CN104636999A - Detection method for building abnormal energy consumption data - Google Patents

Abstract

The invention discloses a detection method for building abnormal energy consumption data. The detection method comprises the following steps of firstly collating original data according to a time series; then, performing judgment by referring to industry indexes so as to find out abnormal data in a specific time slot; then, filtering out an abnormal value in the abnormal energy consumption data by adopting the deviation detection technology in data mining; finally, finding out an abnormal point in the data by adopting a discrete Fourier transform-based time series. The method for performing accurate positioning on the abnormal data in building energy consumption data, provided by the invention, is capable of replacing an empirical threshold judging method generally used in current energy industry; by means of an artificial intelligent analysis method in data mining and the application of the discrete Fourier transform principle, the abnormal energy consumption data is prevented from being misjudged, misreported and omitted, and effective abnormal energy consumption monitoring information is provided for energy consumption departments and personnel.

Description

A kind of building is abnormal with energy data detection method

Technical field

The present invention relates to building energy consumption monitoring field, specifically abnormal the using of a kind of building can data detection method.

Background technology

In building energy consumption monitoring system, it is the important ring ensureing security of system, Effec-tive Function that exception can detect, by this function, operation maintenance personnel can know in real time with can abnormal, accurately fault location, to get rid of or repair system fault in time.Current, judge that exception can data be depend on expert or user experience substantially, with reference to industry index, set fixing limit or threshold value, once monitor data exceeds limit or reaches threshold value and trigger alarm.And in actual motion, with condition impacts such as energy situation climate, personnel, seasons, comparatively complicated by energy data fluctuations situation, dependence arranges single static data and judges extremely obviously to lose precisely.Find by following the tracks of the operation of many systems, abnormal when occurring system acquisition less than, and erroneous judgement wrong report phenomenon happens occasionally, not only can do nothing to help user accurately use can, also can produce interference to user job in some cases.

Summary of the invention

Separate-blas estimation technology in data mining and practical experience judge to combine by the present invention, accurate anomaly sieving data, realize the effective judgement to building exception energy data, to overcome the defect existed in prior art in maximum magnitude.

Object of the present invention is achieved through the following technical solutions:

A kind of building is abnormal with energy data detection method:

Step (1): with set time section for frequency acquisition gathers building energy consumption data, set up the time series D by energy data:

D＝{d _t|t＝t ₀,t ₁,t ₂,...,t _n-1}；

In formula, d _tfor element, represent the image data in t;

Step (2): also based on industry target setting threshold value, compare with this threshold value according to building actual conditions, the abnormal elements in extraction time sequence, i.e. abnormal image data, using the subsequence D of abnormal elements as D ₁; The subsequence D of described abnormal elements as D is removed in time series D ₂: D ₂={ d ₁, d ₂..., d _p;

Step (3): adopt the Outlier mining method based on deviation, with regular length m by D ₂be split as s subsequence, D ₂subsequence D _2sbe expressed as:

D _2s＝{d _s,d _s+1,...,d _s+m-1}，2≤m≤p

In formula, m is the element number of subsequence, and p represents D ₂element number;

Calculate the smoothing factor of each subsequence with dissimilarity function, by smoothing factor with define threshold value comparison, judge whether to exceed: the subsequence exceeded is designated as D ₂₁; The subsequence do not exceeded is designated as normal data sequence;

Step (4): the Time Series Similarity based on discrete Fourier transformation searches D respectively ₁and D ₂₁in abnormal elements, by D ₁and D ₂₁be split as multiple subsequence with regular length respectively, with discrete Fourier transformation, subsequence time series data transformed to domain space from time domain space, obtain Fourier coefficient sequence, Fourier coefficient D _f:

$D_f=\frac{1}{\sqrt{n}}\underset{t=0}{\overset{n-1}{\mathrm{\Σ}}}{d}_t\exp (-\frac{j2\mathrm{\πft}}{n})$

In formula, n is the element number of subsequence, d _tfor the element in subsequence, f=0 ..., n-1;

Given standard energy consumption sequence, setting threshold value, by D ₁and D ₂₁in the subsequence that splits out calculate Euclidean distance respectively and between standard energy consumption sequence, the sequence being greater than given threshold value is abnormal data sequence, and what be less than or equal to threshold value is then normal data sequence.

