CN103559655B - The Forecasting Methodology of the novel feeder line load of microgrid based on data mining - Google Patents

Abstract

The Forecasting Methodology of the novel feeder line load of microgrid based on data mining, relates to the Forecasting Methodology that a kind of microgrid feeder line is loaded. The present invention loads as object taking the novel feeder line of microgrid, and application data digging technology is analyzed all kinds of feeder line load characteristics on microgrid; According to weather data, adopt cluster analysis to set up all kinds of feeder line load patterns; Choose optimization method, determine all kinds of feeder line load forecasting model, carry out derivation algorithm design. The novel feeder line of microgrid provided by the invention load can change the electric energy passive situation of accepting merely system, can be on one's own initiative and major network cooperation, play a positive role at demand response; The novel feeder line load prediction of microgrid can not only be predicted the load total amount of microgrid, can also predict microgrid distributed power generation (wind-powered electricity generation, photovoltaic generation, energy storage) power; Preferably forecast model, revises predicated error, can improve all kinds of feeder line load prediction precision.

Description

The Forecasting Methodology of the novel feeder line load of microgrid based on data mining

Technical field

The present invention relates to the Forecasting Methodology of a kind of microgrid feeder line load, be specifically related to a kind of microgrid based on data mining newThe Forecasting Methodology of type feeder line load.

Background technology

The novel feeder line load of microgrid is by traditional feeder line load (positive carry), photovoltaic and wind-power electricity generation distributed renewable energySource (load) and energy storing device form. The load prediction of novel microgrid feeder line, should consider traditional feeder line load curve peakThe tendency that paddy rises and falls, also will pay close attention to correlation and operational mode between novel microgrid feeder line load. At present domestic not yet haveThe report of correlation technique.

Summary of the invention

The object of this invention is to provide the Forecasting Methodology of the novel feeder line load of a kind of microgrid based on data mining, with microgridNovel feeder line load is object, and application data digging technology is analyzed all kinds of feeder line load characteristics on microgrid; According to weatherData, adopt cluster analysis to set up all kinds of feeder line load patterns; Choose optimization method, as robust regression, BP neutral net, ARMAThe methods such as time prediction model, determine all kinds of feeder line load forecasting model, carry out derivation algorithm design.

The Forecasting Methodology of the novel microgrid feeder line load based on data mining of the present invention, comprises following content:

(1) to the raw data analysis of loading from the undressed all kinds of feeder lines of electrical network collection in worksite, design is realBody-contact figure;

(2) according to feeder line load and weather data structure, set up data warehouse, extract giving birth to data, change, clearlyAfter reason, integrated processing, be loaded in the feeder line load data warehouse of subject-oriented;

(3) adopt fuzzy clustering, all kinds of new by setting up working day and festivals or holidays to all kinds of feeder line loads in data warehouseType feeder line load pattern;

(4) according to all kinds of novel feeder line load operation rules, in method base, choose optimization method and determine novel feeder lineLoad forecasting model, trains and learns by historical data sample, calculates or estimation prediction model parameters;

(5) test that actual observation sample predicts the outcome, statistical analysis predicated error, revises forecast model.

Tool of the present invention has the following advantages:

1, the novel feeder line of microgrid load can change the electric energy passive situation of accepting merely system, can be on one's own initiative and major networkCooperation, plays a positive role at demand response;

2, the magnanimity feeder line load data of electrical network is analyzed, discarded the dross and selected the essential, after data pretreatment, store face intoIn the data warehouse of theme;

3, can obtain the novel all kinds of feeder line Load Characteristic Analysis of microgrid, all kinds of novel feeder line load prediction trend etc. relevantDecision information;

4, the novel feeder line load prediction of microgrid can not only be predicted the load total amount of microgrid, can also predict distributed of microgridElectricity (wind-powered electricity generation, photovoltaic generation, energy storage) power;

5, preferred forecast model, revises predicated error, can improve all kinds of feeder line load prediction precision.

Brief description of the drawings

Fig. 1 is based on the novel feeder line E-R figure being the theme that loads;

Fig. 2 is all kinds of novel feeder line load prediction flow chart based on data mining.

Detailed description of the invention

Below in conjunction with accompanying drawing, technical scheme of the present invention is further described, but is not limited to this, every to thisInvention technical scheme is modified or is equal to replacement, and does not depart from the spirit and scope of technical solution of the present invention, all should containIn protection scope of the present invention.

