CN109376907A - Adapt to the high-voltage distribution network transformer substation load forecasting method of transmission and distribution network integration planning - Google Patents
Adapt to the high-voltage distribution network transformer substation load forecasting method of transmission and distribution network integration planning Download PDFInfo
- Publication number
- CN109376907A CN109376907A CN201811115905.8A CN201811115905A CN109376907A CN 109376907 A CN109376907 A CN 109376907A CN 201811115905 A CN201811115905 A CN 201811115905A CN 109376907 A CN109376907 A CN 109376907A
- Authority
- CN
- China
- Prior art keywords
- grid
- load
- layer
- sample
- year
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention discloses a kind of high-voltage distribution network transformer substation load forecasting methods of adaptation transmission and distribution network integration planning.Currently, the prediction mode of power distribution network is difficult to adapt to entire districts and cities' development, it is more difficult to run through transmission and distribution network load prediction with a kind of method.Technical solution of the present invention includes: to plan characteristics of demand in conjunction with transmission & distribution net integration, the high-voltage distribution network load forecasting method combined using Elman neural network model and Spatial Load Forecasting, it is prediction unit with high-voltage distribution network transformer substation donor site, it is combined from historical load and spatial prediction and high-voltage distribution network Mid-long Term Load is predicted, introduce the donor site carry calculation model based on grid exploitation degree.The present invention combines the distant view peak load density data advantage having in space planning, while dynamically introducing substation's measured data over the years in calculating sample and being modified to grid exploitation degree；After prediction obtains grid exploitation degree, year power transformation donor site that looks to the future divides variation and recalculates to obtain high-voltage distribution network donor site load condition.
Description
Technical field
The present invention relates to substation donor site load prediction field, specifically a kind of adaptation transmission and distribution network integration planning
High-voltage distribution network transformer substation load forecasting method.
Background technique
Currently, the method for electricity demand forecasting mainly divides two major classes: traditional prediction method and intelligent Forecasting.
Traditional prediction method specifically includes that elastic coefficient method, regression analysis, time series forecasting, output value unit consumption method
And their derivation method.Intelligent Forecasting mainly includes neural network, fuzzy logic, expert system etc..The above method is equal
With stronger specific aim, it is suitble to the more difficult reasonable segmentation load of prediction mode in whole area, the prediction mode of power distribution network is difficult
To adapt to entire districts and cities' development, it is more difficult to run through transmission and distribution network load prediction with a kind of method.
Summary of the invention
To solve the above-mentioned problems of the prior art, the present invention provides one kind and adapts to transmission and distribution network integration planning need
It asks, for the prediction technique of substation donor site load in high voltage distribution network.
The technical solution adopted by the invention is as follows: a kind of high-voltage distribution network transformer substation for adapting to transmission and distribution network integration planning is negative
Lotus prediction technique comprising:
Characteristics of demand is planned in conjunction with transmission & distribution net integration, is mutually tied using Elman neural network model with Spatial Load Forecasting
The high-voltage distribution network load forecasting method of conjunction is prediction unit with high-voltage distribution network transformer substation donor site, from historical load and spatial prediction
It combines and high-voltage distribution network Mid-long Term Load is predicted, introduce the donor site carry calculation model based on grid exploitation degree.
Present invention introduces grid exploitation degree concepts, breach conventionally employed firm demand mode corresponding with grid, linking
Substation donor site historical load trend and tessellated mesh load condition, meet the need of regional distribution network substation load prediction
It asks.
Supplement as above-mentioned technical proposal, the particular content of the donor site carry calculation model based on grid exploitation degree is such as
Under:
High voltage distribution network carries out effective connection by the method for operation and superior voltage grade sharing of load, passes through website donor site
Interior space and geographical is divided to be adapted with junior's distribution network load forecast of distribution；If districts and cities are divided into m grid according to urban planning,
The gross area is Sm, k 110 kv substations, using 110 kv substation donor sites as analyzed area, power transformation in 1 year are shared in range
Stand total e in donor siteiA grid, the vector set of grid block area are Si={ s (1i),s(2i),…,s(ei), have
When calculating grid block load, only note and 10 kilovolts and following large user, peak load intensity vector integrate as Pi max
={ pmax(1i),pmax(2i),…,pmax(ei), grid inside maximum volume rate vector set is βiMax={ βmax(1i),βmax
(2i),…,βmax(ei), introduce region exploitation degree vector γi={ γ (1i),γ(2i),…,γ(ei) indicate grid internal loading
Exploitation degree, wherein γ (ei) ∈ [0,1], introduce grid land use degree θi={ θ (1i),θ(2i),…,θ(ei) represent certain period
The ratio of interior grid plot ratio and target construction scale, wherein θ (ei)∈[0,1]；
35 kilovolts and above direct-furnish large user load be with load density due to that can not multiply to obtain in grid, according to
There is Y by applying to install data statisticsi={ y (1i),y(2i),…,y(ei), non-industry class value is 0, and it is Δ Y that current year, which increases the amount of applying to install newly,i,
Have:
In formula, uiFor simultaneity factor；WiFor gridding synthetic load in 1 year 110 kilovolts of donor site；Due to general soil benefit
There is certain positive relationship with construction progress and load development progress, to accelerate to calculate convergence, if | θ (ei)-γ(ei)|∈[0,
0.2]；Pi,βiRespectively 1 year actual load intensity vector and actual volume rate vector.
The present invention describes grid year by year and regional economy-load grid planning maximum value ratio situation using exploitation degree,
By limiting exploitation degree feature and predicting, to be fitted under practical area development economic growth situation to net region load
Influence.