Building actual conditions and industry index comprise: spring, time period in autumn, air conditioning electricity should be less than standby power consumption amount; Night rest periods, public building room lighting electricity consumption and pole low energy consumption should be maintained with water; Office building power consumption festivals or holidays should be less than power consumption on working day; Commercial building non operation time section energy consumption should lower than business hours section energy consumption; Power consumption values fluctuation range should be relatively steady in normal use procedure for the building such as office building, teaching building.

Preferably, in step (3), dissimilarity function is defined as:

$\mathrm{DST}(A,B)=\frac{({\mathrm{\Σ}}_{i=1}^m{A}_i\⊕B_i)}{m}$

In formula, A is adjacent two numerical value d _q-1, d _qbetween difference, B be follow-up adjacent two number d _q, d _q+1between difference, m is the length of collection object and the number of subsequence element, be xor operator, be defined as $A\⊕B=\left\{\begin{array}{cc}1,& A\≠B\\ 0,& A=B\end{array}\right..$

In step (4), the Euclidean distance of two sequences is expressed as: x in formula _fand y _fbe respectively sequence X and the coefficient of sequence Y after discrete Fourier transformation.

Beneficial effect of the present invention:

The invention provides a kind of method to the abnormal data precise positioning in energy for building data, the empirical value determining method generally used in current energy industry can be replaced, by the analytical approach of artificial intelligence in data mining and the application of discrete Fourier transform (DFT) principle, prevent the erroneous judgement to exception energy data, report by mistake and fail to report, for energy unit and personnel provide effective exception energy monitoring information.

Accompanying drawing explanation

Fig. 1 is the flow chart of steps of the inventive method.

Embodiment

Below in conjunction with specific embodiment, the invention will be further described.

Embodiment 1

A kind of building is abnormal with energy data detection method:

Step (1): with set time section for frequency acquisition gathers building energy consumption data, set up the time series D by energy data:

D＝{d _t|t＝t ₀,t ₁,t ₂,...,t _n-1}；

In formula, d _tfor element, represent the image data in t;

Step (2): also based on industry target setting threshold value, compare with this threshold value according to building actual conditions, the abnormal elements in extraction time sequence, i.e. abnormal image data, using the subsequence D of abnormal elements as D ₁; The subsequence D of described abnormal elements as D is removed in time series D ₂: D ₂={ d ₁, d ₂..., d _p;

Step (3): adopt the Outlier mining method based on deviation, with regular length m by D ₂be split as s subsequence, D ₂subsequence D _2sbe expressed as:

D _2s＝{d _s,d _s+1,...,d _s+m-1}，2≤m≤p

In formula, m is the element number of subsequence, and p represents D ₂element number;

Calculate the smoothing factor of each subsequence with dissimilarity function, by smoothing factor with define threshold value comparison, judge whether to exceed: the subsequence exceeded is designated as D ₂₁; The subsequence do not exceeded is designated as normal data sequence;

Step (4): the Time Series Similarity based on discrete Fourier transformation searches D respectively ₁and D ₂₁in abnormal elements, by D ₁and D ₂₁be split as multiple subsequence with regular length respectively, with discrete Fourier transformation, subsequence time series data transformed to domain space from time domain space, obtain Fourier coefficient sequence, Fourier coefficient D _f:

$D_f=\frac{1}{\sqrt{n}}\underset{t=0}{\overset{n-1}{\mathrm{\Σ}}}{d}_t\exp (-\frac{j2\mathrm{\πft}}{n})$

In formula, n is the element number of subsequence, d _tfor the element in subsequence, f=0 ..., n-1;

Given standard energy consumption sequence, setting threshold value, by D ₁and D ₂₁in the subsequence that splits out calculate Euclidean distance respectively and between standard energy consumption sequence, the sequence being greater than given threshold value is abnormal data sequence, and what be less than or equal to threshold value is then normal data sequence.

