drift detection.
Figure 18 (c) is the comparison diagram that dynamic adjustment and drift detection is added in the experiment 4 that the embodiment of the present disclosure provides.
Figure 19 is the dynamic intuitionistic fuzzy Cognitive Map for the milk that the experiment 4 that the embodiment of the present disclosure provides constructs.
Figure 20 is 4 predicted value of experiment that provides of the embodiment of the present disclosure compared with true value.
Figure 21 is the Time Series Forecasting Methods flow chart that the embodiment of the present disclosure provides.
Specific embodiment
The invention will be further described with embodiment with reference to the accompanying drawing.
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has logical with the application person of an ordinary skill in the technical field
The identical meanings understood.
It should be noted that term used herein above and is not intended to restricted root evidence only for describing specific embodiment
The illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
Fuzzy Cognitive Map:
The domain E of a given discussion, fuzzy set are defined as:
A={ < x, μA(x)>|x∈E} (1)
Here μA: E → [0,1] indicates that the element of each x ∈ E belongs to setDegree of membership.For each element x
∈ E, there is 0≤μA(x)≤1。
Fuzzy Cognitive Map is that Fuzzy inferential engine is added in Cognitive Map (Cognitive Map) to get, and is that one kind has
Xiang Tu.As shown in Figure 1.Fig. 1 describes the FCM for having 5 nodes and 6 cum rights arcs.Its interior joint A1, A2..., Ac
It is concept, it can be event, target, emotion and the trend etc. of system, for reflecting attribute, feature, quality and the shape of system
State has certain state value, and state value is the fuzzy value on [- 1,1] section, indicates degree existing for concept status.Concept
Between causality by ω=[ωij], i, j=1,2 ..., c, ωij∈ [- 1,1] value indicates (i.e. weight), and there are three types of classes
The causality of type: work as wijWhen > 0, indicating to increase with the state value of j-th of node, the state value of the i-th node also increases,
Work as wijIt when < 0, indicates to increase with the state value of j-th of node, the state value of i-th of node will be reduced, and work as wijWhen=0,
It indicates between jth node and i-th of node without any incidence relation.
Activation level of the Fuzzy Cognitive Map at the current K moment is described as A (k)=[A1(k), A2(k) ..., Ac(k)], this
In activation level meaning be exactly the current K moment time series be mapped to c cluster degree of membership, obtained down by it
The activation level at one moment is A (k+1)=[A1(k+1), A2(k+1) .., Ac(k+1)] it, is described as follows:
F is the sigmoid function with steepness parameter σ under normal conditions, is expressed as follows:
Intuitionistic Fuzzy Sets:
One intuitionistic Fuzzy Sets can be regarded as the fuzzy set of a broad sense, give a domain E, and intuitionistic Fuzzy Sets are determined
Justice are as follows:
A={ < x, μA(x), γA(x)>|x∈E} (4)
Here μA: E → [0,1], γA: E → [0,1] respectively indicates element x ∈ E and belongs to setDegree of membership and
Non-affiliated degree, 0≤μA(x)≤1,0≤γA(x)≤1, herein:
0≤μA(x)+γA(x)≤1 (5)
For each x ∈ E, if γA(x)=1- μA(x), A means that a fuzzy set.Member is concentrated in intuitionistic fuzzy
Plain x ∈ E belongs to setHesitation degree be expressed as follows:
πA(x)=1- μA(x)-γA(x) (6)
Hesitation can be considered as the uncertainty degree that x is under the jurisdiction of in A, and it usually has more intuitively than non-affiliated degree
It explains.For intuitionistic Fuzzy Sets calculating there are many kinds of definition, among these include the summation to two fuzzy sets and multiplication,
They are defined as:
Model:
Model is divided into four parts, first with fuzzy C-means clustering then data clusters is utilized at model node
PSO carries out parameter learning, constitutes initial intuitionistic fuzzy Cognitive Map by this two step and utilizes dynamic analog when new data comes
Paste four kinds of situations of C mean cluster point are adjusted model, and drift detection is introduced when adjusting parameter, concept drift occurs
When, illustrate that the parameter of model at this time is not suitable for the variation tendency of current data, so parameter update is carried out to model, here it is
The dynamic intuitionistic fuzzy Cognitive Map building process that the present embodiment proposes is as shown in Figure 2.
Building dynamic intuitionistic fuzzy Cognitive Map first has to build initial intuitionistic fuzzy Cognitive Map using historical data, and
It is updated on this basis.For the initial intuition Fuzzy Cognitive Map (IFCM-I) of medical decision making, it is used for symptom and medicine
Decision-making diagnosis connects.It is not only merely according to their illnesss when expert makes a policy according to the symptom of patient
Analysis, also includes their intuition, so there is the influence of hesitation degree here, its hesitation degree is exactly for closing between symptom
Then the hesitation of system makes a policy according to the influence generated to hesitation degree between symptom, thus reach one it is more pre-
It surveys.
For the initial intuition Fuzzy Cognitive graph model to time series forecasting, the place different from decision model is to save
Hesitation degree between point does not artificially provide, and is one kind between weight and true weight required in learning process
Error, for reducing the error between predicted value and true value.Fig. 3, which is a tool, to be recognized there are four the initial intuition of node is fuzzy
Know figure.Herein according to the hesitation degree bring negative influence between concept, the inference form of initial intuition Fuzzy Cognitive Map is obtained
Are as follows:
HereinIndicate the weighted value between concept i and concept j,Indicate concept i and concept j it
Between weight hesitation angle value, but there is also certain limitations, and in order to overcome these limitations, we have done some variations:
What initial intuition Fuzzy Cognitive Map only considered is the hesitation degree to weight relationship between node, what the present embodiment proposed
New dynamic intuitionistic fuzzy Cognitive Map further includes the hesitation that is subordinate to of the time series to concept at current time.It is poly- by Fuzzy C
The degree of membership for only belonging to C cluster that class obtains, not can determine which cluster it particularly belongs to, and it is straight that here it is dynamics
Feel Fuzzy Cognitive Map.