The Forecasting Methodology of the novel microgrid feeder line load based on data mining of the present invention, comprises following content:

(1) to the raw data analysis of loading from the undressed all kinds of feeder lines of electrical network collection in worksite, design is realBody-contact figure;

(2) according to feeder line load and weather data structure, set up data warehouse, extract giving birth to data, change, clearlyAfter reason, integrated processing, be loaded in the feeder line load data warehouse of subject-oriented;

(3) adopt fuzzy clustering, all kinds of new by setting up working day and festivals or holidays to all kinds of feeder line loads in data warehouseType feeder line load pattern;

(4) according to all kinds of novel feeder line load operation rules, in method base, choose optimization method and determine novel feeder lineLoad forecasting model, trains and learns by historical data sample, calculates or estimation prediction model parameters;

(5) test that actual observation sample predicts the outcome, statistical analysis predicated error, revises forecast model.

In foregoing, specific requirement is as follows:

1, database entity relation design

Data analysis is whole database design, sets up the first step in ground process, is also an of paramount importance step simultaneously.The data analysis of load database will be determined to the various data types of required use in novel feeder line load prediction process.

In forecasting process, in the middle of the various data types of required use, most importantly all kinds of novel feeder lines of distribution are loadedHistorical data. Can be divided into two classes: (1) positive carry: i.e. the sub-load identical with conventional energy flow direction; (2) negativeLotus: with the sub-load of energy flow opposite direction in traditional distribution feeder, this sub-load by near wind-force user's side,Photovoltaic, the generating of energy storage device small distributed power supply provide.

In order to determine the corresponding relation between all kinds of solid data attributes and data, the present invention is to from electrical network collection in worksiteThe raw data analysis of undressed all kinds of feeder lines load after, designed Entity-Relationship figure (EntityRelationshipDiagram, E-R figure), as shown in Figure 1.

2, Data Warehouse Design

Multidimensional structure in database is divided into two class tables: a class is fact table, be used for storing true metric andThe code value of each dimension; Another kind of is dimension table. According to feeder line load and weather data structure, foundation taking feeder line historical load data asFact table and the data warehouse forming with some dimension tables of the factor compositions such as date, time, weather, area, microgrid. Data binsStorehouse is the particular database of subject-oriented, is data platform important in data mining technology, and it can reach efficient data and look intoAsk and retrieval. It is generally made up of fact table (theme) and several dimension tables being associated (time, position, weather parameters etc.).Often there are snowflake type, the model such as star-like. The Data Warehouse Design of Novel distribution network feeder line load, in order to improve retrieval effectiveness, adopts snowFlower pattern type. Corresponding each table specific object design of this data warehouse is as follows:

(1) feeder line information on load is fact table, comprises time numbering, Position Number, and direction numbering, feeder line historical load hasMerit power, feeder line historical load reactive power.

(2) date and time information dimension table, comprises: date numbering, year, month, day, week, season and solar term that the date is corresponding.

(3) temporal information dimension table, comprises: time numbering, date numbering, hour, minute information;

(4) regional information dimension table, comprises: area number, area name;

(5) microgrid information dimension table, comprises: area number, microgrid numbering, microgrid title;

(6) feeder line positional information dimension table, comprises: Position Number, area number, microgrid numbering, feeder line title;

(7) load type information dimension table, comprises: load type numbering, type direction, typonym.

(8) Weather information dimension table, comprises: date numbering, maximum temperature, minimum temperature, weather condition, humidity, air pressure.

The concrete structure design of each table is as shown in table 1-8:

Table 1 feeder line load fact table

Field name	Type	Length	Attribute implication
				Time_ID	Numeral (Int)	4	Time numbering
Date_ID	Numeral (Int)	4	Date numbering
				Location_ID	Numeral (Int)	4	Position Number
Type_ID	Numeral (Int)	4	Load type numbering
				LoadP	Numeral (Double)	8	Feeder line load active power
LoadQ	Numeral (Double)	8	Feeder line reactive load power

Table 2 date and time information dimension table

Table 3 temporal information dimension table

Table 4 regional information dimension table

Field name	Type	Length	Attribute implication
				Region_ID	Numeral (Int)	4	Area ID
Region	Text	20	Netherlands

Table 5 microgrid information dimension table

Table 6 feeder line positional information dimension table

Table 7 load type information dimension table

Table 8 Weather information dimension table

3, data pretreatment

The electric network data transmission being gathered by FTU (line feed terminals unit), RTU (far-end measurement and control unit) device and intelligent electric meterThe host computer system of giving control centre, these data are undressed raw data, they originate different, structure confusion, some numberAccording to imperfect, also often there is bad data or network topology mistake, therefore need these data to re-start extraction, clearReason, the pretreatment such as integrated, and use the identification of state estimation method and revise them, then according to the need of the novel feeder line load prediction of microgridAsk, by data transaction by different data according to certain rule, correction, sequence, polymerization, comprehensive, concentrate after, according to dataWarehouse structure, is loaded in data warehouse, thereby ensures integrality and the reliability of data, reaches and makes full use of various data sourcesObject.

4, set up feeder line load pattern

By the integrated data of solar term factor and climatic factor is classified, set up feeder line load pattern.

If cluster sample weather day, positive carry data type vector was:

$D_k^{\left(1\right)}=(D_{k1}^{\left(1\right)},D_{k2}^{\left(1\right)},D_{k3}^{\left(1\right)},D_{k4}^{\left(1\right)},D_{k5}^{\left(1\right)},D_{k6}^{\left(1\right)})---\left(1\right)$

Wherein k=1,2, L, p; P is the number of days of or two solar term (containing working day or festivals or holidays);It is k dayDay maximum temperature;It is the Daily minimum temperature of k day;It is the rain or shine situation of weather of k day;Be the relative of k dayHumidity;It is the day type (working day or festivals or holidays) of k day;Be the solar term of k day.

If the negative load data type vector of cluster sample weather day photovoltaic is:

$D_k^{\left(2\right)}=(D_{k1}^{\left(2\right)},D_{k2}^{\left(2\right)},D_{k3}^{\left(2\right)},D_{k4}^{\left(2\right)},D_{k5}^{\left(2\right)},D_{k6}^{\left(2\right)})---\left(2\right)$

Wherein k=1,2, L, p; P is the number of days of or two solar term (containing working day or festivals or holidays);Be respectively intensity of solar radiation maximum, mean value, the minimum of a value of k day;Be respectively maximum temperature, mean temperature, the minimum temperature of k day.

Weather integrated data is standardized, by attribute data bi-directional scaling, make it to fall into specific [0,1] districtBetween. Standardize initial data is carried out to linear transformation with min-max. Standardization integrated data can calculate in the following manner:

If the cluster data type vector after standardization is:

$D_k^{\left(s\right)}=({D_{k1}^{\left(s\right)}}^{\′},{D_{k2}^{\left(s\right)}}^{\′},{D_{k3}^{\left(s\right)}}^{\′},{D_{k4}^{\left(s\right)}}^{\′},{D_{k5}^{\left(s\right)}}^{\′},{D_{k6}^{\left(s\right)}}^{\′})---\left(3\right)$

${D_{kl}^{\left(s\right)}}^{\′}=\frac{D_{kl}^{\left(s\right)}-D_l^{\left(\min \right)}}{D_l^{\left(\max \right)}-D_l^{\left(\min \right)}}---\left(4\right)$

WhereinBe respectively dataMinimum and maximum, k=1,2 ..., p; L=1,2 ...,6，s＝1，2。

Use C-mean fuzzy clustering, the feeder line load sample of prediction closer to each other day can be classified as to a class, be designated as respectively X₁(t)，X₂(t)，...，X_n(t), and form thus a feeder line load pattern M, that is:

M＝(X₁(t)，X₂(t)，...，X_n(t))(5)。

Wherein n is the number of feeder line load curve in pattern, and t is the time that feeder line load sample is corresponding.

5, forecast model is chosen

According to all kinds of feeder line load operation rules, set up Forecasting Methodology storehouse (when regression analysis, neural network prediction, ARMABetween the model such as forecast model), from method base, choose suitable method and determine feeder line load forecasting model, design solves calculationMethod, and look-ahead feeder line information on load, to meet the demand of decision-makings at different levels.

6, forecast model derivation algorithm

All kinds of novel feeder line Short Term Load algorithm flow charts as shown in Figure 2. Take 1hour as feeder line load curveEach interval period. All kinds of novel feeder line Short Term Load algorithm steps are as follows:

Step 1: the year, month, day of selecting prediction; Select area, microgrid, the feeder line of prediction.

Step 2: according to solar term and climatic factor, select integrated data type (1) formula.

Step 3: by cluster sample data typeBy (4) formula minimax standardization data type

Step 4: use C-mean fuzzy clustering, choose and n the novel feeder line load curve X that predicts that day is close₁(t)，X₂(t)，...，X_n(t) as the sample that belongs to identical feeder load pattern (5) formula.

Step 5: determine forecast model output vector Y and input matrix X. Y=X_n(t) be at novel feeder line load pattern MIn from day nearest sample curve of prediction, X=(X₁(t)，X₂(t)，...，X_n-1(t)) be other sample curve in same M.

Step 6: according to all kinds of feeder line load characteristics, choose forecast model (robust regression model, L-M neural network predictionModel, ARMA time prediction model etc.) Y=CX, wherein C=(c₁，c₂，....，c_n-1)。

Step 7: the weight coefficient vector C that calculates forecast model. Error of calculation E, that is:IfE≤ε, execution step 8, otherwise, again revise weight coefficient vector C. Wherein ε is precision of prediction, and t is feeder line load sample correspondenceTime.

Step 8: export novel feeder line load power curve Y (t).

Claims (3)

1. the Forecasting Methodology of the novel feeder line load of the microgrid based on data mining, is characterized in that Forecasting Methodology is:

(1) to the raw data analysis of loading from the undressed all kinds of feeder lines of electrical network collection in worksite, design entity-LianThe figure of system;

(2) according to feeder line load and weather data structure, set up data warehouse, raw data are extracted, change, clear up, collectedAfter becoming to process, be loaded in the novel feeder line load data warehouse of subject-oriented;

(3) adopt fuzzy clustering, negative by setting up working day and festivals or holidays all kinds of feeder lines to all kinds of feeder line loads in data warehouseLotus pattern;

(4), according to all kinds of novel feeder line load operation rules, in method base, choose optimization method and determine novel feeder line loadForecast model, trains and learns by historical data sample, calculates or estimation prediction model parameters;

(5) test that actual observation sample predicts the outcome, statistical analysis predicated error, revises forecast model;

Concrete steps are as follows:

Step 1: the year, month, day of selecting prediction; Select area, microgrid, the feeder line of prediction;

Step 2: according to solar term and climatic factor, select according to the following formula integrated data type: $D_k^{\left(1\right)}=(D_{k1}^{\left(1\right)},D_{k2}^{\left(1\right)},D_{k3}^{\left(1\right)},D_{k4}^{\left(1\right)},D_{k5}^{\left(1\right)},D_{k6}^{\left(1\right)});$ Wherein k=1,2, L, p; P is the number of days of or two solar term (containing working day or festivals or holidays);It is the day highest temperature of k dayDegree;It is the Daily minimum temperature of k day;It is the rain or shine situation of weather of k day;Be the relative humidity of k day;It is the day type (working day or festivals or holidays) of k day;Be the solar term of k day;

Step 3: by cluster sample data typeBy following formula minimax standardization data type $D_{\mathrm{kl}}^{{\left(s\right)}^{\′}}=\frac{D_{\mathrm{kl}}^{\left(s\right)}-D_l^{\left(\min \right)}}{D_l^{\left(\max \right)}-D_l^{\left(\min \right)}};$ WhereinBe respectively dataMinimum and maximum, k=1，2，...，p；l＝1，2，...，6，s＝1，2；

Step 4: use C-mean fuzzy clustering, choose and n the novel feeder line load curve X that predicts that day is close₁(t)，X₂(t)，...，X_n(t) as the sample that belongs to identical feeder load pattern, wherein feeder line load pattern M=(X₁(t)，X₂(t)，...，X_n(t))；

Step 5: determine forecast model output vector Y and input matrix X, Y=X_n(t) be from advance in novel feeder line load pattern MSurvey a day nearest sample curve, X=(X₁(t)，X₂(t)，...，X_n-1(t)) be other sample curve in same M;

Step 6: according to all kinds of feeder line load characteristics, choose forecast model Y=CX, wherein C=(c₁，c₂，....，c_n-1)；

Step 7: calculate the weight coefficient vector C of forecast model, the error of calculation,If E≤ε, carries outStep 8, otherwise, again revise weight coefficient vector C, wherein t is the time that feeder line load sample is corresponding;

Step 8: export novel feeder line load power curve Y (t).

2. the Forecasting Methodology of the novel feeder line load of the microgrid based on data mining according to claim 1, is characterized in thatIf described data warehouse is by forming taking feeder line historical load data as fact table with date, time, weather, area, microgridDry dimension table composition.

3. the Forecasting Methodology of the novel feeder line load of the microgrid based on data mining according to claim 1, is characterized in thatDescribed data warehouse is snowflake type.