CN110443376A - State analysis method and its application module based on non-supervisory machine learning algorithm - Google Patents
State analysis method and its application module based on non-supervisory machine learning algorithm Download PDFInfo
- Publication number
- CN110443376A CN110443376A CN201910815181.6A CN201910815181A CN110443376A CN 110443376 A CN110443376 A CN 110443376A CN 201910815181 A CN201910815181 A CN 201910815181A CN 110443376 A CN110443376 A CN 110443376A
- Authority
- CN
- China
- Prior art keywords
- point
- sample
- end module
- sample point
- remote end
- 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.)
- Granted
Links
- 238000010801 machine learning Methods 0.000 title claims abstract description 36
- 238000004458 analytical method Methods 0.000 title claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 19
- 230000008030 elimination Effects 0.000 claims description 8
- 238000003379 elimination reaction Methods 0.000 claims description 8
- 230000001174 ascending effect Effects 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 2
- 101100391182 Dictyostelium discoideum forI gene Proteins 0.000 claims 1
- 230000007812 deficiency Effects 0.000 abstract description 5
- 230000000694 effects Effects 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000000034 method Methods 0.000 description 4
- 238000013450 outlier detection Methods 0.000 description 3
- 230000032683 aging Effects 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000003750 conditioning effect Effects 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 230000006837 decompression Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Remote Monitoring And Control Of Power-Distribution Networks (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention discloses a kind of state analysis methods based on non-supervisory machine learning algorithm, comprising the following steps: 1, remote end module is numbered;2, the output signal of remote end module is acquired;3, the output signal of collected remote end module is arranged, establishes sample point set, and the point in the output signal of remote end module collected and sample point set is corresponded;4, the point in sample point set is substituted into non-supervisory machine learning algorithm, to determine the working condition of remote end module.The invention also discloses a kind of application modules of the state analysis method based on non-supervisory machine learning algorithm described in realize, comprising: energy management apparatus, collection of simulant signal device and digital signal processing device.Have many advantages, such as to solve the problems, such as that the sample of forward and reverse can not obtain, compensates for the deficiency that can not carry out fault modeling and working condition anticipation using Supervised machine learning algorithm.
Description
Technical field
The present invention relates to state analysis technical fields, and in particular to a kind of state based on non-supervisory machine learning algorithm point
Analyse method and its application module.
Background technique
For mesh when light measurement system is run, remote end module missing data before being out of order after a failure can not be defeated
Data out cause the sample of forward and reverse that can not obtain, and also just can not carry out failure using the machine learning algorithm for having supervision
Modeling and anticipation, therefore, it is necessary to a kind of fault diagnosis machine learning algorithms, to the normal remote end module output data of current live
Carry out sample collection, how in numerous fault diagnosis machine learning algorithms, formed can solve above-mentioned technical problem can
Row scheme prejudges the failure of remote end module, is problem urgently to be resolved at present.
Summary of the invention
In view of the deficiencies of the prior art, the object of the present invention is to provide a kind of states based on non-supervisory machine learning algorithm
Analysis method, which is directed to the output signal of direct current transportation light measurement system remote end module, using non-supervisory machine
Device learning algorithm is analyzed the sample characteristics of output signal of the remote end module under different working condition, is realized to remote end module
The working conditions such as aging prejudged.
In view of the deficiencies of the prior art, it is a further object of the present invention to provide a kind of realize to be calculated based on non-supervisory machine learning
The state analysis method and its application module of method.
To achieve the purpose of the present invention, technical solution below is taken: a kind of based on non-supervisory machine learning algorithm
State analysis method may comprise steps of:
N remote end module is numbered in step 1, and the n is positive integer;
Step 2 is respectively acquired the output signal of n remote end module;
Step 3 arranges the output signal of collected n remote end module, establishes sample point set, and will be adopted
The output signal of n remote end module of collection and the point in sample point set correspond;
Point in sample point set is substituted into non-supervisory machine learning algorithm by step 4, to determine the work shape of remote end module
State.
In step 4, the non-supervisory machine learning algorithm can with specifically includes the following steps:
Step a1, the Euclidean distance between sample point and ascending order arrangement are calculated;
Step a2, kth distance and the kth field of each sample point are calculated;
Step a3, the reachable density and reach distance of each sample point are calculated.