The weight matrix and hesitation degree of the time series of dynamic intuitionistic fuzzy Cognitive Map are also to be obtained by study, currently
The hesitation that is subordinate to of the time series at moment to concept, as in medical decision making the expiratory dyspnea of patient be it is appropriate,
Hesitation to patient respiration difficulty be it is low, be mapped in time series, one can consider that degree of membership it is high hesitation degree it is low.
For the ease of the readability of equation, { < v is usedμ,vγ>}iSimplification representation indicate the input at current time, vμRepresentative is worked as
The degree of membership at preceding moment, vγThe non-affiliated degree at current time is represented, degree of membership, non-affiliated degree and hesitation degree abide by formula (6).
Hesitation degree is asked according to current degree of membership, is mapped to first by the time series that fuzzy C-means clustering acquires current time
The degree of membership of c cluster, then take out c cluster in two degrees of membership it is maximum, when current to their corresponding cluster representatives
The time series at quarter has hesitation, and others think that hesitation degree is 0.Think hesitation degree and degree of membership there are certain interconnecting relation,
If degree of membership is big, it is believed that the true value of the data at current time closer to this cluster, for this result hesitation degree just
Relative to a little bit smaller, if degree of membership is small, it is believed that the probability that the true value of the data at current time is clustered close to this is relatively
It is small, for this result hesitation degree with respect to more greatly.It is proposed the formula of a calculating hesitation degree:
vπ=f (vμ)+random uniform (0,0.2) v μ (11)
Allow degree of membership by Fermi's formula f activation, satisfaction is subordinate to the small principle of big obtained hesitation degree, then gives
The interference of a fractional value out, this disturbance range can be arranged according to specific circumstances, this interference and the value of degree of membership have
It closes.Fermi's formula are as follows:
F (x)=1/ (1+math.exp (x/k)) (12)
With { < ωμ,ωγ> indicate influence of the node j to node i.Indicate that dynamic intuitionistic fuzzy recognizes with following equation
The structure of figure:
Here count summation and multiplication does some variations with formula (7) and (8).It utilizes (8) and (13) to calculate, obtains:
And the right item of the equation is calculated by using (7) by symbolThe intuitionistic fuzzy of description is summed, it becomes:
micIt is calculated by (16):
In this equationIt indicates summation, can be replaced with (7).
Equation (16) can be obtained by recursive calculation:
It is available according to formula (17):
The IFCM-II model there are four node is enumerated, as shown in Figure 4.M is calculated with (19)ijIt is divided into four steps:
By using m in (15)i4C=4 is substituted into obtain the reasoning equation of the model.So calculated from (15) i-th
The degree of membership of node and the value of non-affiliated degree, iteration K+1 times formation { < vμ,vγ>}iForm:
Obviously the complexity of (15) is higher than the complexity of (2) and (10), and result obtained will provide more information, because
It can quantify the hesitation degree propagated by (15).
The parameter learning of model:
Building intuitionistic fuzzy Cognitive Map most importantly finds the node and weight matrix of model.Time series can be with very
It is multi-form to indicate, the amplitude variation of amplitude space is only focused on herein, and wherein time series is retouched with the amplitude at continuous moment
It states, and the variation of amplitude can also be depicted in two dimensional amplitude space.Time series xK, k=1,2 ..., NIt indicates,
Here a processing is made to data and is converted to two-dimensional space zk=[xk,Δxk] form, it can intuitively be mapped to model
Node (concept) on.
What building graph structure utilized is fuzzy C-means clustering.Fuzzy C-mean algorithm forms c by minimizing some objective function
A information.Fuzzy set (performance indicator) Q indicates the Weighted distance between data and the representative (cluster centre) of data
The sum of.
And subordinated-degree matrix U=[uik] and a series of cluster centre v1,v2,...,vc, it is that performance is minimized by iteration
Index Q is obtained.The node (concept) of each cluster centre and model is associated, the activation water of c node of data to model
Flat (degree of membership) is expressed as follows:
Herein | | | | it is a normal form, there are two basic parameters, i.e. cluster number (c) and fuzzy coefficient for fuzzy clustering
(m).The quantity wherein clustered is exactly the quantity of the node (concept) of model, and back experimental section is described, and is obscured
Coefficient is usually arranged as 2.0.
After model is built up, each data correspond to an activation vector for model, each to activate the c of vector and model
Node (concept) is associated, that is, when data clusters at c node, each data have one to be subordinate to c node
Degree, then we find out hesitation degree and non-affiliated degree according to degree of membership.
Only compare degree of membership herein, target is expressed as targetk, k=1,2 ..., N, target is understood to be data
The degree of membership of c node is actually corresponded to, the activation level that we are obtained by model prediction should be as close possible to this mesh
Scale value, that is,For the prediction model of time series, using current defeated
Enter data namely to give a forecast to the activation vector of subsequent time with the activation vector at the i-th moment, the activation vector of prediction more connects
The performance of close-target value model is better.
About the optimization of parameter, weight matrix is only adjusted, steepness factor here is set as 5.This stage directly affects mould
The estimated performance of type.It realizes parameter optimization, selects a good optimization algorithm, such as Gradient learning, genetic optimization, particle first
Group's optimization, what we selected here is particle swarm optimization algorithm.