Supplement as above-mentioned technical proposal, substation donor site load forecasting method are using improved Elman network
Model is trained historical sample:
Weight is corrected using following various calculating for history gridding load data:
For " S " type curve amendment sample, amendment weight is calculated using following methods:
In the above formulas, k indicates that the sequence of calculation number, i, j respectively indicate the i-th row of matrix, jth column；Respectively
For the connection weight that input layer is calculated to hidden layer kth, k+1 times,For the amendment weight of input layer to hidden layer；The connection weight that layer is calculated to hidden layer kth, k+1 times is respectively accepted,To accept layer to hidden layer
Amendment weight；The respectively connection weight that is calculated to output layer kth, k+1 times of hidden layer,For
Hidden layer to output layer amendment weight；ηh、ηc、ηoutThe respectively learning rate factor of hidden layer, undertaking layer, output layer；Real output value is calculated for i-th of sample kth time,Acquisition value, i.e. sample value are calculated for i-th of sample kth time,
EpFor error target function, a, b, c, d are sensibility control coefrficient, according to search sensitivity adjustment；Xh、Respectively
For the output valve of hidden layer, the output valve of i-th sample hidden layer, the output valve of j-th sample hidden layer；Xc、Point
The output valve of layer Wei not accepted, i-th of sample accepts the output valve of layer, j-th of sample accepts the output valve of layer；For
- 1 calculating acquisition value of j-th of sample kth；
The amendment weight being calculatedTo connection weightIt is updated,
Obtain the connection weight that prediction calculates.
The calculating logic of supplement as above-mentioned technical proposal, the Elman neural network model is as follows:
For the neural network model, input vectorFor n-dimensional vector, hidden layer and undertaking layer output vectorAnd
For n+1 dimensional vector, export as single value；Consider above situation, WhFor the connection weight of input layer to hidden layer, dimension is (n+
1) × n, WcFor the connection weight for accepting layer to hidden layer, dimension is (n+1) × (n+1), WoutFor the company of hidden layer to output layer
Weight is connect, the operational formula of dimension n+1, Elman neural network model is as follows:
Xh(k)=f [Wh·Xc(k)+Wc·Sin(k-1)],
Xc(k)=Xh(k-1),
Sout(k)=g (Wout·Xh(k)),
In formula, k indicates the sequence of calculation number, f [Wh·Xc(k)+Wc·SinIt (k-1)] is hidden layer element excitation function, g
(Wout·XhIt (k)) is the excitation function of output layer unit；When sample training, error target function is as follows:
Real output value is calculated for i-th of sample kth time,It calculates and obtains for i-th of sample kth time.
Supplement as above-mentioned technical proposal is learnt while being trained using historical sample using " S " type curve
Sample is as amendment sample；
Historical sample:
Counties and cities area entire for past i carries out gridding decomposition and calculating, and regional classification is m subregion, grid regions
The vector set of block area is S={ s (1), s (2) ..., s (m) }, for grid s (m), is had:
βim=si(m)/s (m),
Wherein, βimFor the plot ratio of 1 year m grid,Maximum volume rate, s are planned for m gridiIt (m) is m grid
1 year exploitation area；θimFor 1 year m grid land use degree；
Calculate 1 year grid exploitation degree γi(m), have:
Meeting the 110 kv substation donor sites for covering m grid simultaneously has:
γ'i={ γ ' (1i),γ'(2i),…,γ'(ei),
If | γ 'i(m)-γi(m) | > ζ,
γi(m)=| γ 'i(m)+γi(m) |/2,
Wherein, wimIt is 1 year critical point load of m grid, w'imIt is 1 year predicted value to 1 year critical point load on m grid,
yimIt is large user's load, γ ' in the gridiIt is the 110 kv substation donor sites comprising m grid in 1 year donor site Intranet
Lattice exploitation degree vector set, W'iIt is 110 kv substation in 1 year load measurement data, γ 'iIt (m) is negative by 1 year
The m grid exploitation degree that lotus Actual measurement obtains, ζ is divergence threshold, and exploitation degree is checked and corrected；Not for m grid
To be saturated the load in period, Δ yiFor 1 year newly-increased large user's load, Δ Yi={ Δ y1,...ΔynBig for what is increased newly
Customer charge Vector Groups, uiFor simultaneity factor；
" S " type curve learning sample:
Considering economy and electricity consumption development, planning the year two thousand forty is that electricity consumption is saturated year, with 1991 to count initial year,
50 years periods power load experience " S " type curve reaches saturated level；
" S " type curve is added in sample and is trained jointly, the fitting degree of exploitation degree sample training is promoted, for assisting sample
This, there is matched curve:
It is described in conjunction with area grid exploitation degree are as follows:
Wherein, i is time, cmFor the land status of m grid, cm={ living, commercial affairs, municipal administration, finance, industry, amusement } is right
Answer cm={ 1,2,3,4,5,6 }, as land character difference is modified curve, learning sample uses history actual grid
Numerical value and matched curve learn jointly.