In step 4, the working condition of the described judgement remote end module can with specifically includes the following steps:
Step b1, the local outlier factor of each sample point in sample point set is calculated;
If step b2, the part of sample point peels off the factor greater than 1, which is outlier, corresponding to the sample point
Remote end module be the remote end module to peel off;
If step b3, the part of sample point peel off the factor be in close proximity to 1 or be equal to 1, the point be normal point, it is corresponding
Remote end module is normal remote end module.
Step 4 specifically includes the following steps:
Euclidean distance and ascending order arrangement between step 41, calculating sample point: sample point set G is set, n detection is shared
Sample, data dimension s, forFor any two number in sample point set G
Strong point, β=(xj1,xj2,…,xjs), it is d (α, β) from point α to the Euclidean distance of point β:
Wherein, α=(xi1,xi2,…,xis), β=(xj1,xj2,…,xjs) be n tie up theorem in Euclid space two o'clock coordinate,
xi1,…,xisAnd xj1,…,xjsRespectively indicate point α and specific location of the point β in coordinate system;
Step 42, the kth distance for calculating each sample point: d is definedk(α) is the kth distance of point α, dk(α)=d (α, β),
Meet the following conditions:
(1) the k point of at least elimination point α outside, β ' ∈ G { x ≠ α } meet d (α, β ')≤d in sample point set G
(α,β);
(2) at most there is the k-1 point of elimination point α outside in sample point set G, β ' ∈ G { x ≠ α } meets d (α, β ')
≤d(α,β);That is point β is nearest k-th point of range points α, and elimination point α is outside;
Step 43, the kth field for calculating each sample point: N is setk(α) is point α kth apart from neighborhood, is met:
Nk(α)=β ' ∈ G { x ≠ α } | d (α, β ')≤d (α, β) },
Wherein, β ' is the point of elimination point α outside in sample point set G, d (α, β ') and d (α, β) be respectively point α to β ' with
The distance of point α to β, in the sample point set G comprising all to point α distance less than the point of point α kth neighborhood distance, then have
Nk(α)≥k;
Step 44, the reach distance d for calculating each sample pointk(α, β):
dk(α, β)=max { dk(α), d (α, β) },
Wherein, dk(α) is the kth distance of point α, and max expression takes dk(α, β) and dkThe maximum value of (α);
That is the kth reach distance of point β to point α is at least the kth distance of point α, and k nearest point of distance alpha point, they arrive point
The reach distance of α is considered equal, and is equal to dk(α);
Step 45, the reachable density p for calculating point αk(α):
Wherein, dk(α, β) is the reach distance of each sample point, Nk(α) is point α kth apart from neighborhood, expression in point α the
The average reach distance of all point-to-point α in k neighborhood;
Step 46, the local outlier factor LOF for calculating each sample point and descending arrangement, the part of point α peel off the factor
LOFk(α) is defined as:
Wherein, ρk(α), ρk(β) is the reachable density of point α, β;LOFkThe neighborhood N of (α) expression point αkOther points in (α)
The average of the ratio between the local reachability density of local reachability density and point α.If ρk(α) and ρkThe ratio of (β) is equal to 1, illustrates a little
The similar density of point in the neighborhood of α, then point α and neighborhood belong to cluster;If ρk(α) and ρkThe ratio of (β) illustrates a little less than 1
The density of α is higher than its neighborhood dot density, then point α is point off density;If ρk(α) and ρkThe ratio of (β) is greater than 1, illustrates that point α's is close
Degree is less than its neighborhood dot density, then point α is abnormal point;
The factor LOF if the part of point α peels offk(α) > > 1, then sample point set G is the point set that peels off, corresponding distal end
Module is the remote end module to peel off, and otherwise, sample point set G is normal point set, and corresponding remote end module is normal remote
End module.
Another object to realize the present invention takes technical solution below: a kind of realize is based on non-supervisory engineering
Practise the state analysis method and its application module of algorithm, comprising: energy management apparatus, collection of simulant signal device and digital signal
Processing unit, one end of one end connection digital signal processing device of the energy management apparatus, the energy management apparatus
The other end connects one end of collection of simulant signal device, and the other end of the digital signal processing module connects collection of simulant signal
The other end of module, the collection of simulant signal device include sequentially connected Overvoltage protecting unit, voltage lifting unit and width
Adjustment unit is spent, the digital signal processing device includes sequentially connected A/D converter and fpga logic processor, the energy
Measuring managing device includes sequentially connected PPC battery, voltage voltage regulation unit and voltage reference cell.