Assuming that having the molecular group of N number of grain, wherein i-th of particle is expressed as in the target search space of D dimension
The vector of one D dimension:
Xi=(xi1,xi2,...,xiD), i=1,2 ..., N
The flying speed of i-th of particle is also the vector of D dimension, is denoted as:
Vi=(vi1,vi2,...,viD), i=1,2 ..., N
The optimal location that i-th of particle searches so far is individual extreme value, is denoted as:
Pbest=(pi1,pi2,...,piD), i=1,2 ..., N
The optimal location that entire population searches so far is global extremum, is denoted as:
gbest=(g1,g2,...,gD)
When finding the two optimal values, particle updates speed and the position of oneself according to (26) and (27):
xij(t+1)=xij(t)+vij(t+1) (27)
The dynamic inertia weight value of PSO herein uses linear decrease weight strategy, and inertia weight ω is used to adjust grain
Influence of the sub- current iteration speed to next iteration speed has very big shadow to overall situation and partial situation's search capability of algorithm
It rings.Since big inertia weight is conducive to global search, small inertia weight is conducive to local search, and this paper ω is with algorithm
The progress of iteration and linear reduction improve convergence for taking into account the balance of global search and local search.TmaxFor
Maximum number of iterations, t are current iteration number, ωmaxFor maximum weighted coefficient, ωminFor minimum weight coefficient, general ωmax
=0.9, ωmin=0.4, expression formula are as follows:
When with the weight matrix of PSO algorithm training pattern, the weight section of weight matrix is provided first, to weight matrix
It is initialized, the element for then defining population position vector is the weight matrix of model, is iterated optimizing.Herein
Stopping criterion for iteration is set as meeting maximum number of iterations.For Optimization Solution model weight matrix (including hesitation degree square
Battle array), fitness function is chosen are as follows:
Wherein N is number of training, A in formulaiFor the activation vector of model reality output,For desired activation
Vector, it follows that error amount eiSmaller, model performance is better.
Dynamic fuzzy C mean cluster:
Initial intuitionistic fuzzy Cognitive Map is obtained using fuzzy C-means clustering and PSO, then uses dynamic fuzzy C mean value
Cluster adjusts according to position of the current time data to node.Wherein each cluster is indicated by its cluster centre, and is all distributed
The data that one radix is used to record each moment distribute to the degree of membership of the cluster.
uck(i) be the i moment k-th cluster radix, ukiIt is the degree of membership that i time data belongs to k-th cluster, meter
Calculation method is the formula (5) of fuzzy C-means clustering.When radix is increasing with the addition of data, then the new data come in
Influence for cluster centre can be relatively small, cluster centre will not easily because of a data sudden change newly inputted, if
It just will be updated the position of cluster centre when influencing excessive, update method is as follows:
When there is new data xiWhen coming in, progress data processing first obtains zi, then calculate the cluster centre of it and model
viDistance Di(j), when there is the fact that following, it just will increase a new cluster:
Dmin=argmin (Di(j)) > kd
kdMaximum distance of the initial value between initial cluster center, with the change of cluster centre during later
Change, kdValue also will do it adjustment.
The method for calculating error are as follows:
In the activation level (degree of membership) at K moment, it is described as A (k)=[A1(k),A2(k),...,Ac(k)], passed through by it
Calculate and obtain by the activation of sigmoid function the activation level (degree of membership) of subsequent time.B (k+1)=[B1(k+1),
B2(k+1),...,Bc(k+1)] what is represented is subsequent time to the true degree of membership of c cluster centre.In eiUsed in return
One change is carried out in the quantity (c) of node (concept) and data volume (N-1), so that characteristic of its value independently of map
With the quantity of modeling process, this permission can be compared analysis to result under different condition for study.
It is to be made comparisons according to the predicted value of model with true target value herein, and then the performance of judgment models, and such as
Fruit eiValue be more than model performance that we need, then we, which can consider, is added new cluster centre, or changes one
The position of lower cluster centre is also possible to adjust lower weight, reduces error amount as far as possible.
The present embodiment is compared according to the error of the true value and predicted value that find out with the performance ke of model is needed, and dynamic is adjusted
Integral mould is roughly divided into following four situation:
The first situation:
||ei| | > ke, Dmin≥kdSuch case is considered as that a new cluster centre, the cluster centre being newly added is added
For the data at current time.
Second situation:
||ei||≤ke,Dmin< kdSuch case is a model satisfaction institute model performance in need at this stage, must not
Adjust model.
The third situation:
||ei||≤ke,Dmin≥kdSuch case shows that model cannot limit the new data just inputted well, but mould
Type can receive the data newly come in, therefore can use the cluster centre of dynamic fuzzy C mean value more new model, adjust in lower cluster
The position of the heart.
4th kind of situation:
||ei| | > ke, Dmin< kdSuch case is that the data of rigid input model successfully fall into an existing collection
Group, however the error of model is unsatisfactory for our requirement, judges whether concept drift occurred using drift detection this when
It moves, if occurring adjusting the weight matrix of model, otherwise without modification.Detailed process is as shown in Figure 5.
Drift detection:
In real life, a kind of problem is that the concept that data are included may change over time.Automatic metaplasia
In producing line, product can continuously occur the problem of close reason, and then the feature of defective product also changes therewith, and commercial affairs are living
In dynamic, the purchase interest time to time change of customer, in network security, the access module of network is different with user and changes this
The common feature of a little problems is: constantly generating new data, the concept that data include is at any time there may be variation, this in data
The variation of the conception of species is in referred to as concept drift.Data are divided into 2 seed types: one is same points of Dynamic data exchange that data source generates
Cloth, referred to as stable data；Another kind is the data not independent same distribution that data source generates, it is believed that is sent out in data generating procedure
It has given birth to " concept drift ".Since movement is the essence of substance, concept drift is also the essence of data.Detection concept is drifted about substantially
There are 3 kinds of methods: performance method, Furthest Neighbor and property method.
