CN104318489A - Transformer grouping method based on load characteristic analysis - Google Patents

Abstract

Disclosed is a transformer grouping method based on load characteristic analysis. The transformer grouping method based on load characteristic analysis includes the following steps that firstly, transformer load data are processed to form load rate data, and dimensionality reduction is performed to extract three transformer load characteristics such as annual average load rate, load rate fluctuation and load rate trend; then transformer load characteristic samples are formed through index calculation, and a characteristic sample set is subjected to standardization and to clustering grouping processing by a k-medoids algorithm; finally, grouping results are subjected to characteristic analysis. The transformer grouping method based on load characteristic analysis is provided in consideration of differences of different transformer load characteristics, transformer operation grouping results and suggestions are obtained, and a technological means is provided for grasping transformer economical operation conditions and trends.

Description

A kind of transformer grouping method analyzed based on load characteristic

Technical field

The present invention relates to a kind of by setting up transformer load characteristic index and realizing with this method that transformer load hives off.

Background technology

Along with socioeconomic development and industrial economy benefit consciousness improve constantly, reduce the wastage, saves energy day by day become power supply industry pay close attention to focal issue.Therefore, the economical operation adopting technological means to realize transformer controls, and to reducing the wastage, saves energy is of great practical significance.Wherein, the load characteristic analyzing transformer is the basis of transformer being carried out to economical operation control.

Traditional Example of Transformer Economical Run, mainly according to transformer technology parameter, in conjunction with actual load situation, simply judges transformer economic operation situation by the size of load factor.The method can reflect the running status that transformer is current, but be difficult to the account of the history and the future trends that definitely reflect load, the analytical approach of its load factor also only reflects the relation of Rate of average load and maximum load rate, the intensity of load, and do not reflect the low ebb loading condition of operation, the degree of scatter of load, therefore, be necessary comprehensively to analyze transformer load rate, seek the characteristic quantity of more scientific ground reflected load situation, Rational Classification is carried out to the performance driving economy of transformer, and can with this according to the reasonable operation array mode arranging transformer, the economy that in effective raising local distribution network, transformer runs, reduce the electric energy loss that distribution network system is overall.

Summary of the invention

Technical matters to be solved by this invention is the load data according to transformer, analyzes running status and the future load trend of transformer.

Solve the problems of the technologies described above, the technical solution used in the present invention calculates load factor data according to transformer load data and extracts the load characteristic of transformer, according to load characteristic, transformer cluster is hived off again, analyze the internal feature that each hives off, identify each running state of transformer that hives off.

A kind of transformer grouping method based on load characteristic analysis of the present invention specifically comprises the following steps:

Step 1: utilize transformer load data computational load rate data;

Step 2: through Data Dimensionality Reduction, obtains per day load factor load factor data processing;

Step 3: extract transformer load feature, obtains annual load factor, load factor fluctuation and load factor trend, morphogenesis characters sample set;

Step 4: standardization is carried out to feature samples collection, then carry out transformer load k-medoids cluster and hive off;

Step 5: signature analysis is carried out to grouping result.

Load data in described step 1 comprises the time t that gain merit in P_H-high-pressure side, Q_H-high-pressure side is idle and monitor;

Described load factor data are capacity V according to transformer and load factor computing formula calculate load factor the data sequence { (f of transformer ₁, t ₁), (f ₂, t ₂) ..., (f _s, t _s), wherein t ₁, t ₂..., t _sthe ordered arrangement carried out according to time order and function order, (f _i, t _i) represent at t _ithe load factor of moment transformer is f _i.

The Data Dimensionality Reduction of described step 2 is in units of sky, calculates the Rate of average load of each transformer every day, this day Rate of average load

The extraction transformer load feature of described step 3 is specially: the annual load factor of a transformer, load factor fluctuate and the computing formula of load factor trend is respectively:

(1) annual load factor day_f _jit is the per day load factor in jth sky;

(2) load factor fluctuation

(3) calculating of load factor trend is divided into two steps:

The first step: the monthly average load factor calculating each month: day_f _j, wherein ml is the number of days of the l month;

Second step: carry out fitting a straight line with the monthly average load factor of the 1-12 month, obtain the slope of this straight line, be load factor trend wherein

Calculate the characteristic index value of all transformers, morphogenesis characters sample set.