The application module can also include LED, and the LED is connected with digital signal processing unit.
Extra-high voltage or electric current obtain the pressure in-a 10V~+10V range after divider resistance or shunt resistance
Drop;And the effect measured is exactly acquired to the analog voltage of input and signal condition, and in order to accurately measure this pressure drop, sensing
The front stage circuits part of device selects the operational amplifier of low noise to carry out the conditioning of signal, so that it is guaranteed that signal can preferably by
A/D converter is acquired and converts.
Overvoltage protecting unit: impact and destruction of the abnormal input signal to sensor circuit in order to prevent is needed to defeated
Enter signal to be limited, i.e., adds an overvoltage crowbar in signal input part.It is generally realized using zener diode, surely
Pressure diode is a kind of with the semiconductor devices all before critical breakdown reverse voltage with high-ohmic.
Voltage is lifted unit: this part includes two-stage amplifying circuit and filter circuit, and first order amplifying circuit constitutes instead
To attenuator, while shifted signal being added, second level amplifying circuit constitutes sign-changing amplifier, and two-stage amplifying circuit acts on simultaneously,
So that the range of input signal is -10V~+10V.
Voltage-adjusting unit: the input signal of design is -10V~+10V, the A/D converter energy received signal of sensor
Range is 0V~2.5V, this just needs us to improve input signal.Minimum voltage is raised to 0V, ceiling voltage by us
It is raised to 2.5V.In this way, -10V voltage signal is with regard to corresponding A/D converter output " 0 " ,+10V voltage is with regard to corresponding A/D converter
" 4096 " of output.
The effect of A/D converter is that analog signal is converted to digital signal, for the requirement for meeting measurement accuracy, translation bit
Number is 12.
Data processing equipment selects logic processor FPGA (Field-Programmable Gate Array), main
Effect is that the signal of A/D converter acquisition is calculated, encoded and sent, the final simulation electricity realized from " 0~2.5V "
It is pressed onto the conversion of the Serial No. of " 0~4096 ".In addition, logic processor FPGA also needs to rise before sending digital signal
Beginning position, under-voltage marker and check bit be added in digital signal.Such low and high level is modulated onto LED infraluminescence
The launch wavelength of Guan Shang, LED are 850nm.Electric signal is converted to optical signal, sends finally by optical fiber.
The energy of entire remote end module derives from laser, and the principle of laser is: photovoltaic energy converter (PPC) will connect
The laser energy received is converted into electric signal, then is the energy supply of entire module for power supply.Meanwhile also believing comprising synchronised clock in laser
Number;And the effect of energy management apparatus is then that decompression and voltage reversal are carried out to the voltage of photovoltaic energy converter, is driven
The voltage of device work.
The advantages of the present invention:
1, the present invention is directed to the output signal of direct current transportation light measurement system remote end module, is calculated using non-supervisory machine learning
Method analyzes the sample characteristics of output signal of the remote end module under different working condition, realizes the aging shape to remote end module
State is prejudged, to solve the problems, such as that the sample of forward and reverse can not obtain, compensating for not using has supervision machine
Learning algorithm carries out the deficiency of fault modeling and working condition anticipation.
2, the state analysis method of the invention based on non-supervisory machine learning algorithm is believed for remote end module output data
Number sample characteristics Working state analysis is carried out to remote end module using LOF outlier detection algorithm, to find distal end
Output signal feature before module failure, and then prejudged before remote end module is out of order, to compensate for out
Remote end module missing data before failure, after being out of order can not output data, cause the sample of forward and reverse that can not obtain,
The deficiency of fault modeling and anticipation can not be carried out using the machine learning algorithm for having supervision.