When carrying out dynamic adjustment to model, drift detection joined, the thought for detection of drifting about is very simple, is exactly
Control the online error-rate of the algorithm (the online error rate of control algolithm).If sample
Data are Stable distritations, then the error rate of model will be gradually reduced with the input of data.When probability distribution occurs
When variation, the error rate of model will rise.So concept drift is exactly the error rate in online control model training process.
Two threshold values can be arranged for error rate in drift detection, and one is warning, the other is drift.Work as sample data
In the input of w-th of data when, error rate has reached warning value, illustrate to have the omen of sample probability distribution change, if
The data inputted in succession do not allow lower error rate, and error rate has reached drift value when d-th of data inputs, then really
This probability distribution of random sample is changed, and in order to adapt to new sample data, model will just be learnt with the data after w,
And if the data inputted in succession allow lower error rate, illustrate to be a false alarm, model will not change.
It about the size of the two threshold values of warning and drift, is determined by the probability distribution of error rate.It is false
If the sample sequence of input is such (xi,yi), xiIt is the degree of membership at a current time, yiIt is the true person in servitude of subsequent time
Category degree, model is to xiPredicted value beBy comparingAnd yiThe prediction result for being assured that i-th of sample is False
Or True, so finally can be obtained by such a Bernoulli distribution.I-th point of error rate is just
The probability observation for being the False of this point is pi, while the standard deviation of this point is also available: si=sqrt (pi*(1-
pi)/i), warning and drift level is determined according to confidence level (confidence level).Warning is set as
95% confidence level, i.e. pi+si>==min (p)+2*min (s), drift are set as 99% confidence level, i.e. pi+si>==
min(p)+3*min(s)。
By being compared to obtain p of the wrong numerical value number for calculating current time with threshold valuei, because in reasoning
There is a hesitation when which is clustered to the data membership at calculating current time in journey, so current time is calculated
Number of errors also have a hesitation.It is added up by their maximum two hesitation degree all incorrect data,
Then an average value is sought, for calculating piAnd si。
di=s/ (2*o) (33)
diRepresent to occurring the hesitation of number of errors summation at the i moment, behalf wrong moment hesitation degree summation, o
It represents and the summation of number of errors occurs at the i moment.Pass through diWe obtain the p at current timeiAnd si。
pi=(o*di)/i (34)
si=sqrt (pi*(1-pi)/i) (35)
In another embodiment, a kind of and dynamic intuitionistic fuzzy Cognitive Map construction method institute as shown in Figure 2 is additionally provided
Corresponding dynamic intuitionistic fuzzy Cognitive Map constructs system, comprising:
(1) cluster module is used to that normalized temporal sequence data to be clustered into intuitionistic fuzzy using fuzzy C-means clustering
The node of Cognitive Map；
(2) weight matrix training module is used to train the weight matrix of intuitionistic fuzzy Cognitive Map, obtains initial intuition mould
Paste Cognitive Map；
Specifically, in the weight matrix training module, particle swarm optimization algorithm training intuitionistic fuzzy Cognitive Map is utilized
Weight matrix, process are as follows:
The weight section for providing weight matrix first, initializes weight matrix, then define population position to
The element of amount is the weight matrix of intuitionistic fuzzy Cognitive Map, is iterated optimizing, until obtaining optimal weights matrix or meeting to change
For termination condition.
(3) sequence data receiving module is used to receive new normalized temporal sequence data and is input to initial intuition
Fuzzy Cognitive Map；
(4) module is adjusted, the position using dynamic fuzzy C mean cluster adjustment intuitionistic fuzzy Cognitive Map cluster centre is used for
It sets, obtains dynamic intuitionistic fuzzy Cognitive Map；
During adjusting the position of intuitionistic fuzzy Cognitive Map cluster centre, if current time data successfully fall into one
Existing cluster and it is greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value
When value, judge whether current time data concept drift occur using drift detection, if so, adjustment intuitionistic fuzzy Cognitive Map
Weight matrix；Otherwise, intuitionistic fuzzy Cognitive Map is not changed.
In an alternative embodiment, in the adjustment module, if current time data do not fall within any one existing cluster
And when being greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value, add
Enter a new cluster centre, the cluster centre being newly added is current time data.
In an alternative embodiment, in the adjustment module, if current time data successfully fall into an existing cluster and
It is less than or equal to default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value
When, do not need adjustment intuitionistic fuzzy Cognitive Map.
In an alternative embodiment, in the adjustment module, if current time data successfully fall into an existing cluster and
When being greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value, utilize
Dynamic fuzzy C mean value updates the position of the cluster centre of intuitionistic fuzzy Cognitive Map.
As shown in figure 21, the present embodiment additionally provides a kind of Time Series Forecasting Methods comprising:
Time series data to be predicted is input in dynamic intuitionistic fuzzy Cognitive Map；The dynamic intuitionistic fuzzy cognition
Figure is realized using dynamic intuitionistic fuzzy Cognitive Map construction method as shown in Figure 2；
The degree of membership that should be predicted by dynamic intuitionistic fuzzy Cognitive Map output phase；
Anti fuzzy method, the truthful data accordingly predicted are carried out to the degree of membership of prediction.