Supplement as above-mentioned technical proposal, input, output valve are calculated for exploitation degree and land development degree:
Single input is carried out using 5 years historical datas in the past, land status, and 1 year, the input of m grid are as follows:
xm(i)={ cm,sm,γi-4,γi-3,γi-2,γi-1,γi,θi,εi-4,εi-3,εi-2,εi-1,εi}；
Wherein, xmIt (i) is the input quantity of 1 year m grid, cm,smThe respectively land status and area of m grid, by upper
One trifle grid is calculated 5 years (containing current year) exploitation degree situation γi-4~γi, this districts and cities is obtained 5 years by statistical data
Interior economic growth rate εi-4~εi；
Output quantity ym(i)=(γi+1,θi+1)；
Using historical data and matched curve as learning sample, elman neural network model is trained, letter is obtained
Number accepts layer, input layer to middle layer and middle layer to the connection weight of output layerPrediction is brought into calculate；Often
Secondary prediction generates the predicted value of an exploitation degree and land use degree using the every wheel iteration of aforementioned elman neural network model
γi+1,θi+1, and next calculating is carried it into, iteration 3 times, predicts 3 years backward gridding load developments, obtains m altogether
3 years exploitation degree of a grid and land use degree, are denoted as { γ (mi+1),γ(mi+2),γ(mi+3),θ(mi+1),θ(mi+2),θ
(mi+3)}；Whole m grid in city obtains 3 years exploitation degree predicted values of each grid in the whole city after carrying out prediction calculating over the ground, according to 110
The annual supply district of kv substation is split grid, indicates are as follows: Si+b={ s (1i+b),s(2i+b),…,s(ei+b), b
∈ { 1,2,3 } calculates donor site predicted load Wi+b:
The present invention is contacted grid space-time load density and planning maximum mesh load density using exploitation degree concept, is extracted
In load development process, have continuous tendency variable (exploitation degree), the historical data base that formation data can calculate, with " S "
Type curve is calculated as aid sample using the prediction for improving Eman neural network progress grid exploitation degree；After obtaining predicted value,
Using 110 kilovolts of donor sites as scope of statistics, the donor site internal loading predicted value under the following grid exploitation degree and following donor site range is calculated
Situation.
The present invention solidifies abundant adaptive mess, and the characteristic of power transformation donor site variation carries out Modeling and Design, combines space rule
The distant view peak load density data advantage having in drawing, while substation's measured data over the years is dynamically introduced in calculating sample
Grid exploitation degree is modified；After prediction obtains grid exploitation degree, year power transformation donor site that looks to the future divides variation and recalculates
Obtain high-voltage distribution network donor site load condition.
Detailed description of the invention
Fig. 1 is modified Delphi approach model structural form figure in the embodiment of the present invention；
Fig. 2 is the flow chart of conventional electrical distribution net gridding load forecasting method；
Fig. 3 is exploitation degree progress curve figure in the embodiment of the present invention；
Fig. 4 is the flow chart of load forecasting method in the embodiment of the present invention.
Specific embodiment
With reference to the accompanying drawings of the specification and specific embodiment the present invention is further described.
1.Elman neural network model
It is made of input layer, hidden layer, undertaking layer, output layer, hidden layer uses nonlinear activation function, accepts layer and obtains
Hidden layer n-th is taken to export, feedback effect is calculated in hidden layer (n+1)th time.It is needed in conjunction with the present invention, network input layer vector dimension
Degree is n, and considering that implicit level accepts layer dimension is n+1, is considered for computational accuracy and algorithm logic, herein for property defeated
Layer number of elements is set as 1 out.
For the neural network, input vectorFor n-dimensional vector, hidden layer and undertaking layer output vectorAndFor n+
1 dimensional vector exports as single value.Consider above situation, WhFor the connection weight of input layer to hidden layer, dimension be (n+1) ×
N, WcFor the connection weight for accepting layer to hidden layer, dimension is (n+1) × (n+1), WoutFor the connection weight of hidden layer to output layer
Value, dimension n+1.The operation of neural network is as follows:
Xh(k)=f [Wh·Xc(k)+Wc·Sin(k-1)],
Xc(k)=Xh(k-1),
Sout(k)=g (Wout·Xh(k)),
In formula, k indicates the sequence of calculation number, f [Wh·Xc(k)+Wc·SinIt (k-1)] is hidden layer element excitation function, g
(Wout·XhIt (k)) be the excitation function f (k) of output layer unit is hidden layer element excitation function.When sample training, error mesh
Scalar functions EpIt is as follows:
Real output value is calculated for i-th of sample kth time,Acquisition value is calculated for i-th of sample kth time
(sample value).
2. gridding analytic approach
Classified zoning predicted method is the Spatial Load Forecasting method based on classified zoning principle, is also known as load in many documents
Density index method, this method are the use according to city regulatory detailed planning figure (hereinafter referred to as regulatory control figure) by planning from different places
Electrical property is classified, then carries out subregion to planning area, i.e., will plan that area carries out classified zoning according to regulatory control figure.It completes
On the basis of classified zoning, each partition load historical data is collected, is referred in conjunction with the load density in domestic and international flourishing city or area
Mark, selects the load density target of different load nature of electricity consumed and different subregions, to predict the power load distributing feelings in planning area
Condition and saturation loading.Conventional mesh load forecasting method is as shown in Figure 2.
3. granularity selects
Load grid granularity is broadly divided into large, medium and small three classes, adapts to the analytical calculation for not having to bore.Wherein, bulky grain
Number using administrative section as an area Ge great, or using the physical relieves such as the communal facilitys such as major trunk roads, national highway, greenbelt or river as
Boundary；Medium-grain is generally 3-5km2, generally with land character, street, greening, river etc. for boundary；Small particle size with
City regulatory control figure is standard, and general every piece of region area is in 1km2Below.Consider that present invention is generally directed to high voltage distribution network developments
It calculates, needs to consider to refine development in donor site, therefore use small particle size.
Since city regulatory control figure has detailed planning to each functional areas, planning personnel more can easily determine different
Subregion load type.Each cell load variation is influenced smaller by power equipment or substation's supply district variation.Due to each
The power load property of cell will not change with power equipment and substation's supply district and be changed, therefore the history of each cell
Load data is still one of the reliable basis for analyzing its load rule of development.
Embodiment
The present embodiment provides it is a kind of adaptation transmission and distribution network integration planning high-voltage distribution network transformer substation load forecasting method,
Include: that the integration of transmission & distribution net is combined to plan characteristics of demand, is combined using Elman neural network model and Spatial Load Forecasting
High-voltage distribution network load forecasting method is prediction unit with high-voltage distribution network transformer substation donor site, mutually ties from historical load with spatial prediction
High-voltage distribution network Mid-long Term Load is predicted in conjunction, introduces the donor site carry calculation model based on grid exploitation degree.Above-mentioned high pressure
The detailed process of distribution network transformer substation load forecasting method is detailed in Fig. 4.
Consider in integrated planning process, substation donor site and area, county's range are important base's range, and in power grid
In the case that layering and zoning is basically completed, High-Voltage Network (such as 110 kilovolts) website donor site is normally within the scope of district, transregional turn of confession
Load is smaller to entire effect after Distribution Network Frame is made in order substantially.Traditional province, city and gridding load prediction are with space and administration
Zoning range carries out, and effective is connected after prediction with more difficult for balance, the calculating of power grid, website etc. needed for planning.
The particular content of donor site carry calculation model based on grid exploitation degree is as follows:
High voltage distribution network carries out effective connection by the method for operation and superior voltage grade sharing of load, passes through website donor site
Interior space and geographical is divided to be adapted with junior's distribution network load forecast of distribution；If districts and cities are divided into m grid according to urban planning,
The gross area is Sm, k 110 kv substations, using 110 kv substation donor sites as analyzed area, power transformation in 1 year are shared in range
Stand total e in donor siteiA grid, the vector set of grid block area are Si={ s (1i),s(2i),…,s(ei), have
Consider 35 kilovolts and above networking direct-furnish large user is directly accessed 220 kilovolts and above becomes
Power station, when calculating grid block load, only note and 10 kilovolts and following large user (35 kilovolts or more isobaric grade Firsthand Users
Individually consider in subsequent calculating), peak load intensity vector integrates as Pi max={ pmax(1i),pmax(2i),…,pmax(ei),
Maximum volume rate vector set is β inside gridi max={ βmax(1i),βmax(2i),…,βmax(ei), introduce region exploitation degree to
Measure γi={ γ (1i),γ(2i),…,γ(ei) indicate grid internal loading exploitation degree, wherein γ (ei) ∈ [0,1], introduce grid
Land use degree θi={ θ (1i),θ(2i),…,θ(ei) represent the ratio of certain period Intranet lattice plot ratio and target construction scale
Example, wherein θ (ei)∈[0,1]；
35 kilovolts and above direct-furnish large user load be with load density due to that can not multiply to obtain in grid, according to
There is Y by applying to install data statisticsi={ y (1i),y(2i),…,y(ei) (non-industry class value is 0), it is Δ Y that current year, which increases the amount of applying to install newly,i,
Have:
In formula, uiFor simultaneity factor (all grids and it is non-concurrent reach load maximum value, integral load is represented most with simultaneity factor
When big, average meshes load accounts for the ratio of itself maximum value)；WiFor gridding synthetic load in 1 year 110 kilovolts of donor site；
Since general land use construction progress and load development progress have certain positive relationship, to accelerate to calculate convergence, if | θ
(ei)-γ(ei)|∈[0,0.2]；Pi,βiRespectively 1 year actual load intensity vector and actual volume rate vector.
The present invention describes grid year by year and regional economy-load grid planning maximum value ratio situation using exploitation degree,
By limiting exploitation degree feature and predicting, to be fitted under practical area development economic growth situation to net region load
Influence.
Substation donor site load forecasting method is trained historical sample using improved Elman network model:
Weight is corrected using following various calculating for history gridding load data:
For " S " type curve amendment sample, amendment weight is calculated using following methods:
In the above formulas, k indicates that the sequence of calculation number, i, j respectively indicate the i-th row of matrix, jth column；Respectively
For the connection weight that input layer is calculated to hidden layer kth, k+1 times,For the amendment weight of input layer to hidden layer；The connection weight that layer is calculated to hidden layer kth, k+1 times is respectively accepted,To accept layer to hidden layer
Amendment weight；The respectively connection weight that is calculated to output layer kth, k+1 times of hidden layer,For
Hidden layer to output layer amendment weight；ηh、ηc、ηoutThe respectively learning rate factor of hidden layer, undertaking layer, output layer；Real output value is calculated for i-th of sample kth time,Acquisition value, i.e. sample value are calculated for i-th of sample kth time,
EpFor error target function, a, b, c, d are sensibility control coefrficient, according to search sensitivity adjustment；Xh、Respectively
For the output valve of hidden layer, the output valve of i-th sample hidden layer, the output valve of j-th sample hidden layer；Xc、Point
The output valve of layer Wei not accepted, i-th of sample accepts the output valve of layer, j-th of sample accepts the output valve of layer；For
- 1 calculating acquisition value of j-th of sample kth；
The amendment weight being calculatedTo connection weightIt is updated,
Obtain the connection weight that prediction calculates.
The calculating logic of the Elman neural network model is as follows:
For the neural network model, input vectorFor n-dimensional vector, hidden layer and undertaking layer output vectorAnd
For n+1 dimensional vector, export as single value；Consider above situation, WhFor the connection weight of input layer to hidden layer, dimension is (n+
1) × n, WcFor the connection weight for accepting layer to hidden layer, dimension is (n+1) × (n+1), WoutFor the company of hidden layer to output layer
Weight is connect, the operational formula of dimension n+1, Elman neural network model is as follows:
Xh(k)=f [Wh·Xc(k)+Wc·Sin(k-1)],
Xc(k)=Xh(k-1),
Sout(k)=g (Wout·Xh(k)),
In formula, k indicates the sequence of calculation number, f [Wh·Xc(k)+Wc·SinIt (k-1)] is hidden layer element excitation function, g
(Wout·XhIt (k)) is the excitation function of output layer unit；When sample training, error target function is as follows:
Real output value is calculated for i-th of sample kth time,It calculates and obtains for i-th of sample kth time.