Embodiment 2:

As abnormal in building a kind of in embodiment 1 with energy data detection method, spring, time period in autumn, air conditioning electricity should be less than standby power consumption amount; Night rest periods, public building room lighting electricity consumption and pole low energy consumption should be maintained with water; Office building power consumption festivals or holidays should be less than power consumption on working day; Commercial building non operation time section energy consumption should lower than business hours section energy consumption; Power consumption values fluctuation range should be relatively steady in normal use procedure for the building such as office building, teaching building.

Embodiment 3:

As abnormal in building a kind of in embodiment 1 or 2 with energy data detection method, dissimilarity function is defined as:

$\mathrm{DST}(A,B)=\frac{({\mathrm{\Σ}}_{i=1}^m{A}_i\⊕B_i)}{m}$

In formula, A is adjacent two numerical value d _q-1, d _qbetween difference, B be follow-up adjacent two number d _q, d _q+1between difference, m is the length of collection object, be xor operator, be defined as $A\⊕B=\left\{\begin{array}{cc}1,& A\≠B\\ 0,& A=B\end{array}\right..$

In the embodiment above, the subsequence D that again filters out in finding step (2), (3) of the Time Series Similarity changed based on discrete Fourier of step (4) ₁and D ₂₁in abnormal elements.

The energy consumption data of building, there is the change of certain law in its time series, the air conditioning electricity data variation that such as seasonal variations produces, Campus buildings energy consumption regular reduction, commercial building energy consumption regularity during festivals or holidays between winter and summer vacations raises etc.Each class standard energy consumption sequence can be made based on above-mentioned energy consumption data rule change.

Adopt discrete Fourier change principle, multiple data point is connected into curve by time shaft by the method based on time series similarity analysis, line is expanded to by point, similarity between the lines or diversity factor are analyzed, isolated energy consumption data can be connected into regular energy behavior curve thus, therefrom find out outlier, location abnormal data.

By D ₁and D ₂₁be split as multiple subsequence with regular length respectively, by discrete Fourier transform (DFT), subsequence time series data transformed to domain space from time domain space, obtain following fourier coefficient:

$D_f=\frac{1}{\sqrt{n}}\underset{t=0}{\overset{n-1}{\mathrm{\Σ}}}{d}_t\exp (-\frac{j2\mathrm{\πft}}{n})$

Wherein, n is the element number of subsequence, d _tfor the element in subsequence, f=0 ..., n-1;

Theoretical according to Parserval, time-domain and frequency-domain energy spectrum function is identical, also namely for sequence X and sequence Y: $\left|\right|X-Y\left|\right|=\left|\right|\stackrel{\‾}{X}-\stackrel{\‾}{Y}\left|\right|$

Whether similarly use Euclidean distance to weigh two sequences, if the Euclidean distance of two sequences is less than given threshold epsilon, then thinks and namely meet following formula by these two sequence similarities:

$\left|\right|X-Y\left|\right|=\sqrt{\underset{f=0}{\overset{n-1}{\mathrm{\Σ}}}{|x_f-y_f|}^2}\≤\mathrm{\ϵ}$

Theoretical according to Parserval, following equation is also set up:

$\left|\right|\stackrel{\‾}{X}-\stackrel{\‾}{Y}\left|\right|=\sqrt{\underset{f=0}{\overset{n-1}{\mathrm{\Σ}}}{|x_f-y_f|}^2}\≤\mathrm{\ϵ}$

By D ₁and D ₂₁in compute euclidian distances between the subsequence that splits out and standard energy consumption sequence, the sequence being greater than given threshold value is abnormal data sequence, and what be less than or equal to threshold value is then normal data sequence.