It 3, mainly include three big devices inside remote end module: energy management apparatus, collection of simulant signal device and number letter
Number processing unit, the effect of collection of simulant signal device be exactly the analog voltage of input is acquired and signal condition, thus
Ensure that signal preferably can be acquired and be converted by A/D converter;The effect of digital signal processing device is to turn analog signal
It is melted into digital signal, and energy management apparatus is to power to energize for entire remote end module.In numerous fault diagnosis machine learning
In algorithm, non-supervisory machine learning algorithm is a kind of typically high-precision fault diagnosis machine learning algorithm based on density, non-
Supervision machine learning algorithm can be widely applied to numerous areas, as Telecoms Fraud Analysis, credit card fraud detection, network are attacked
Behavioral value, medical diagnosis and extreme weather weather forecast etc. are hit, in non-supervisory machine learning algorithm of the invention, inputs and is
Data sample point set by calculating the factor that peels off for depending on neighborhood density to each data sample point, and then judges
Whether the data point is outlier.If the factor that peels off is far longer than 1, which is outlier;If the factor that peels off close to
1, then the data point is normal data points.
Detailed description of the invention
Fig. 1 is the overall flow figure of the state analysis method of the invention based on non-supervisory machine learning algorithm.
Fig. 2 is the embodiment flow chart of the state analysis method of the invention based on non-supervisory machine learning algorithm.
Fig. 3 is the structural block diagram of application module of the present invention.
Specific embodiment
Embodiment
The present invention is further illustrated With reference to embodiment.
As shown in Figure 1, a kind of state analysis method based on non-supervisory machine learning algorithm, comprising the following steps:
N remote end module is numbered in step 1, and the n is positive integer;
Step 2 is respectively acquired the output signal of n remote end module;
Step 3 arranges the output signal of collected n remote end module, establishes sample point set, and will be adopted
The output signal of n remote end module of collection and the point in sample point set correspond;
Point in sample point set is substituted into non-supervisory machine learning algorithm by step 4, to determine the work shape of remote end module
State.
As shown in Fig. 2, using the outlier detection algorithm of local outlier factor (LOF), to light measurement system remote end module
Working condition analyzed, state analysis specifically includes following methods and step:
Step 1: a large amount of remote end modules being numbered, respectively the remote end module of number 1, the distal end mould of number 2
The remote end module of block ... number n-1, the remote end module of number n;
Step 2: the output signal of n remote end module being acquired respectively;
Step 3: edit being carried out to the signal of acquisition: sample point set G is established, by n remote end module collected
Output signal and sample point set G in point correspond, respectively sample point 1, sample point 2 ... sample point n-1, sample
Point n;
Step 4: the point in sample point set G is substituted into the outlier detection algorithm of local outlier factor (LOF) to determine
The working condition of remote end module:
1. calculating Euclidean distance and the ascending order arrangement between sample point: sample point set G is set, n detection sample is shared,
Data dimension is s, forFor any two data point in sample point set G, β
=(xj1,xj2,…,xjs), from point α to the Euclidean distance of point β are as follows:
Wherein, α=(xi1,xi2,…,xis), β=(xj1,xj2,…,xjs) be n tie up theorem in Euclid space two o'clock coordinate;
2. calculating the kth distance of each sample point: defining dk(α) is the kth distance of point α, dk(α)=d (α, β), meet with
Lower condition:
1) the k point in set at least not including α, β ' ∈ G { x ≠ α } meet d (α, β ')≤d (α, β);
2) at most there is the k-1 point not including α in set, β ' ∈ G { x ≠ α } meets d (α, β ')≤d (α, β);
That is point β is nearest k-th point of range points α, does not include α;
3. calculating the kth field of each sample point: setting Nk(α) is point α kth apart from neighborhood, is met:
Nk(α)=β ' ∈ G { x ≠ α } | d (α, β ')≤d (α, β) },
Wherein, β ' is the point in set not including α, and d (α, β ') d (α, β) is respectively the distance of point α to β ' and β,
Comprising all points for being less than point α kth neighborhood distance to point α distance in the set, it is apparent from Nk(α)≥k;
4. calculating the reach distance of each sample point:
dk(α, β)=max { dk(α), d (α, β) },
Wherein, dk(α) is the kth distance of point α;
That is the kth reach distance of point β to point α is at least the kth distance of point α, and k nearest point of distance alpha point, they arrive α
Reach distance be considered equal, and be equal to dk(α);
5. calculating the reachable density of each sample point:
Wherein, dk(α, β) is the reach distance of each sample point, Nk(α) is point α kth apart from neighborhood, expression in point α the
The average reach distance of all point-to-point α in k neighborhood;
6. calculating the local outlier factor LOF of each sample point and descending arrangement: the part of point α peels off because of sub-definite
Are as follows:
Wherein, ρk(α), ρk(β) is the reachable density of point α, β;LOFkThe neighborhood N of (α) expression point αkOther points in (α)
The average of the ratio between the local reachability density of local reachability density and point α.If ρk(α) and ρkThe ratio of (β) is equal to 1, illustrates a little
The similar density of point in the neighborhood of α, then point α and neighborhood belong to cluster;If ρk(α) and ρkThe ratio of (β) illustrates a little less than 1
The density of α is higher than its neighborhood dot density, then point α is point off density;If ρk(α) and ρkThe ratio of (β) is greater than 1, illustrates that point α's is close
Degree is less than its neighborhood dot density, then point α is abnormal point;
The factor LOF if the part of point α peels offk(α) > > 1, then sample point set G is the point set that peels off, corresponding distal end
Module is the remote end module to peel off, and otherwise, sample point set G is normal point set, and corresponding remote end module is normal remote
End module.