While being trained using historical sample, using " S " type curve learning sample as amendment sample；
Historical sample:
Counties and cities area entire for past i carries out gridding decomposition and calculating, and regional classification is m subregion, grid regions
The vector set of block area is S={ s (1), s (2) ..., s (m) }, for grid s (m), is had:
βim=si(m)/s (m),
Wherein, βimFor the plot ratio of 1 year m grid,Maximum volume rate, s are planned for m gridiIt (m) is m grid
1 year exploitation area；θimFor 1 year m grid land use degree；
Calculate 1 year grid exploitation degree γi(m), have:
Meeting the 110 kv substation donor sites for covering m grid simultaneously has:
γ'i={ γ ' (1i),γ'(2i),…,γ'(ei),
If | γ 'i(m)-γi(m) | > ζ,
γi(m)=| γ 'i(m)+γi(m) |/2,
Wherein, wimIt is 1 year critical point load of m grid, w'imIt is 1 year predicted value to 1 year critical point load on m grid,
yimIt is large user's load, γ ' in the gridiIt is the 110 kv substation donor sites comprising m grid in 1 year donor site Intranet
Lattice exploitation degree vector set, W'iIt is 110 kv substation in 1 year load measurement data, γ 'iIt (m) is negative by 1 year
The m grid exploitation degree that lotus Actual measurement obtains, ζ is divergence threshold, and exploitation degree is checked and corrected；Not for m grid
To be saturated the load in period, Δ yiFor 1 year newly-increased large user's load, Δ Yi={ Δ y1,...ΔynBig for what is increased newly
Customer charge Vector Groups, uiFor simultaneity factor (all grids and it is non-concurrent reach load maximum value, integral load is represented with simultaneity factor
When maximum, average meshes load accounts for the ratio of itself maximum value).
" S " type curve learning sample:
Considering economy and electricity consumption development, planning the year two thousand forty is that electricity consumption is saturated year, with 1991 to count initial year,
50 years periods power load experience " S " type curve reaches saturated level, as shown in Figure 3.
" S " type curve is added in sample and is trained jointly, the fitting degree of exploitation degree sample training is promoted, for assisting sample
This, there is matched curve:
It is described in conjunction with area grid exploitation degree are as follows:
Wherein, i is time, cmFor the land status of m grid, cm={ living, commercial affairs, municipal administration, finance, industry, amusement } is right
Answer cm={ 1,2,3,4,5,6 }, as land character difference is modified curve, learning sample uses history actual grid
Numerical value and matched curve learn jointly.
For cmIn the case of={ 1,2,3,4,5,6 }, i ∈ [0,50] (step-length takes 0.2) is substituted into f respectivelym(i), it calculates
To set array F { i, fm(i),cm, " S " type is presented on two-dimensional coordinate, therefore that " S " type learning sample refers to is exactly array F
{i,fm(i),cm}.I=10.2, c are worked as in citingmIn the case of=3, f is calculatedm(i)=0.085, then { 10.2,0.085,3 } are exactly to count
An element in group.
Input, output valve are calculated for exploitation degree and land development degree:
Single input is carried out using 5 years historical datas in the past, land status, and 1 year, the input of m grid are as follows:
xm(i)={ cm,sm,γi-4,γi-3,γi-2,γi-1,γi,θi,εi-4,εi-3,εi-2,εi-1,εi}；
Wherein, xmIt (i) is the input quantity of 1 year m grid, cm,smThe respectively land status and area of m grid, by upper
One trifle grid is calculated 5 years (containing current year) exploitation degree situation γi-4~γi, this districts and cities is obtained 5 years by statistical data
Interior economic growth rate εi-4~εi；
Output quantity ym(i)=(γi+1,θi+1)；
Using historical data and matched curve as learning sample, elman neural network model is trained, letter is obtained
Number accepts layer, input layer to middle layer and middle layer to the connection weight of output layerPrediction is brought into calculate；Often
Secondary prediction generates the predicted value of an exploitation degree and land use degree using the every wheel iteration of aforementioned elman neural network model
γi+1,θi+1, and next calculating is carried it into, iteration 3 times, predicts 3 years backward gridding load developments, obtains m altogether
3 years exploitation degree of a grid and land use degree, are denoted as { γ (mi+1),γ(mi+2),γ(mi+3),θ(mi+1),θ(mi+2),θ
(mi+3)}；Whole m grid in city obtains 3 years exploitation degree predicted values of each grid in the whole city after carrying out prediction calculating over the ground, according to 110
The annual supply district of kv substation is split grid, indicates are as follows: Si+b={ s (1i+b),s(2i+b),…,s(ei+b), b
∈ { 1,2,3 } calculates donor site predicted load Wi+b:
Application examples
Steps are as follows for relevant calculation and realization:
1. by districts and cities' range carry out gridding segmentation, be divided into m grid, can according to urban planning carry out grid segmentation with
It formulates.
2. transferring power grid historical data within the scope of districts and cities, the affiliated 1 year load of substation of corresponding grid and m net are inquired
Lattice critical point data.
3. calculating each grid 1 year exploitation degree historical data using the donor site carry calculation model of grid exploitation degree.
4. calculating all grids and all time history exploitation degree and land use degree, completes historical sample and obtain.
5. all grids in pair area carry out qualitative classification, and use " S " type function fitting amendment sample.
6. historical sample and amendment sample are imported Elman neural network according to input/output format to learn, obtain
The connection weight of grid exploitation degree.
7. historical data is imported Elman neural network, is calculated using above-mentioned weight, obtain the exploitation in i+1 year
Spend predicted value.
8. iterating to calculate using new predicted value, latter 3 years grid exploitation degree and land use degree are obtained.
9. calculative substation donor site range information is imported, using the donor site carry calculation model meter of grid exploitation degree
Calculation obtains latter 3 years substation's donor site predicted loads.
Claims (6)
1. a kind of high-voltage distribution network transformer substation load forecasting method for adapting to transmission and distribution network integration planning characterized by comprising
Characteristics of demand is planned in conjunction with transmission & distribution net integration, is combined using Elman neural network model and Spatial Load Forecasting
High-voltage distribution network load forecasting method is prediction unit with high-voltage distribution network transformer substation donor site, mutually ties from historical load with spatial prediction
High-voltage distribution network Mid-long Term Load is predicted in conjunction, introduces the donor site carry calculation model based on grid exploitation degree.
2. high-voltage distribution network transformer substation load forecasting method according to claim 1, which is characterized in that be based on grid exploitation degree
Donor site carry calculation model particular content it is as follows:
High voltage distribution network carries out effective connection by the method for operation and superior voltage grade sharing of load, by empty in website donor site
Between geographical divide be adapted with junior's distribution network load forecast of distribution；If districts and cities are divided into m grid, total face according to urban planning
Product is Sm, k 110 kv substations are shared in range, using 110 kv substation donor sites as analyzed area, substation is supplied within 1 year
Total e in areaiA grid, the vector set of grid block area are Si={ s (1i),s(2i),…,s(ei), have
When calculating grid block load, only note and 10 kilovolts and following large user, peak load intensity vector integrate as Pi max={ pmax
(1i),pmax(2i),…,pmax(ei), grid inside maximum volume rate vector set is βi max={ βmax(1i),βmax(2i),…,
βmax(ei), introduce region exploitation degree vector γi={ γ (1i),γ(2i),…,γ(ei) indicate grid internal loading exploitation degree,
Wherein γ (ei) ∈ [0,1], introduce grid land use degree θi={ θ (1i),θ(2i),…,θ(ei) represent certain period Intranet lattice
The ratio of plot ratio and target construction scale, wherein θ (ei)∈[0,1]；
35 kilovolts and above direct-furnish large user load with load density due to that can not multiply to obtain in grid, by report
Dress data statistics has Yi={ y (1i),y(2i),…,y(ei), non-industry class value is 0, and it is Δ Y that current year, which increases the amount of applying to install newly,i, have:
In formula, uiFor simultaneity factor；WiFor gridding synthetic load in 1 year 110 kilovolts of donor site；To accelerate to calculate convergence, if | θ
(ei)-γ(ei)|∈[0,0.2]；Pi,βiRespectively 1 year actual load intensity vector and actual volume rate vector.
3. high-voltage distribution network transformer substation load forecasting method according to claim 1 or 2, which is characterized in that using improvement
Elman neural network model is trained historical sample；
Weight is corrected using following various calculating for history gridding load data:
For " S " type curve learning sample, amendment weight is calculated using following methods:
In the above formulas, k indicates that the sequence of calculation number, i, j respectively indicate the i-th row of matrix, jth column；It is respectively defeated
Enter the connection weight that layer is calculated to hidden layer kth, k+1 times,For the amendment weight of input layer to hidden layer；The connection weight that layer is calculated to hidden layer kth, k+1 times is respectively accepted,To accept layer to hidden layer
Amendment weight；The respectively connection weight that is calculated to output layer kth, k+1 times of hidden layer,For
Hidden layer to output layer amendment weight；ηh、ηc、ηoutThe respectively learning rate factor of hidden layer, undertaking layer, output layer；Real output value is calculated for i-th of sample kth time,Acquisition value, i.e. sample value are calculated for i-th of sample kth time,
EpFor error target function, a, b, c, d are sensibility control coefrficient, according to search sensitivity adjustment；Xh、Respectively
For the output valve of hidden layer, the output valve of i-th sample hidden layer, the output valve of j-th sample hidden layer；Xc、Point
The output valve of layer Wei not accepted, i-th of sample accepts the output valve of layer, j-th of sample accepts the output valve of layer；For
- 1 calculating acquisition value of j-th of sample kth；
The amendment weight being calculatedTo connection weightIt is updated, obtains
Predict the connection weight calculated.
4. high-voltage distribution network transformer substation load forecasting method according to claim 3, which is characterized in that the Elman nerve
The calculating logic of network model is as follows:
For the neural network model, input vectorFor n-dimensional vector, hidden layer and undertaking layer output vectorAndFor n+
1 dimensional vector exports as single value；Consider above situation, WhFor the connection weight of input layer to hidden layer, dimension be (n+1) ×
N, WcFor the connection weight for accepting layer to hidden layer, dimension is (n+1) × (n+1), WoutFor the connection weight of hidden layer to output layer
Value, the operational formula of dimension n+1, Elman neural network model are as follows:
Xh(k)=f [Wh·Xc(k)+Wc·Sin(k-1)],
Xc(k)=Xh(k-1),
Sout(k)=g (Wout·Xh(k)),
In formula, k indicates the sequence of calculation number, f [Wh·Xc(k)+Wc·SinIt (k-1)] is hidden layer element excitation function, g (Wout·
XhIt (k)) is the excitation function of output layer unit；When sample training, error target function is as follows:
Real output value is calculated for i-th of sample kth time,It calculates and obtains for i-th of sample kth time.
5. high-voltage distribution network transformer substation load forecasting method according to claim 3, which is characterized in that using historical sample
While being trained, using " S " type curve learning sample as amendment sample；
Historical sample:
Counties and cities area entire for past i carries out gridding decomposition and calculating, and regional classification is m subregion, grid block face
Long-pending vector set is S={ s (1), s (2) ..., s (m) }, for grid s (m), is had:
βim=si(m)/s (m),
Wherein, βimFor the plot ratio of 1 year m grid,Maximum volume rate is planned for m grid, and si (m) is m grid 1 year
Exploitation area；θimFor 1 year m grid land use degree；
Calculate 1 year grid exploitation degree γi(m), have:
Meeting the 110 kv substation donor sites for covering m grid simultaneously has:
γ′i={ γ ' (1i),γ'(2i),…,γ'(ei),
If | γ 'i(m)-γi(m) | > ζ,
γi(m)=| γ 'i(m)+γi(m) |/2,
Wherein, wimIt is 1 year critical point load of m grid, w'imIt is 1 year predicted value to 1 year critical point load on m grid, yimIt is
Large user's load, γ ' in the gridiFor the 110 kv substation donor sites comprising m grid, grid is opened in 1 year donor site
Hair degree vector set, W 'iIt is 110 kv substation in 1 year load measurement data, γ 'iIt (m) is to pass through 1 year load reality
The m grid exploitation degree being calculated is surveyed, ζ is divergence threshold, and exploitation degree is checked and corrected；It is full for m grid future
With the load in period, Δ yiFor 1 year newly-increased large user's load, Δ Yi={ Δ y1,...ΔynIt is newly-increased large user
Load Vector Groups, uiFor simultaneity factor；
" S " type curve learning sample:
Consider economy and electricity consumption development, planning the year two thousand forty is that electricity consumption is saturated year, with 1991 to count initial year, during which
Power load experience " S " type curve reaches saturated level within 50 years；
" S " type curve is added in sample and be trained jointly, the fitting degree of exploitation degree sample training is promoted, for aid sample,
There is matched curve:
It is described in conjunction with area grid exploitation degree are as follows:
Wherein, i is time, cmFor the land status of m grid, cmThe corresponding c of={ living, commercial affairs, municipal administration, finance, industry, amusement }m
={ 1,2,3,4,5,6 }, as land character difference is modified curve；Learning sample uses history actual grid numerical value
And matched curve learns jointly.
6. high-voltage distribution network transformer substation load forecasting method according to claim 5, which is characterized in that input, output valve needle
Exploitation degree and land development degree are calculated:
Single input is carried out using 5 years historical datas in the past, land status, and 1 year, m grid input are as follows: xm(i)={ cm,
sm,γi-4,γi-3,γi-2,γi-1,γi,θi,εi-4,εi-3,εi-2,εi-1,εi}；
Wherein, xmIt (i) is the input quantity of 1 year m grid, cm,smThe respectively land status and area of m grid is small by upper one
Grid 5 years exploitation degree situation γ are calculated in sectioni-4~γi, economic growth rate in this districts and cities 5 years is obtained by statistical data
εi-4~εi；
Output quantity ym(i)=(γi+1,θi+1)；
Using historical data and matched curve as learning sample, elman neural network model is trained, function is obtained and holds
Connect layer, input layer to middle layer and middle layer to output layer connection weightPrediction is brought into calculate；It is pre- every time
Survey the predicted value γ that an exploitation degree and land use degree are generated using the every wheel iteration of aforementioned elman neural network modeli+1,
θi+1, and next calculating is carried it into, iteration 3 times, predicts 3 years backward gridding load developments, obtains m-th of net altogether
3 years exploitation degree of lattice and land use degree, are denoted as { γ (mi+1),γ(mi+2),γ(mi+3),θ(mi+1),θ(mi+2),θ(mi+3)}；
Whole m grid in city obtains 3 years exploitation degree predicted values of each grid in the whole city after carrying out prediction calculating over the ground, is become according to 110 kilovolts
The annual supply district in power station is split grid, indicates are as follows: Si+b={ s (1i+b),s(2i+b),…,s(ei+b), b ∈ 1,2,
3 }, donor site predicted load W is calculatedi+b:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811115905.8A CN109376907B (en) | 2018-09-25 | 2018-09-25 | High-voltage distribution network transformer substation load prediction method suitable for integrated planning of transmission and distribution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811115905.8A CN109376907B (en) | 2018-09-25 | 2018-09-25 | High-voltage distribution network transformer substation load prediction method suitable for integrated planning of transmission and distribution network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109376907A true CN109376907A (en) | 2019-02-22 |
| CN109376907B CN109376907B (en) | 2022-06-28 |
Family
ID=65401765
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811115905.8A Active CN109376907B (en) | 2018-09-25 | 2018-09-25 | High-voltage distribution network transformer substation load prediction method suitable for integrated planning of transmission and distribution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109376907B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902868A (en) * | 2019-02-25 | 2019-06-18 | 国网河南省电力公司电力科学研究院 | A kind of large user's industry expansion aided analysis method and device based on part throttle characteristics |
| CN110009136A (en) * | 2019-03-12 | 2019-07-12 | 国网江西省电力有限公司电力科学研究院 | A kind of load forecasting method of distribution transformer and distribution line |
| CN110765562A (en) * | 2019-10-08 | 2020-02-07 | 杭州阳斯信息技术有限公司 | Target grid frame automatic generation method based on multi-source information multi-constraints |
| CN111401638A (en) * | 2020-03-17 | 2020-07-10 | 国网上海市电力公司 | Space load prediction method based on extreme learning machine and load density index method |