A kind of time is additionally provided in one or more embodiments with Time Series Forecasting Methods corresponding to Figure 21
Sequence prediction system, comprising:
Data reception module is used to receive time series data to be predicted and is input to dynamic intuitionistic fuzzy and recognizes
Know in figure；The dynamic intuitionistic fuzzy Cognitive Map is realized using dynamic intuitionistic fuzzy Cognitive Map construction method as shown in Figure 2；
Degree of membership prediction module is used for the degree of membership that should be predicted by dynamic intuitionistic fuzzy Cognitive Map output phase；
Anti fuzzy method module is used to carry out anti fuzzy method, the truthful data accordingly predicted to the degree of membership of prediction.
In one or more embodiments, a kind of computer readable storage medium is additionally provided, computer is stored thereon with
Program, the program realize the step in dynamic intuitionistic fuzzy Cognitive Map construction method as shown in Figure 2 when being executed by processor.
In one or more embodiments, a kind of computer readable storage medium is additionally provided, computer is stored thereon with
Program, the program realize the step in Time Series Forecasting Methods as shown in figure 21 when being executed by processor.
In one or more embodiments, it additionally provides a kind of computer equipment, including memory, processor and is stored in
On memory and the computer program that can run on a processor, the processor are realized as shown in Figure 2 when executing described program
Dynamic intuitionistic fuzzy Cognitive Map construction method in step.
In one or more embodiments, it additionally provides a kind of computer equipment, including memory, processor and is stored in
On memory and the computer program that can run on a processor, the processor are realized when executing described program such as Figure 21 institute
The step in Time Series Forecasting Methods shown.
Experiment:
Experiment 1 is by taking Copper as an example:
Copper is one by the time series tightly investigated, and is the time sequence from publicly available repository
Column, record is the Celsius temperature of copper mine monthly, it at any time fluctuation generally it is constant.Fig. 6 (a) can clearly be seen
This scalar time sequence is examined, Fig. 6 (b) is the amplitude variation of time series, and Fig. 6 (c) is handled time series, is made
It shows in the two-dimensional space that amplitude changes with amplitude, and Fig. 6 (d) is the cluster original obtained by fuzzy C-means clustering
Type.
First the 2-D data handled well of a part is clustered with Fuzzy C-Means Cluster Algorithm, the number of cluster is c=2
~10.It is only concerned the case where fuzzy coefficient is 2.0 in the present embodiment, can be seen that (performance refers to fuzzy set according to Fig. 7 (a)
Mark) Q constantly reduces with the increase of cluster, and amplitude of variation is intended to gently at 7.Fig. 7 (b) is cluster centre number by c
=2~10 obtained minimum error values, it is our of steepness factor here, as can be seen from the figure minimum the case where being only viewed as 5
Error amount is as fuzzy set (performance indicator) Q, and after reaching certain value, fall is less obvious, in conjunction with two
A figure, the cluster number for selecting the model is 7.
The cluster centre finally obtained after cluster are as follows:
v1=[7.34, -0.90], v2=[- 7.24,12.42], v3=[- 2.07, -7.15], v4=[6.20,8.35],
v5=[- 23.59,3.9], v6=[- 28.91, -4.34], v7=[- 19.57, -13.31]
Weight matrix is learnt using PSO algorithm to obtain initial model, then according to the data pair at current time
The data of subsequent time are predicted, the error amount predicted this when be greater than preset threshold value we just to model into
Row adjustment, however the distribution of model may be there is no changing, the adjustment of this when may be to be not necessarily to, and instead may
It will cause bigger error, so first to observe before being changed to model, introduce drift detection.Fig. 8 (a)
It is that model is built up just without the static fuzzy Cognitive Map having changed, Fig. 8 (b) is carried out according to the data at current time
Prediction is greater than threshold value as long as error amount and changes what model obtained as long as us, and Fig. 8 (c) has carried out drift detection and obtained, and hangs down
Straight dotted line representative is then carried out according to from the data for starting to occur alerting to generation when drifting about at the time of generating drift
Study, changes the weight matrix of model.Fig. 8 (d) is the comparison diagram of three kinds of methods.Therefrom it may be seen that drift inspection is added
The modelling effect of survey is more preferable, and generates after drift to the adjustment of model so that model error significantly reduces.
The true data predicted of anti fuzzy method are finally carried out according to degree of membership after prediction, as shown in figure 9, therefrom I
It can be observed that predicted value and true value tendency are coincide substantially, but also have that there is some gaps.In amplitude peak and most
Prediction at low value is simultaneously unsatisfactory, this just has a very big relationship with cluster centre position, but the fluctuation between peak value
Effect or good.
Experiment 2 is by taking Sunspot data as an example:
This data set comes from, and Figure 10 (a) can clearly observe this scalar time sequence, and Figure 10 (b) is time sequence
The amplitude of variation is arranged, Figure 10 (c) is handled time series, its table in amplitude and the two-dimensional space of amplitude variation is made
It shows and, Figure 10 (d) is the clustering prototype obtained by fuzzy C-means clustering.
Figure 11 (a) is to acquire p according to error numberiAnd si, when meeting pi+si>==min (p)+2*min (s) issues police
It accuses, when reaching pi+siWhen >==min (p)+3*min (s), the phenomenon that drift, change the weight of model.This when occurs
The number of drift is a little more, and effect and bad, has obtained Figure 11 (b) in being the introduction of hesitation degree, drift number obviously subtracts
It is few, it is possible that the phenomenon that will appear rising, show updated weight effect and bad, so introducing a judgement
Mechanism updates weight when drifting about, obtained weight for solving next five data, obtained error amount I
Make comparisons with the error obtained with original weight, if effect is relatively good, with updated weight, if bad, just not
Weight is updated, the Weight prediction after capable of thus making drift is more acurrate.