Described step 4 feature samples collection is standardized as:

Carry out standardization to each characteristic index, standardization formula is as follows:

Annual load factor after standardization $\mathrm{NormUYear}\_f_i=\frac{\mathrm{Year}\_f_i-\stackrel{\‾}{\mathrm{Year}\_f}}{\sqrt{\mathrm{Var}\_\mathrm{Year}\_f}},$ Wherein

Load factor fluctuation after standardization wherein

Load factor trend after standardization wherein

The transformer load k-medoids Cluster Classification of described step 4 comprises following sub-step:

S4.1: a Stochastic choice k transformer to hive off initial central point as cluster, and each central point represents one and hives off.

S4.2: after k central point is selected, the transformer of remaining n-k non-central point is grouped into nearest chooses hiving off representated by transformer from it.More specifically: if num_j is the transformer of a non-central point, num_i is a central point (selected object), and d (num_i, num_j) be num_j and all k central point apart from minimum, then num_j belongs to hiving off representated by num_i;

S4.3: to each central point num_i, in hiving off with this, the transformer num_h of each non-central point replaces original central point num_i, calculate its total cost S=current_cost-past_cost, after wherein current_cost represents that the transformer num_h of non-central point replaces original central point num_i, redistribute non-central point hiving off to nearest central point place, calculate non-central point to belonging to hive off the distance summation of central point; Before past_cost represents that the transformer num_h of non-central point replaces original central point num_i, non-central point to belonging to hive off the distance summation of central point;

S4.4: if replace its total cost S of original central point num_i to have the existence being less than 0 at the transformer num_h of all non-central points, then find out one that total cost S is minimum, and replace original corresponding central point with the transformer of this non-central point, form a new k central point;

S4.5: repeat step S4.3, S4.4 until the transformer num_h of all non-central points replaces total cost S of original central point num_i to be all greater than 0.

Described step 5 is specially: respectively to each three characteristic indexs of hiving off, draw probability density figure, analyzes the distribution situation of the transformer under each different loads characteristic index of hiving off further.

According to the scheme of the invention described above embodiment, contemplated by the invention the otherness of different transformer load characteristic, provide a kind of transformer grouping method analyzed based on load characteristic, obtain transformer and run grouping result and suggestion, empirical tests, obvious to running state of transformer classifying quality, for grasp transformer economic operation situation and trend provide a kind of technological means.

Accompanying drawing explanation

Fig. 1 is the transformer grouping method schematic flow sheet analyzed based on load characteristic of the present invention;

Fig. 2 is the 1 each index probability density figure that hives off of the embodiment of the present invention;

Fig. 3 is the 2 each index probability density figure that hive off of the embodiment of the present invention;

Fig. 4 is the 3 each index probability density figure that hive off of the embodiment of the present invention;

Fig. 5 is the 4 each index probability density figure that hive off of the embodiment of the present invention.

Embodiment

Below in conjunction with drawings and Examples, technical scheme of the present invention is described in detail.

As shown in Figure 1, the transformer grouping method analyzed based on load characteristic of the present invention, first adopt and utilize transformer load data to be processed into load factor data, again through dimensionality reduction, extract three transformer load features: the fluctuation of annual load factor, load factor and load factor trend, then transformer load feature samples is formed through index calculate, carry out standardization to feature samples collection and adopt k-medoids algorithm to carry out cluster to it hiving off process, finally signature analysis is carried out to the result of hiving off.

Transformer grouping method based on load characteristic analysis of the present invention comprises the steps:

Step 1: utilize transformer load data computational load rate data;

Step 2: through Data Dimensionality Reduction, obtains per day load factor load factor data processing;

Step 3: extract transformer load feature, obtains annual load factor, load factor fluctuation and load factor trend, morphogenesis characters sample set;

Step 4: standardization is carried out to feature samples collection, then carry out transformer load k-medoids cluster and hive off;

Step 5: signature analysis is carried out to grouping result.

Described step 1 is described as follows:

Load data comprises the time t that gain merit in P_H-high-pressure side, Q_H-high-pressure side is idle and monitor, according to capacity V and the load factor computing formula of transformer obtain the load factor data sequence of transformer

{ (f ₁, t ₁), (f ₂, t ₂) ..., (f _s, t _s), wherein t ₁, t ₂..., t _sthe ordered arrangement carried out according to time order and function order, (f _i, t _i) represent at t _ithe load factor of moment transformer is f _i;

Described step 2 is described as follows:

Load data is that every 3-5 minute record once in like manner load factor data are also just had a record every 3-5 minute, are not suitable for Direct Analysis because data volume is excessive, need to carry out dimensionality reduction to data;

Dimensionality reduction, in units of sky, calculates the Rate of average load of each transformer every day, this day Rate of average load

Described step 3 is described as follows:

Reaction transformer load characteristic index is arranged out further: the fluctuation of annual load factor, load factor and load factor trend according to the per day load factor of transformer that step 2 calculates.The annual load factor of a transformer, load factor fluctuate and the computing formula of load factor trend is respectively

(1) annual load factor day_f _jit is the per day load factor in jth sky

(2) load factor fluctuation

(3) calculating of load factor trend is divided into two steps, the first step: the monthly average load factor calculating each month day_f _j, wherein ml is the number of days of the l month;

Second step: carry out fitting a straight line with the monthly average load factor of the 1-12 month, obtain the slope of this straight line, be load factor trend wherein calculate the characteristic index value of all transformers, morphogenesis characters sample set, in table 1,

Table 1. feature samples collection example

Transformer is numbered	Annual load factor	Load factor fluctuates	Load factor trend
				num_1	Year_f 1	Var_f 1	Trend_f 1
num_2	Year_f 2	Var_f 2	Trend_f 2
				num_3	Year_f 3	Var_f 3	Trend_f 3

…	…	…	…
				num_n-1	Year_f n-1	Var_f n-1	Trend_f n-1
num_n	Year_f n	Var_f n	Trend_f n

Described step 4 is described as follows:

According to the feature samples collection that step 3 is formed, carry out standardization to each characteristic index, standardization formula is as follows:

Annual load factor after standardization $\mathrm{NormYear}\_f_i=\frac{\mathrm{Year}\_f_i-\stackrel{\‾}{\mathrm{Year}\_f}}{\sqrt{\mathrm{Var})\_\mathrm{Year}\_f}},$ Wherein

Load factor fluctuation after standardization wherein

Load factor trend after standardization wherein

Feature samples collection after standardization is in table 2

Feature samples collection example after table 2. standardization

Transformer is numbered	Annual load factor	Load factor fluctuates	Load factor trend
				num_1	NormYear_f 1	NormVar_f 1	NormTrend_f 1
num_2	NormYear_f 2	NormVar_f 2	NormTrend_f 2
				num_3	NormYear_f 3	NormVar_f 3	NormTrend_f 3
…	…	…	…
				num_n-1	NormYear_f n-1	NormVar_f n-1	NormTrend_f n-1
num_n	NormYear_f n	NormVar_f n	NormTrend_f n

Recycling k-medoids clustering algorithm, distance function selects Euclidean distance

$\begin{array}{c}d(\mathrm{num}\_i,\mathrm{num}\_j)=\\ \sqrt{{(\mathrm{NormYear}\_f_i-\mathrm{NormYear}\_f_j)}^2+{(\mathrm{NormVar}\_f_i-\mathrm{NormVar}\_f_j)}^2+{(\mathrm{NormTrend}\_f_i-\mathrm{NormTrend}\_f_j)}^2}\end{array}$

, select suitable cluster numbers k, hive off to the feature samples clustering after standardization, concrete steps are as follows:

S4.1: a Stochastic choice k transformer to hive off initial central point as cluster, and each central point represents one and hives off.

S4.2: after k central point is selected, the transformer of remaining n-k non-central point is grouped into nearest chooses hiving off representated by transformer from it.More specifically: if num_j is the transformer of a non-central point, num_i is a central point (selected object), and d (num_i, num_j) be num_j and all k central point apart from minimum, then num_j belongs to hiving off representated by num_i;

S4.3: to each central point num_i, in hiving off with this, the transformer num_h of each non-central point replaces original central point num_i, calculate its total cost S=current_cost-past_cost, after wherein current_cost represents that the transformer num_h of non-central point replaces original central point num_i, redistribute non-central point hiving off to nearest central point place, calculate non-central point to belonging to hive off the distance summation of central point; Before past_cost represents that the transformer num_h of non-central point replaces original central point num_i, non-central point to belonging to hive off the distance summation of central point;

S4.4: if replace its total cost S of original central point num_i to have the existence being less than 0 at the transformer num_h of all non-central points, then find out one that total cost S is minimum, and replace original corresponding central point with the transformer of this non-central point, form a new k central point;

S4.5: repeat step S4.3, S4.4 until the transformer num_h of all non-central points replaces total cost S of original central point num_i to be all greater than 0.

Described step 5 is described as follows:

Respectively to each three characteristic indexs of hiving off, draw probability density figure, analyze the distribution situation of the transformer under each different loads characteristic index of hiving off further.

Algorithm example

Below by embodiment, and by reference to the accompanying drawings 1, the inventive method is described in further detail.

Step 1: utilize transformer load data computational load rate data;

Extract Guangdong Power Grid all 500kV transformers depressor load data to analyze, extracting the time period is 1 year, wherein the fractional load data of certain transformer are in table 3, the capacity V=250*3MVA=750MVA of this transformer, corresponding load factor data can be obtained, in table 4 according to load factor formula.

The fractional load data of certain, table 3. transformer

Time	P_H	Q_H	Time	P_H	Q_H
						2013-09-23?15:37	540.97	102.346	2013-09-23?18:36	448.371	121.84
2013-09-23?16:19	545.843	107.219	2013-09-23?18:43	472.739	126.714
						2013-09-23?17:05	531.222	92.5984	2013-09-23?18:46	477.613	131.587
2013-09-23?17:12	526.349	87.7248	2013-09-23?19:39	492.234	131.587
						2013-09-23?17:15	521.475	92.5984	2013-09-23?21:05	467.866	121.84
2013-09-23?17:21	511.728	87.7248	2013-09-23?21:11	462.992	116.966
						2013-09-23?17:28	506.854	82.8512	2013-09-23?21:14	458.118	112.093
2013-09-23?17:34	492.234	77.9776	2013-09-23?21:17	453.245	121.84
						2013-09-23?17:37	477.613	121.84	2013-09-23?21:42	433.75	112.093
2013-09-23?17:40	462.992	126.714	2013-09-23?21:49	428.877	116.966
						2013-09-23?17:43	438.624	116.966	2013-09-23?22:05	419.13	116.966
2013-09-23?18:18	424.003	121.84	2013-09-23?22:08	414.256	112.093
						2013-09-23?18:33	438.624	126.714	2013-09-23?22:14	404.509	107.219

The fractional load rate data of certain, table 4. transformer

Time	Load factor	Time	Load factor
				2013-09-23?15:37	73.40884	2013-09-23?18:36	61.95074
2013-09-23?16:19	74.16984	2013-09-23?18:43	65.25691
				2013-09-23?17:05	71.89761	2013-09-23?18:46	66.05443
2013-09-23?17:12	71.14791	2013-09-23?19:39	67.93585
				2013-09-23?17:15	70.61767	2013-09-23?21:05	64.46271