As shown in figure 3, it is a kind of realize described in the state analysis method based on non-supervisory machine learning algorithm application module,
It include: energy management apparatus, collection of simulant signal device, digital signal processing device and LED, the one of the energy management apparatus
One end of end connection digital signal processing device, the one of the other end connection collection of simulant signal device of the energy management apparatus
End, the other end of the other end connection collection of simulant signal module of the digital signal processing module, the collection of simulant signal
Device includes sequentially connected Overvoltage protecting unit, voltage lifting unit and amplitude adjustment unit, the Digital Signal Processing dress
It sets including sequentially connected A/D converter and fpga logic processor, the energy management apparatus includes sequentially connected PPC electricity
Pond, voltage voltage regulation unit and voltage reference cell, the LED are connected with digital signal processing unit.
Above-listed detailed description is illustrating for possible embodiments of the present invention, and the embodiment is not to limit this hair
Bright the scope of the patents, all equivalence enforcements or change without departing from carried out by the present invention, is intended to be limited solely by the scope of the patents of this case.
Claims (6)
1. a kind of state analysis method based on non-supervisory machine learning algorithm, which comprises the following steps:
N remote end module is numbered in step 1, and the n is positive integer;
Step 2 is respectively acquired the output signal of n remote end module;
Step 3 arranges the output signal of collected n remote end module, establishes sample point set, and will be collected
The output signal of n remote end module and the point in sample point set correspond;
Point in sample point set is substituted into non-supervisory machine learning algorithm by step 4, to determine the working condition of remote end module.
2. the state analysis method according to claim 1 based on non-supervisory machine learning algorithm, which is characterized in that in step
In rapid 4, the non-supervisory machine learning algorithm specifically includes the following steps:
Step a1, the Euclidean distance between sample point and ascending order arrangement are calculated;
Step a2, kth distance and the kth field of each sample point are calculated;
Step a3, the reachable density and reach distance of each sample point are calculated.
3. the state analysis method according to claim 1 based on non-supervisory machine learning algorithm, which is characterized in that in step
In rapid 4, the working condition of the described judgement remote end module specifically includes the following steps:
Step b1, the local outlier factor of each sample point in sample point set is calculated;
If step b2, the part of sample point peels off the factor greater than 1, which is outlier, remote corresponding to the sample point
End module is the remote end module to peel off;
If step b3, the part of sample point peels off the factor equal to 1, which is normal point, and corresponding remote end module is normal
Remote end module.