Figure 12 (a) is static fuzzy Cognitive Map, and Figure 12 (b) is predicted according to the data at current time, as long as error
Value is greater than threshold value, and we just change what model obtained, and Figure 12 (c) has carried out drift detection and obtained, and vertical dotted line represents
Be generate drift at the time of.Figure 12 (d) is the comparison diagram of three kinds of methods.Therefrom it may be seen that drift detection is added
Modelling effect is much better.
The true data that anti fuzzy method is predicted finally are carried out according to degree of membership after prediction, as shown in figure 13, therefrom
It can be observed that prediction effect is all well and good.
Experiment 3 is by taking Oldman Time Series as an example:
Third time series Oldman Time Series, Order is graceful, comes from repository, describes from January 1 in 1988
The mean daily flow for the Order Man He that day on December 31st, 1991 is reported.When Figure 14 (a) can clearly observe this scalar
Between sequence, Figure 14 (b) be time series amplitude variation, Figure 14 (c) is handled time series, make its amplitude with
It shows in the two-dimensional space of amplitude variation, Figure 14 (d) is the clustering prototype obtained by fuzzy C-means clustering.
Figure 15 (a) is that model is built up just without the static fuzzy Cognitive Map having changed, and Figure 15 (b) is according to current
The data at moment are predicted, are greater than threshold value as long as us as long as error amount and are changed what model obtained.Figure 15 (c) is to be floated
It moves detection to obtain, what vertical dotted line represented is at the time of generating drift, and then we are according to from starting to occur alerting to production
Data when raw drift are learnt, and the weight matrix of model is changed.Figure 15 (d) is the comparison diagram of three kinds of methods.It therefrom can be with
See that the modelling effect that drift detection is added is more preferable, and generates after drift to the adjustment of model so that model error obviously drops
It is low.
The true data that anti fuzzy method is predicted finally are carried out according to degree of membership after prediction, as can be seen from Figure 16
Prediction effect is pretty good.
Experiment 4 is by taking monthly-milk-production as an example:
4th time series monthly-milk-production, describes the yield of monthly milk, Figure 17 (a) can
Clearly to observe this scalar time sequence, Figure 17 (b) is the amplitude variation of time series, and Figure 17 (c) is to time series
It handles, it is made to show in the two-dimensional space that amplitude changes with amplitude, Figure 17 (d) is to pass through fuzzy C-means clustering
Obtained clustering prototype.
As long as it is more than that our threshold values of defined are changed model that Figure 18 (a), which is once there is error amount,.
Figure 18 (b) introduces drift detection, when occur error amount it is big when, first do not change model, but observed, reached
Model profile can be reminded to may have occurred variation when warning, the more new model when reaching drift, but the model is with number
According to input, error rate is gradually reduced, illustrate that the sample data is Stable distritation, then model is just not necessarily to change,
Weight before meets the needs of model.As shown in Figure 18 (c), is obtained by comparison, joined the model prediction of drift mechanism
Effect is more preferable.
Figure 19 is the dynamic intuitionistic fuzzy Cognitive Map of the milk of building, including four nodes, uses Descartes's language respectively
Speech is described.If data are that a negative less than normal is indicated with (NS), others be successively negative in (NM), bear big (NH),
Just small (PS) is hit exactly (PM), and honest (PH), the amplitude that (PS × NH) can be understood as data is integrally positive number less than normal, it
Amplitude variation is to reduce, and the amplitude of reduction is bigger.By the qualitative description to model, visualization is reached
Purpose.
The true data for finally obtaining prediction, we can see that prediction effect is pretty good from Figure 20.
As shown in table 1, we are the dynamic fuzzy Cognitive Map (DFCMS-II-DDM) for having drift to detect of proposition and previous
Static fuzzy Cognitive Map (FCMS) and static fuzzy intuition Cognitive Map (FCMS-II) have or not and the dynamic of drift detection be added
State Fuzzy Cognitive Map (DFCMS-II) is compared, and effect is obviously relatively good.
The model performance comparison diagram of 1 four kinds of distinct methods of table building
It should be understood by those skilled in the art that, embodiment of the disclosure can provide as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the disclosure
Formula.Moreover, the disclosure, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The disclosure be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that can be realized by computer program instructions each in flowchart and/or the block diagram
The combination of process and/or box in process and/or box and flowchart and/or the block diagram.It can provide these computers
Processor of the program instruction to general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices
To generate a machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute
For realizing the function of being specified in one or more flows of the flowchart and/or one or more blocks of the block diagram
Device.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that instruction stored in the computer readable memory generation includes
The manufacture of command device, the command device are realized in one box of one or more flows of the flowchart and/or block diagram
Or the function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer
Or the instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or box
The step of function of being specified in figure one box or multiple boxes.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage and be situated between
In matter, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be
Magnetic disk, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random
AccessMemory, RAM) etc..
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art
The various modifications or changes that can be made are not needed to make the creative labor still within protection scope of the present invention.
Claims (10)
1. a kind of dynamic intuitionistic fuzzy Cognitive Map construction method characterized by comprising
Normalized temporal sequence data is clustered into using fuzzy C-means clustering the node of intuitionistic fuzzy Cognitive Map；
The weight matrix of training intuitionistic fuzzy Cognitive Map, obtains initial intuition Fuzzy Cognitive Map；
It receives new normalized temporal sequence data and is input to initial intuition Fuzzy Cognitive Map；
Using the position of dynamic fuzzy C mean cluster adjustment intuitionistic fuzzy Cognitive Map cluster centre, obtains dynamic intuitionistic fuzzy and recognize
Know figure；
Adjust intuitionistic fuzzy Cognitive Map cluster centre position during, if current time data successfully fall into one it is existing
Cluster and when being greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value,
Judge whether current time data concept drift occur using drift detection, if so, the weight square of adjustment intuitionistic fuzzy Cognitive Map
Battle array；Otherwise, intuitionistic fuzzy Cognitive Map is not changed.