2013-09-23?17:21	69.22571	2013-09-23?21:11	63.67175
				2013-09-23?17:28	68.47745	2013-09-23?21:14	62.8843
2013-09-23?17:34	66.44962	2013-09-23?21:17	62.57809
				2013-09-23?17:37	65.72118	2013-09-23?21:42	59.73332
2013-09-23?17:40	64.0025	2013-09-23?21:49	59.27211
				2013-09-23?17:43	60.52688	2013-09-23?22:05	58.01931
2013-09-23?18:18	58.82154	2013-09-23?22:08	57.22049
				2013-09-23?18:33	60.87473	2013-09-23?22:14	55.797

Step 2: through Data Dimensionality Reduction, obtains per day load factor load factor data processing;

According to the load factor data that step 1 obtains, in units of sky, calculate the Rate of average load of each transformer every day, this day Rate of average load

Step 3: extract transformer load feature, obtains annual load factor, load factor fluctuation and load factor trend, morphogenesis characters sample set;

Fluctuated by the annual load factor of transformer, load factor and the computing formula of load factor trend, obtain feature samples collection, in table 5.

Table 5. feature samples collection

Step 4: standardization is carried out to feature samples collection, then carry out transformer load k-medoids cluster and hive off;

After carrying out standardization to feature samples collection, selected cluster number k is 4, and carry out transformer load k-medoids cluster and hive off, the cluster result of output is in table 6 and table 7.

Table 6. cluster result (one)

Table 7. cluster result (two)

Step 5: signature analysis is carried out to grouping result

According to the grouping result that step 4 obtains, each is hived off, respectively for annual load factor, load factor fluctuation and load factor trend feature index, draw probability density figure, the distribution situation of the transformer of each different loads characteristic index of hiving off of further analysis, evaluates different out-of-limit running status of hiving off.

1) hive off 1 signature analysis (see Fig. 2)

As can be seen from Figure 2, for hiving off 1, the annual load factor of transformer is in medium level, roughly between 20% to 60%; Load factor fluctuation is comparatively large, roughly between 0.1 to 0.4; Load factor trend is roughly between-0.4 to 0.2.Illustrate hive off 1 transformer running and comparing economy but not steady.

2) hive off 2 signature analysises (see Fig. 3)

As can be seen from Figure 3, for hiving off 2, the annual load factor of transformer is in reduced levels, roughly between 10% to 40%; Load factor fluctuation is less, roughly between 0 to 0.15; Load factor trend is roughly between-0.1 to 0.4.Although illustrate hive off 2 transformer running and comparing steady, run not too economical, loading trends is in ascent stage simultaneously, need notice user power utilization situation, at the growth of load to a certain extent, considers dilatation.

3) hive off 3 signature analysises (see Fig. 4)

As can be seen from Figure 4, for hiving off 3, the annual load factor of transformer is in medium level, roughly between 10% to 60%; Load factor fluctuation is comparatively large, roughly between 0 to 0.4; Load factor trend is roughly between 0 to 1.2.Illustrate hive off 3 transformer run less expensive, but run steadily and also loading trends be in ascent stage, user power utilization situation need be noticed, at the growth of load to a certain extent, consider dilatation.

4) hive off 4 signature analysises (see Fig. 5)

As can be seen from Figure 5, for hiving off 4, the annual load factor of transformer is in medium level, roughly between 30% to 60%; Load factor fluctuates for medium level, roughly between 0 to 0.2; Load factor trend is roughly between-0.2 to 0.4.Illustrate hive off 4 transformer run less expensive simultaneously more steady.

Claims (7)

1., based on the transformer grouping method that load characteristic is analyzed, it is characterized in that comprising the following steps:

Step 1: utilize transformer load data computational load rate data;

Step 2: through Data Dimensionality Reduction, obtains per day load factor load factor data processing;

Step 3: extract transformer load feature, obtains annual load factor, load factor fluctuation and load factor trend, morphogenesis characters sample set;

Step 4: standardization is carried out to feature samples collection, then carry out transformer load k-medoids cluster and hive off;

Step 5: signature analysis is carried out to grouping result.