Threshold value described in the present invention all provides by limited number of time experiment or experience.

Above embodiment is only for illustration of technical scheme of the present invention; but not limiting the scope of the invention; although done to explain to the present invention with reference to preferred embodiment; those of ordinary skill in the art is to be understood that; can modify to technical scheme of the present invention or equivalent replacement, and not depart from essence and the scope of technical solution of the present invention.

Claims (4)

1. building is abnormal with an energy data detection method, it is characterized in that: method is

Step (1): with set time section for frequency acquisition gathers building energy consumption data, set up the time series D by energy data:

D＝{d _t|t＝t ₀,t ₁,t ₂,...,t _n-1}；

In formula, d _tfor element, represent the image data in t;

Step (2): also based on industry target setting threshold value, compare with this threshold value according to building actual conditions, the abnormal elements in extraction time sequence, using the subsequence D of abnormal elements as D ₁; The subsequence D of described abnormal elements as D is removed in time series D ₂: D ₂={ d ₁, d ₂..., d _p;

Step (3): adopt the Outlier mining method based on deviation, with regular length m by D ₂be split as s subsequence, D ₂subsequence D _2sbe expressed as:

D _2s＝{d _s,d _s+1,...,d _s+m-1}，2≤m≤p

In formula, m is the element number of subsequence, and p represents D ₂element number;

Calculate the smoothing factor of each subsequence with dissimilarity function, by smoothing factor with define threshold value comparison, judge whether to exceed: the subsequence exceeded is designated as D ₂₁; The subsequence do not exceeded is designated as normal data sequence;

Step (4): the Time Series Similarity based on discrete Fourier transformation searches D respectively ₁and D ₂₁in abnormal elements, by D ₁and D ₂₁be split as multiple subsequence with regular length respectively, with discrete Fourier transformation, subsequence time series data transformed to domain space from time domain space, obtain Fourier coefficient sequence, Fourier coefficient D _f:

$D_f=\frac{1}{\sqrt{n}}\underset{t=0}{\overset{n-1}{\mathrm{\Σ}}}{d}_t\exp (-\frac{j2\mathrm{\πft}}{n})$

In formula, n is the element number of subsequence, d _tfor the element in subsequence, f=0 ..., n-1;

Given standard energy consumption sequence, setting threshold value, by D ₁and D ₂₁in the subsequence that splits out calculate Euclidean distance between standard energy consumption sequence respectively, the sequence that Euclidean distance is greater than given threshold value is abnormal data sequence, and what be less than or equal to threshold value is then normal data sequence.

2. one building exception energy data detection method according to claim 1, is characterized in that: building actual conditions and industry index comprise: spring, time period in autumn, air conditioning electricity should be less than standby power consumption amount; Night rest periods, office building, the electricity consumption of teaching building room lighting and pole low energy consumption should be maintained with water; Office building power consumption festivals or holidays should be less than power consumption on working day; Commercial building non operation time section energy consumption should lower than business hours section energy consumption; Power consumption values fluctuation range should be relatively steady in normal use procedure for the building such as office building, teaching building.

3. one building exception energy data detection method according to claim 1, is characterized in that: dissimilarity function is defined as:

$\mathrm{DST}(A,B)=\frac{({\mathrm{\Σ}}_{i=1}^m{A}_i\⊕B_i)}{m}$

In formula, A is adjacent two numerical value d _q-1, d _qbetween difference, B be follow-up adjacent two number d _q, d _q+1between difference, m is the length of collection object, be xor operator, be defined as $A\⊕B=\left\{\begin{array}{cc}1,& A\≠B\\ 0,& A=B\end{array}\right..$

4. one building exception energy data detection method according to claim 1, is characterized in that: the Euclidean distance of two sequences is expressed as: x in formula _fand y _fbe respectively sequence X and the coefficient of sequence Y after discrete Fourier transformation.