2. a kind of dynamic intuitionistic fuzzy Cognitive Map construction method as described in claim 1, which is characterized in that excellent using population
Change the weight matrix of algorithm training intuitionistic fuzzy Cognitive Map, process are as follows:
The weight section for providing weight matrix first, initializes weight matrix, then defines population position vector
Element is the weight matrix of intuitionistic fuzzy Cognitive Map, is iterated optimizing, until obtaining optimal weights matrix or meeting iteration end
Only condition.
3. a kind of dynamic intuitionistic fuzzy Cognitive Map construction method as described in claim 1, which is characterized in that in adjustment intuition mould
During the position for pasting Cognitive Map cluster centre, if current time data do not fall within any one existing cluster and utilize adjustment
When the error of the intuitionistic fuzzy Cognitive Map true value found out and predicted value afterwards is greater than default error threshold, be added one it is new poly-
Class center, the cluster centre being newly added are current time data；
Or
Adjust intuitionistic fuzzy Cognitive Map cluster centre position during, if current time data successfully fall into one it is existing
Cluster and it is less than or equal to default error using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value
When threshold value, adjustment intuitionistic fuzzy Cognitive Map is not needed；
Or
Adjust intuitionistic fuzzy Cognitive Map cluster centre position during, if current time data successfully fall into one it is existing
Cluster and when being greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value,
The position of the cluster centre of intuitionistic fuzzy Cognitive Map is updated using dynamic fuzzy C mean value.
4. a kind of dynamic intuitionistic fuzzy Cognitive Map constructs system characterized by comprising
Cluster module is used to that normalized temporal sequence data to be clustered into intuitionistic fuzzy Cognitive Map using fuzzy C-means clustering
Node；
Weight matrix training module is used to train the weight matrix of intuitionistic fuzzy Cognitive Map, obtains initial intuition Fuzzy Cognitive
Figure；
Sequence data receiving module is used to receive new normalized temporal sequence data and is input to initial intuition Fuzzy Cognitive
Figure；
Module is adjusted, is used to obtain using the position of dynamic fuzzy C mean cluster adjustment intuitionistic fuzzy Cognitive Map cluster centre
Dynamic intuitionistic fuzzy Cognitive Map；
Adjust intuitionistic fuzzy Cognitive Map cluster centre position during, if current time data successfully fall into one it is existing
Cluster and when being greater than default error threshold using the error of the intuitionistic fuzzy Cognitive Map true value found out adjusted and predicted value,
Judge whether current time data concept drift occur using drift detection, if so, the weight square of adjustment intuitionistic fuzzy Cognitive Map
Battle array；Otherwise, intuitionistic fuzzy Cognitive Map is not changed.
5. a kind of dynamic intuitionistic fuzzy Cognitive Map as claimed in claim 4 constructs system, which is characterized in that in the weight square
In battle array training module, the weight matrix of particle swarm optimization algorithm training intuitionistic fuzzy Cognitive Map, process are utilized are as follows:
The weight section for providing weight matrix first, initializes weight matrix, then defines population position vector
Element is the weight matrix of intuitionistic fuzzy Cognitive Map, is iterated optimizing, until obtaining optimal weights matrix or meeting iteration end
Only condition.
6. a kind of dynamic intuitionistic fuzzy Cognitive Map as claimed in claim 4 constructs system, which is characterized in that in the adjustment mould
In block, if current time data are not fallen within any one existing cluster and are found out using intuitionistic fuzzy Cognitive Map adjusted true
When the error of real value and predicted value is greater than default error threshold, a new cluster centre is added, the cluster centre being newly added is
Current time data；
Or
In the adjustment module, if current time data successfully fall into an existing cluster and utilize intuitionistic fuzzy adjusted
When the error of true value and predicted value that Cognitive Map is found out is less than or equal to default error threshold, does not need adjustment intuitionistic fuzzy and recognize
Know figure；
Or
In the adjustment module, if current time data successfully fall into an existing cluster and utilize intuitionistic fuzzy adjusted
When the error of true value and predicted value that Cognitive Map is found out is greater than default error threshold, intuition is updated using dynamic fuzzy C mean value
The position of the cluster centre of Fuzzy Cognitive Map.
7. a kind of Time Series Forecasting Methods characterized by comprising
Time series data to be predicted is input in dynamic intuitionistic fuzzy Cognitive Map；The dynamic intuitionistic fuzzy Cognitive Map is adopted
It is realized with the dynamic intuitionistic fuzzy Cognitive Map construction method as described in any one of claim 1-3；
The degree of membership that should be predicted by dynamic intuitionistic fuzzy Cognitive Map output phase；
Anti fuzzy method, the truthful data accordingly predicted are carried out to the degree of membership of prediction.
8. a kind of time series forecasting system characterized by comprising
Data reception module is used to receive time series data to be predicted and is input to dynamic intuitionistic fuzzy Cognitive Map
In；The dynamic intuitionistic fuzzy Cognitive Map uses the dynamic intuitionistic fuzzy Cognitive Map as described in any one of claim 1-3 to construct
Method is realized；
Degree of membership prediction module is used for the degree of membership that should be predicted by dynamic intuitionistic fuzzy Cognitive Map output phase；
Anti fuzzy method module is used to carry out anti fuzzy method, the truthful data accordingly predicted to the degree of membership of prediction.
9. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is held by processor
The step in the dynamic intuitionistic fuzzy Cognitive Map construction method as described in any one of claim 1-3 is realized when row；
Or
The program realizes the step in Time Series Forecasting Methods as claimed in claim 7 when being executed by processor.
10. a kind of computer equipment including memory, processor and stores the meter that can be run on a memory and on a processor
Calculation machine program, which is characterized in that the processor is realized as described in any one of claim 1-3 when executing described program
Step in dynamic intuitionistic fuzzy Cognitive Map construction method；
Or
The processor realizes the step in Time Series Forecasting Methods as claimed in claim 7 when executing described program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910213353.2A CN110084359A (en) | 2019-03-20 | 2019-03-20 | Dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910213353.2A CN110084359A (en) | 2019-03-20 | 2019-03-20 | Dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110084359A true CN110084359A (en) | 2019-08-02 |
Family
ID=67413374
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910213353.2A Pending CN110084359A (en) | 2019-03-20 | 2019-03-20 | Dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084359A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112766633A (en) * | 2020-12-22 | 2021-05-07 | 国网浙江省电力有限公司绍兴供电公司 | Electric wireless heterogeneous network management method and device based on flow balance |
| CN114068012A (en) * | 2021-11-15 | 2022-02-18 | 北京智精灵科技有限公司 | Cognitive decision-oriented multi-dimensional hierarchical drift diffusion model modeling method |
| CN114866290A (en) * | 2022-04-14 | 2022-08-05 | 中国科学技术大学 | Fuzzy behavior decision method and system based on expert system |
| WO2023087917A1 (en) * | 2021-11-17 | 2023-05-25 | 北京智精灵科技有限公司 | Cognitive decision-making evaluation method and system based on multi-dimensional hierarchical drift diffusion model |
-
2019
- 2019-03-20 CN CN201910213353.2A patent/CN110084359A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112766633A (en) * | 2020-12-22 | 2021-05-07 | 国网浙江省电力有限公司绍兴供电公司 | Electric wireless heterogeneous network management method and device based on flow balance |
| CN112766633B (en) * | 2020-12-22 | 2023-10-24 | 国网浙江省电力有限公司绍兴供电公司 | Flow balance-based power wireless heterogeneous network management method and device |
| CN114068012A (en) * | 2021-11-15 | 2022-02-18 | 北京智精灵科技有限公司 | Cognitive decision-oriented multi-dimensional hierarchical drift diffusion model modeling method |
| CN114068012B (en) * | 2021-11-15 | 2022-05-10 | 北京智精灵科技有限公司 | Cognitive decision-oriented multi-dimensional hierarchical drift diffusion model modeling method |
| WO2023087917A1 (en) * | 2021-11-17 | 2023-05-25 | 北京智精灵科技有限公司 | Cognitive decision-making evaluation method and system based on multi-dimensional hierarchical drift diffusion model |
| CN114866290A (en) * | 2022-04-14 | 2022-08-05 | 中国科学技术大学 | Fuzzy behavior decision method and system based on expert system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084359A (en) | Dynamic intuitionistic fuzzy Cognitive Map construction method, Time Series Forecasting Methods and system | |
| Bacanin et al. | Artificial neural networks hidden unit and weight connection optimization by quasi-refection-based learning artificial bee colony algorithm | |
| Angelov et al. | Autonomous learning multimodel systems from data streams | |
| Wang et al. | A cluster-based competitive particle swarm optimizer with a sparse truncation operator for multi-objective optimization | |
| CN113408703A (en) | Multi-modal big data machine automatic learning system based on nerves and symbols | |
| Luo et al. | Fuzzy forecasting for long-term time series based on time-variant fuzzy information granules | |
| CN111079860B (en) | Method for constructing multi-granularity response model of emergency based on scene similarity | |
| CN106530082A (en) | Stock predication method and stock predication system based on multi-machine learning | |
| CN108764577A (en) | Online time series prediction technique based on dynamic fuzzy Cognitive Map | |
| Shan et al. | The data-driven fuzzy cognitive map model and its application to prediction of time series | |
| CN109344962A (en) | Time Series Forecasting Methods based on intuitionistic fuzzy Cognitive Map | |
| CN111444949A (en) | Rule optimization-based data-driven granularity modeling method | |
| CN108985455A (en) | A kind of computer application neural net prediction method and system | |
| Jin et al. | PRECOM: A parallel recommendation engine for control, operations, and management on congested urban traffic networks | |
| Al-Laham et al. | Parameter identification of a multilayer perceptron neural network using an optimized salp swarm algorithm | |
| Hou et al. | Wind power forecasting method of large-scale wind turbine clusters based on DBSCAN clustering and an enhanced hunter-prey optimization algorithm | |
| Du et al. | FELight: Fairness-aware traffic signal control via sample-efficient reinforcement learning | |
| Chen et al. | An adaptive latent factor model via particle swarm optimization for high-dimensional and sparse matrices | |
| Zhang et al. | The prediction of online time series with concept drift based on dynamic intuitionistic fuzzy cognitive map | |
| Khidirova et al. | Neuro-fuzzy algorithm for clustering multidimensional objects in conditions of incomplete data | |
| Slim | Improved Crude Oil Price Forecasting With Statistical Learning Methods | |
| Liu | Fuzzy cognitive maps in GIS data analysis | |
| Ruan et al. | Low-rank Adaptation for Spatio-Temporal Forecasting | |
| He | Bank Customer Churn Prediction Analysis Based on Improved WOA-SVM | |
| Qi et al. | Disease classification model based on Qubit neural tree networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190802 |