2. the transformer grouping method analyzed based on load characteristic according to claim 1, is characterized in that: the load data in described step 1 comprises the time t that gain merit in P_H-high-pressure side, Q_H-high-pressure side is idle and monitor;

Described load factor data are capacity V according to transformer and load factor computing formula calculate load factor the data sequence { (f of transformer ₁, t ₁), (f ₂, t ₂) ..., (f _s, t _s), wherein t ₁, t ₂..., t _sthe ordered arrangement carried out according to time order and function order, (f _i, t _i) represent at t _ithe load factor of moment transformer is f _i.

3. the transformer grouping method analyzed based on load characteristic according to claim 1, is characterized in that: the Data Dimensionality Reduction of described step 2 is in units of sky, calculates the Rate of average load of each transformer every day, this day Rate of average load

4. the transformer grouping method analyzed based on load characteristic according to claim 1, is characterized in that: the extraction transformer load feature of described step 3 is specially: the annual load factor of a transformer, load factor fluctuate and the computing formula of load factor trend is respectively:

(1) annual load factor day_f _jit is the per day load factor in jth sky;

(2) load factor fluctuation

(3) calculating of load factor trend is divided into two steps:

The first step: the monthly average load factor calculating each month: wherein m _lfor the number of days in January;

Second step: carry out fitting a straight line with the monthly average load factor of the 1-12 month, obtain the slope of this straight line, be load factor trend wherein calculate the characteristic index value of all transformers, morphogenesis characters sample set.

5. the transformer grouping method analyzed based on load characteristic according to claim 1, is characterized in that: being standardized as feature samples collection of described step 4:

Carry out standardization to each characteristic index, standardization formula is as follows:

Annual load factor after standardization $\mathrm{NormYear}\_f_i=\frac{\mathrm{Year}\_f_i-\stackrel{\‾}{\mathrm{Year}\_f}}{\sqrt{\mathrm{Var}\_\mathrm{Year}\_f}},$ Wherein

Load factor fluctuation after standardization $\mathrm{NormVar}\_f_i=\frac{\mathrm{Var}\_f_i-\stackrel{\‾}{\mathrm{Var}\_f}}{\sqrt{\mathrm{Var}\_\mathrm{Var}\_f}},$ Wherein

Load factor trend after standardization $\mathrm{NormTrend}\_f_i=\frac{\mathrm{Trend}\_f_i-\stackrel{\‾}{\mathrm{Trend}\_f}}{\sqrt{\mathrm{Var}\_\mathrm{Trend}\_f}},$ Wherein

6. the transformer grouping method analyzed based on load characteristic according to claim 1, is characterized in that: the transformer load k-medoids Cluster Classification of described step 4 comprises following sub-step:

S4.1: a Stochastic choice k transformer to hive off initial central point as cluster, and each central point represents one and hives off;

S4.2: after k central point is selected, the transformer of remaining n-k non-central point is grouped into nearest chooses hiving off representated by transformer from it;

More specifically: if num_j is the transformer of a non-central point, num_i is a central point (selected object), and d (num_i, num_j) be num_j and all k central point apart from minimum, then num_j belongs to hiving off representated by num_i;

S4.3: to each central point num_i, in hiving off with this, the transformer num_h of each non-central point replaces original central point num_i, calculate its total cost S=current_cost-past_cost, after wherein current_cost represents that the transformer num_h of non-central point replaces original central point num_i, redistribute non-central point hiving off to nearest central point place, calculate non-central point to belonging to hive off the distance summation of central point; Before past_cost represents that the transformer num_h of non-central point replaces original central point num_i, non-central point to belonging to hive off the distance summation of central point;

S4.4: if replace its total cost S of original central point num_i to have the existence being less than 0 at the transformer num_h of all non-central points, then find out one that total cost S is minimum, and replace original corresponding central point with the transformer of this non-central point, form a new k central point;

S4.5: repeat step S4.3, S4.4 until the transformer num_h of all non-central points replaces total cost S of original central point num_i to be all greater than 0.

7. the transformer grouping method analyzed based on load characteristic according to claim 1, it is characterized in that: described step 5 is specially: respectively to each three characteristic indexs of hiving off, draw probability density figure, analyze the distribution situation of the transformer under each different loads characteristic index of hiving off further.