4. the state analysis method according to claim 1 based on non-supervisory machine learning algorithm, which is characterized in that step
4 specifically includes the following steps:
Euclidean distance and ascending order arrangement between step 41, calculating sample point: sample point set G is set, n detection sample is shared
This, data dimension s, forI=1,2 ... s;For any two number in sample point set G
Strong point, β=(xj1, xj2..., xjs), it is d (α, β) from point α to the Euclidean distance of point β:
Wherein, α=(xi1, xi2..., xis), β=(xj1, xj2..., xjs) be n tie up theorem in Euclid space two o'clock coordinate,
xi1..., xisAnd xj1..., xjsRespectively indicate point α and specific location of the point β in coordinate system;
Step 42, the kth distance for calculating each sample point define dk(α) is the kth distance of point α, dk(α)=d (α, β) meets
The following conditions:
(1) the k point of at least elimination point α outside, β ' ∈ G { x ≠ α } in sample point set G, meet d (α, β ')≤d (α,
β);
(2) at most there is the k-1 point of elimination point α outside in sample point set G, β ' ∈ G { x ≠ α } meets d (α, β ')≤d
(α, β);That is point β is nearest k-th point of range points α, and elimination point α is outside;
Step 43, the kth field for calculating each sample point: N is setk(α) is point α kth apart from neighborhood, is met:
Nk(α)=β ' ∈ G { x ≠ α } | d (α, β ')≤d (α, β) },
Wherein, β ' is the point of elimination point α outside in sample point set G, and d (α, β ') and d (α, β) are respectively that point α to β ' and point α are arrived
The distance of β, in the sample point set G comprising all to point α distance less than the point of point α kth neighborhood distance, then have Nk(α)
≥k;
Step 44, the reach distance d for calculating each sample pointk(α, β):
dk(α, β)=max { dk(α), d (α, β) },
Wherein, dk(α) is the kth distance of point α, and max expression takes dk(α, β) and dkThe maximum value of (α);
That is the kth reach distance of point β to point α is at least the kth distance of point α, and k nearest point of distance alpha point, they are to point α's
Reach distance is considered equal, and is equal to dk(α);
Step 45, the reachable density p for calculating point αk(α):
Wherein, dk(α, β) is the reach distance of each sample point, Nk(α) is point α kth apart from neighborhood, indicates adjacent in the kth of point α
The average reach distance of all point-to-point α in domain;
Step 46, the local outlier factor LOF for calculating each sample point and descending arrangement: the part of point α peels off factor LOFk
(α) is defined as:
Wherein, ρk(α), ρk(β) is the reachable density of point α, β;LOFkThe neighborhood N of (α) expression point αkThe part of other points can in (α)
Up to the average of the ratio between the local reachability density of density and point α;If ρk(α) and ρkThe ratio of (β) is equal to 1, then point α and neighborhood
Belong to cluster;If ρk(α) and ρkThe ratio of (β) is less than 1, then point α is point off density;If ρk(α) and ρkThe ratio of (β) is greater than
1, then point α is abnormal point;
The factor LOF if the part of point α peels offk(α) > > 1, then sample point set G is the point set that peels off, corresponding remote end module
For the remote end module to peel off, otherwise, sample point set G is normal point set, and corresponding remote end module is normal distal end mould
Block.
5. a kind of application module for realizing the state analysis method based on non-supervisory machine learning algorithm described in claim 1,
It is characterized in that, comprising: energy management apparatus, collection of simulant signal device and digital signal processing device, the energy management dress
The one end for one end connection digital signal processing device set, the other end connection collection of simulant signal dress of the energy management apparatus
The one end set, the other end of the other end connection collection of simulant signal module of the digital signal processing module, the simulation letter
Number acquisition device includes sequentially connected Overvoltage protecting unit, voltage lifting unit and amplitude adjustment unit, the digital signal
Processing unit includes sequentially connected A/D converter and fpga logic processor, and the energy management apparatus includes being sequentially connected
PPC battery, voltage voltage regulation unit and voltage reference cell.