| CN113723719A (en) * | 2021-11-03 | 2021-11-30 | 国网北京市电力公司 | Distribution transformation-based near-intermediate load prediction method, device, equipment and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104599055A (en) * | 2015-01-04 | 2015-05-06 | 湘潭大学 | Power load demand management method for iron and steel enterprises |
| CN104636822A (en) * | 2015-01-21 | 2015-05-20 | 广州市香港科大霍英东研究院 | Residential load prediction method of elman-based neural network |
| CN105303259A (en) * | 2015-11-06 | 2016-02-03 | 国网上海市电力公司 | Gridding concept based power distribution network planning management method |
| CN107067104A (en) * | 2017-03-30 | 2017-08-18 | 昆明能讯科技有限责任公司 | A kind of power supply grid load forecasting method based on GIS-Geographic Information System |
| CN107169589A (en) * | 2017-04-12 | 2017-09-15 | 佛山电力设计院有限公司 | A kind of low pressure grid based on KNN and roulette algorithm becomes more meticulous load forecasting method |
-
2018
- 2018-09-25 CN CN201811115905.8A patent/CN109376907B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104599055A (en) * | 2015-01-04 | 2015-05-06 | 湘潭大学 | Power load demand management method for iron and steel enterprises |
| CN104636822A (en) * | 2015-01-21 | 2015-05-20 | 广州市香港科大霍英东研究院 | Residential load prediction method of elman-based neural network |
| CN105303259A (en) * | 2015-11-06 | 2016-02-03 | 国网上海市电力公司 | Gridding concept based power distribution network planning management method |
| CN107067104A (en) * | 2017-03-30 | 2017-08-18 | 昆明能讯科技有限责任公司 | A kind of power supply grid load forecasting method based on GIS-Geographic Information System |
| CN107169589A (en) * | 2017-04-12 | 2017-09-15 | 佛山电力设计院有限公司 | A kind of low pressure grid based on KNN and roulette algorithm becomes more meticulous load forecasting method |
Non-Patent Citations (2)
| Title |
|---|
| XI HAN等: "Power Load Forecasting based on improved Elman Neural Network", 《2018 ICEI》 * |
| 傅荣荣: "基于多级聚类和改进Elman神经网络的空间负荷预测方法", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902868A (en) * | 2019-02-25 | 2019-06-18 | 国网河南省电力公司电力科学研究院 | A kind of large user's industry expansion aided analysis method and device based on part throttle characteristics |
| CN110009136A (en) * | 2019-03-12 | 2019-07-12 | 国网江西省电力有限公司电力科学研究院 | A kind of load forecasting method of distribution transformer and distribution line |
| CN110765562A (en) * | 2019-10-08 | 2020-02-07 | 杭州阳斯信息技术有限公司 | Target grid frame automatic generation method based on multi-source information multi-constraints |
| CN110765562B (en) * | 2019-10-08 | 2023-09-01 | 杭州阳斯信息技术有限公司 | Automatic target net rack generation method based on multi-source information and multi-constraint |
| CN111401638A (en) * | 2020-03-17 | 2020-07-10 | 国网上海市电力公司 | Space load prediction method based on extreme learning machine and load density index method |
| CN111401638B (en) * | 2020-03-17 | 2024-02-02 | 国网上海市电力公司 | Spatial load prediction method based on extreme learning machine and load density index method |
| CN113723719A (en) * | 2021-11-03 | 2021-11-30 | 国网北京市电力公司 | Distribution transformation-based near-intermediate load prediction method, device, equipment and medium |
| CN113723719B (en) * | 2021-11-03 | 2022-02-11 | 国网北京市电力公司 | Distribution transformation-based near-intermediate load prediction method, device, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109376907B (en) | 2022-06-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109376907A (en) | Adapt to the high-voltage distribution network transformer substation load forecasting method of transmission and distribution network integration planning | |
| CN112765800B (en) | Design method of distributed water resource configuration model | |
| Chen et al. | Improving flood forecasting capability of physically based distributed hydrological models by parameter optimization | |
| Wang et al. | Dynamic simulation patterns and spatiotemporal analysis of land-use/land-cover changes in the Wuhan metropolitan area, China | |
| CN105930931A (en) | Electric power engineering cost management method | |
| CN108304978A (en) | A kind of mid-term Electric Power Load Forecast method based on data clusters theory | |
| CN106355334A (en) | Farmland construction area determining method | |
| Yang et al. | Nonlinear optimization set pair analysis model (NOSPAM) for assessing water resource renewability | |
| CN105184455A (en) | High dimension visualized analysis method facing urban electric power data analysis | |
| CN107516168A (en) | A kind of Synthetic Assessment of Eco-environment Quality method | |
| Li et al. | Evaluation of urban green space landscape planning scheme based on PSO-BP neural network model | |
| CN103886220B (en) | Water data discretization method for setting weight based on BP network and Gini coefficient | |
| CN103049651A (en) | Method and device used for power load aggregation | |
| CN108520345A (en) | Evaluation for cultivated-land method and system based on GA-BP neural network models | |
| CN108510107A (en) | A kind of agricultural planting guidance method, electronic equipment and storage medium | |
| Yuan et al. | Regional land-use allocation using a coupled MAS and GA model: From local simulation to global optimization, a case study in Caidian District, Wuhan, China | |
| CN104809522B (en) | A kind of comprehensive energy prediction technique | |
| Zhou et al. | Socio-hydrological water balance for water allocation between human and environmental purposes in catchments | |
| CN107248007A (en) | The rural residential area intelligent optimization collocation method that a kind of double-deck domain is oriented to | |
| CN109360018A (en) | A kind of fuzzy zone land price estimation method based on artificial neural network | |
| Kannan et al. | Development of an automated procedure for estimation of the spatial variation of runoff in large river basins | |
| Petrushenko et al. | Innovations in Management of Fiscal Equalisation: Decentralization Reform | |
| CN103886519B (en) | Economic water consumption data spatial discretization method based on RBF neural network | |
| CN102360453A (en) | Horizontal arrangement method of protection forest | |
| Liu et al. | A quantitative analysis framework for water-food-energy nexus in an agricultural watershed using WEAP-MODFLOW |
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 |