6. application module according to claim 5, which is characterized in that it further include LED, the LED and Digital Signal Processing
Unit is connected.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910815181.6A CN110443376B (en) | 2019-08-30 | 2019-08-30 | State analysis method based on non-supervision machine learning algorithm and application module thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910815181.6A CN110443376B (en) | 2019-08-30 | 2019-08-30 | State analysis method based on non-supervision machine learning algorithm and application module thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110443376A true CN110443376A (en) | 2019-11-12 |
CN110443376B CN110443376B (en) | 2024-05-17 |
Family
ID=68438578
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910815181.6A Active CN110443376B (en) | 2019-08-30 | 2019-08-30 | State analysis method based on non-supervision machine learning algorithm and application module thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110443376B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113702894A (en) * | 2021-09-23 | 2021-11-26 | 云南电网有限责任公司电力科学研究院 | Fault diagnosis method and device for remote module and merging unit |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106412841A (en) * | 2016-11-23 | 2017-02-15 | 沈阳航空航天大学 | Indoor location method based on DBSCAN algorithm |
CN106446500A (en) * | 2016-07-14 | 2017-02-22 | 江苏信息职业技术学院 | Improved supervised orthogonal neighborhood preserving embedding dimension reduction-based fault identification method |
CN109100143A (en) * | 2018-07-06 | 2018-12-28 | 华中科技大学 | Fault Diagnosis of Roller Bearings and equipment based on CEEMDAN and CFSFDP |
CN109446184A (en) * | 2018-08-28 | 2019-03-08 | 湖南大唐先科技有限公司 | Power generation big data preprocess method and system based on big data analysis platform |
CN109934354A (en) * | 2019-03-12 | 2019-06-25 | 北京信息科技大学 | Abnormal deviation data examination method based on Active Learning |
-
2019
- 2019-08-30 CN CN201910815181.6A patent/CN110443376B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106446500A (en) * | 2016-07-14 | 2017-02-22 | 江苏信息职业技术学院 | Improved supervised orthogonal neighborhood preserving embedding dimension reduction-based fault identification method |
CN106412841A (en) * | 2016-11-23 | 2017-02-15 | 沈阳航空航天大学 | Indoor location method based on DBSCAN algorithm |
CN109100143A (en) * | 2018-07-06 | 2018-12-28 | 华中科技大学 | Fault Diagnosis of Roller Bearings and equipment based on CEEMDAN and CFSFDP |
CN109446184A (en) * | 2018-08-28 | 2019-03-08 | 湖南大唐先科技有限公司 | Power generation big data preprocess method and system based on big data analysis platform |
CN109934354A (en) * | 2019-03-12 | 2019-06-25 | 北京信息科技大学 | Abnormal deviation data examination method based on Active Learning |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113702894A (en) * | 2021-09-23 | 2021-11-26 | 云南电网有限责任公司电力科学研究院 | Fault diagnosis method and device for remote module and merging unit |
Also Published As
Publication number | Publication date |
---|---|
CN110443376B (en) | 2024-05-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6336528B2 (en) | Methods and systems | |
CN108008176B (en) | A kind of photovoltaic array real-time state monitoring and fault location system | |
JP5584622B2 (en) | Fault detection method for photovoltaic system | |
Leva et al. | PV module fault diagnosis based on microconverters and day-ahead forecast | |
EP2495577A2 (en) | Systems and methods for identifying faulty sensors within a power generation system | |
KR102144350B1 (en) | Failure diagnosing system of solar power generating system | |
WO2023169098A1 (en) | Isolation forest-based method for diagnosing open-circuit fault of modular multilevel converter | |
Zaki et al. | Fault detection and diagnosis of photovoltaic system using fuzzy logic control | |
CN115575754A (en) | Intelligent industrial park electricity information abnormity identification method and system | |
CN110084158B (en) | Electric equipment identification method based on intelligent algorithm | |
CN112003564B (en) | Distributed photovoltaic system branch power abnormity early warning method based on intelligent terminal | |
CN110443376A (en) | State analysis method and its application module based on non-supervisory machine learning algorithm | |
CN113687164A (en) | Non-invasive load event two-stage self-adaptive detection method | |
Dong et al. | Fault diagnosis and classification in photovoltaic systems using scada data | |
CN112734637A (en) | Thermal infrared image processing method and system for monitoring temperature of lead | |
CN210488596U (en) | Application module based on unsupervised machine learning algorithm | |
CN116388685A (en) | Solar cell panel operation control system | |
CN108228800B (en) | Photovoltaic power generation system anomaly detection system and method based on data mining | |
JP2019201533A (en) | Method and apparatus for determining deterioration of solar cell module | |
CN115511111A (en) | Power station operation and maintenance state monitoring method based on time sequence data multi-scale features | |
Guerriero et al. | Real time monitoring of solar fields with cost/revenue analysis of fault fixing | |
CN113848347A (en) | Health state detection method for wind meter of wind driven generator | |
CN104991150A (en) | Storage-state vacuum electronic device residual life detection method | |
CN106301213A (en) | Solar energy equipment diagnostic method | |
CN111398810A (en) | Slip ring electric spark detection and diagnosis system and detection and diagnosis method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |