CN108830418A - A kind of Short-Term Load Forecasting Method - Google Patents
A kind of Short-Term Load Forecasting Method Download PDFInfo
- Publication number
- CN108830418A CN108830418A CN201810613378.7A CN201810613378A CN108830418A CN 108830418 A CN108830418 A CN 108830418A CN 201810613378 A CN201810613378 A CN 201810613378A CN 108830418 A CN108830418 A CN 108830418A
- Authority
- CN
- China
- Prior art keywords
- value
- formula
- day
- bat
- quantify
- 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.)
- Pending
Links
- 238000013277 forecasting method Methods 0.000 title claims abstract description 10
- 230000001537 neural effect Effects 0.000 claims abstract description 26
- 238000005457 optimization Methods 0.000 claims abstract description 19
- 230000000694 effects Effects 0.000 claims abstract description 6
- 238000003062 neural network model Methods 0.000 claims abstract description 4
- 238000013139 quantization Methods 0.000 claims description 60
- 238000010606 normalization Methods 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000000034 method Methods 0.000 claims description 11
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 6
- 239000003595 mist Substances 0.000 claims description 4
- 239000004576 sand Substances 0.000 claims description 4
- 241000288673 Chiroptera Species 0.000 claims description 3
- 230000006978 adaptation Effects 0.000 claims description 3
- 239000012141 concentrate Substances 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 230000007340 echolocation Effects 0.000 claims description 3
- 230000003203 everyday effect Effects 0.000 claims description 3
- 210000003462 vein Anatomy 0.000 claims description 3
- 210000001367 artery Anatomy 0.000 claims 1
- 210000004218 nerve net Anatomy 0.000 claims 1
- 238000013528 artificial neural network Methods 0.000 abstract description 5
- 238000012098 association analyses Methods 0.000 abstract description 4
- 230000007547 defect Effects 0.000 abstract description 4
- 230000007812 deficiency Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000002354 daily effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Water Supply & Treatment (AREA)
- Physiology (AREA)
- Operations Research (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of Short-Term Load Forecasting Methods, its main feature is that, it includes choosing similar day collection, construct the RBF neural prediction model of bat algorithm optimization, using the RBF neural network model of bat algorithm optimization to prediction day load forecast, the present invention is on the basis of conventional gray association analysis, the similar day of higher similarity is chosen using distance similarity and the associated synthetical grey relation degree of shape proximity, conventional gray correlation fractal dimension is compensated for when selecting similar day, only consider the geometric similarity degree between data sequence and ignores the deficiency of numerical value degree of closeness, improve precision of prediction；Using the weight of bat algorithm optimization radial basis function (RBF) neural network, RBF neural can be overcome to be easily trapped into the defect of local optimum, speed whole network convergence rate, operation efficiency improves；With scientific and reasonable, the advantages that strong applicability, effect is good.
Description
Technical field
The present invention relates to load forecast fields, are a kind of Short-Term Load Forecasting Methods.
Background technique
Short-term electric load prediction is power grid production programming, the important component of traffic control, it is mainly to following several
When, one day even several days electric load is predicted, be power grid peace for arranging short term scheduling plan, reply emergency
Therefore the basis of full stabilization and economical operation is very necessary using advanced prediction technique.
Association analysis is the method for each correlate degree in a kind of analysis system of gray system theory proposition, basic
Thought is to judge correlation degree according to the similarity degree of data sequence, is usually used in load forecast field and finds and prediction day
Historical load day with similar Correlative Influence Factors.But the similar day that this method is found often only has good load
Trend similitude causes precision of prediction not high without having good load curve shape similarity.
Existing radial basis function (Radial Basis Function, RBF) neural network is a kind of partial approximation network,
Having many advantages, such as that learning time is short, calculation amount is small, network performance is excellent, pace of learning is thousands of times faster than usual BP learning algorithm,
It is widely used in load forecast field.But RBF neural is during prediction, between network hidden layer and output layer
The selection of connection weight be affected to precision of prediction, acquired frequently with gradient descent method, be easily trapped into locally optimal solution, ask
Solve the low defect of precision.
In view of the above-mentioned problems existing in the prior art, the present invention provides a kind of association of improved grey model and bat algorithm optimization diameter
To the Short-Term Load Forecasting Method of basis function neural network.This method uses on the basis of conventional gray association analysis
Shape similarity and the associated synthetical grey relation degree of closely located property choose the similar day of higher similarity, recycle similar
Day sample set carries out load forecast after carrying out the radial basis function neural network prediction model of training bat algorithm optimization, not only
Conventional gray correlation fractal dimension is compensated for when selecting similar day, the geometric similarity degree between data sequence is only considered and ignores
The deficiency of numerical value degree of closeness, and RBF neural can be overcome to be easily trapped into the defect of local optimum, receive whole network
Speed, operation efficiency and precision of prediction is held back to be improved.
Summary of the invention
The technical problem to be solved by the present invention is to:A kind of scientific and reasonable, strong applicability, the good short term power of effect are provided
Load forecasting method is puted forth effort to solve the problems, such as that load forecast precision is low and RBF neural is easily trapped into part most
Problem excellent, solving precision is low.
Solve its technical problem the technical solution adopted is that：A kind of Short-Term Load Forecasting Method, characterized in that it is wrapped
Include following steps：
S1:Choose similar day collection
S1.1:Choose similar day rough set
The influence factor of temperature, weather condition, date type as similar day rough set is chosen, is selected 60 days a few days ago to be predicted
Range of each 30 day data as data sample before and after data, preceding L same date, L indicates the time limit of sample data, depending on being
The possessed data volume of system, value range 2-6；
S1.2:Quantization influence factor
It because the order of magnitude and unit of temperature, weather condition and date type are different from, cannot further calculate, need
Prediction day and each influence factor of similar day rough set are quantified,
S1.2.1:Quantization for temperature：When the highest temperature >=40 DEG C, quantization value is 2.0；33~39 DEG C of the highest temperature
When, quantization value is 1.9；At 26~32 DEG C of the highest temperature, quantization value is 1.8；At 19~25 DEG C of the highest temperature, quantify value
It is 1.7；At 12~18 DEG C of the highest temperature, quantization value is 1.6；At 5~11 DEG C of the highest temperature, quantization value is 1.5；Highest gas
At -2~4 DEG C of temperature, quantization value is 1.4；When the highest temperature≤- 3 DEG C, quantization value is 1.3；When the lowest temperature≤- 25 DEG C, amount
Changing value is 2.0；When the lowest temperature -24~-18 DEG C, quantization value is 1.9；When the lowest temperature -17~-11 DEG C, quantify value
It is 1.8；When the lowest temperature -10~-4 DEG C, quantization value is 1.7；When the lowest temperature -3~3 DEG C, quantization value is 1.6；It is minimum
At 4~10 DEG C of temperature, quantization value is 1.5；At 11~17 DEG C of the lowest temperature, quantization value is 1.4；The lowest temperature >=18 DEG C
When, quantization value is 1.3；When temperature on average≤- 20 DEG C, quantization value is 2.0；When temperature on average -19~-12 DEG C, quantization is taken
Value is 1.9；When temperature on average -11~-4 DEG C, quantization value is 1.8；When temperature on average -3~4 DEG C, quantization value is 1.7；It is flat
At equal 5~12 DEG C of temperature, quantization value is 1.6；At 13~20 DEG C of temperature on average, quantization value is 1.5；Temperature on average 21~28
DEG C when, quantization value be 1.4；When temperature on average >=29 DEG C, quantization value is 1.3；
Quantization of the S1.2.2 for weather condition：When fine, quantify value 0.1；When cloudy, quantify value 0.2；When negative, amount
Change value 0.3；When mist, quantify value 0.4；When haze, quantify value 0.5；When sand, quantify value 0.6；When light rain, quantization is taken
Value 0.7；When shower, quantify value 0.8；When thunder shower, quantify value 0.9；When drizzle or moderate rain, quantify value 1.0；When moderate rain,
Quantify value 1.1；When moderate rain or heavy rain, quantify value 1.2；When raining heavyly, quantify value 1.3；When heavy or torrential rain, quantify value 1.4；
When heavy rain, quantify value 1.5；When rain and snow mixed, quantify value 1.6；When slight snow, quantify value 1.7；When light to moderate snow, quantization is taken
Value 1.8；When moderate snow, quantify value 1.9；When moderate or heavy snow, quantify value 2.0；When heavy snow, quantify value 2.1；Heavy to torrential snow
When, quantify value 2.2；When severe snow, quantify value 2.3；
Quantization of the S1.2.3 for date type：When Monday, quantify value 0.35；When Tuesday, quantify value 0.2；Wednesday
When, quantify value 0.15；When Thursday, quantify value 0.2；When Friday, quantify value 0.35；When Saturday, quantify value 0.6；Week
When day, quantify value 0.7；When festivals or holidays, quantify value 1.0,
S1.3:Similar day collection is chosen using improved grey model correlating method
S1.3.1:Construction feature matrix
By each influence factor value composition characteristic vector X after step S1.2 quantization, wherein prediction day character vector X0=
[x0(1), x0(2) ..., x0(m)] it indicates, m is the dimension of feature vector, every day character vector X in similar day rough set1,
X2..., XnIt indicates, n is the number of sample in similar day rough set, the eigenmatrix of this n+1 Sequence composition one m × (n+1) is such as
Shown in formula (1),
In formula：x0It (k) is k-th of feature of prediction day, xiIt (k) is k-th of feature of i-th of sample in similar day rough set；
S1.3.2:Calculate shape similarity grey relational grade
The difference of each component of sample sequence in prediction day sequence and similar day rough set is calculated, matrix of differences, such as formula are formed
(2) shown in：
In formula：For k-th of feature of i-th of sample in the value and similar day rough set of k-th of feature of prediction day
Value difference, willIt introduces formula (3) and calculates its shape similarity grey relational grade：
In formula：γ1(x0(k), xiIt (k)) is the kth of i-th of sample in k-th of the feature and similar day rough set of prediction day
The shape similarity grey relational grade of a feature；
S1.3.3:Calculate closely located property grey relational grade
The quotient of each component of sample sequence in prediction day sequence and similar day rough set is calculated, quotient's matrix is formed, such as formula (4) institute
Show：
In formula：K-th for i-th of sample in the value and similar day rough set of k-th of feature of prediction day is special
The quotient of the value of sign, willIt introduces formula (5) and calculates its closely located property grey relational grade：
In formula：γ2(x0(k), xiIt (k)) is the kth of i-th of sample in k-th of the feature and similar day rough set of prediction day
The closely located property grey relational grade of a feature；
S1.3.4:Calculate synthetical grey relation degree
The synthetical grey relation degree for calculating each history day and prediction day in similar day rough set, as shown in formula (6)：
In formula：For the synthetical grey relation degree of i-th of sample in prediction day and similar day rough set；
S1.3.5:Choose similar day
All similar day rough set samples of the synthetical grey relation degree greater than 0.85 in step S1.3.4 are selected to form similar day
Sample set,
S2:Construct the RBF neural prediction model of bat algorithm optimization
S2.1:Data prediction
S2.1.1:Confirm the input quantity and output quantity of load forecasting model
The input vector includes the load data at t-1, t, t+1 moment and the loading effects of similar day on the day before similar day
Factor, the output vector are the load data of similar day t moment,
S2.1.2:Input vector obtained by step S2.1.1 and output vector are normalized by formula (7), formula (8)
Processing obtains normalization input vector and normalized output vector, wherein normalization formula is：
In formula：M is input layer number, and N is output layer number of nodes, dτAnd yτRespectively it is originally inputted before normalized
τ-component in the original output vector of vector sum, dτ, maxAnd dτ, minτ points are respectively originally inputted in vector before normalized
The maximum value and minimum value of amount, yτ, maxAnd yτ, minRespectively before normalized in original output vector τ-component maximum value
And minimum value,Withτ-component in the original output vector of vector sum is originally inputted respectively after normalized；
S2.2:Initialize RBF network model
Using after the obtained normalization of step S2.1.2 input vector and output vector as training sample building RBF mind
Through Network Prediction Model, RBF neural is three-decker, including input layer, hidden layer and output layer, wherein hidden layer node
Number is determined according to the empirical equation (9) of Hecht-Nielsen：
In formula：NhFor node in hidden layer, M is input layer number, and N is output layer number of nodes, formula (9) resulting value to
Upper round numbers,
In RBF neural prediction model, the output of each neuron of hidden layer is：
In formula：For the training sample of input, B is number of samples；| | | | it is European norm；εa
The center vector for indicating a-th of neuronal kernel function of hidden layer, is determined by K-means algorithm；σaFor a-th of neuron of hidden layer
Extension constant；G () takes Gaussian function, i.e.,：
S2.3:The weight of bat algorithm optimization RBF model
S2.3.1:Bat population is initialized, initializes the bat individual amount S of entire population, every bat individual is most
Big pulse loudness of a sound A0With maximum impulse emissivity R0；The c bat is generated at random in the position at the t ' momentSpeedArteries and veins
Rush loudness of a soundWith impulse ejection rateC=1,2 ..., S；Initialize the maximum frequency f of bat population echolocationmaxAnd minimum
Frequency fmin, pulse loudness of a sound attenuation coefficient α, impulse ejection rate increase coefficient μ, entire population maximum number of iterations Z, current iteration
Number z, search precision θ；
S2.3.2:Current iteration number is recorded, calculates the c bat ideal adaptation angle value Fitness according to formula (12)
(c)：
In formula：B is that training sample concentrates number of samples, and N is output layer number of nodes；Q-th for c-th of sample is defeated
The predicted value of egress,For the corresponding actual value of q-th of output node of c-th of sample；Similarly found out using formula (12)
The fitness of remaining S-1 bat, then the current adaptive optimal control of the smallest conduct of fitness value is found out in all fitness values
Spend Fbest, corresponding position is current optimal solution, is labeled as Pbest；
S2.3.3:Judge whether current iteration number z reaches maximum number of iterations Z or current adaptive optimal control degree is less than search
Precision θ enters step S2.3.8 if meeting above-mentioned termination condition；Otherwise S2.3.4 is entered step,
S2.3.4:Speed and the position of every bat are updated according to formula (13), formula (14), formula (15), and more
Position after new calculates the fitness of new explanation according to formula (12), updates the history optimal solution of every bat as current location
With the globally optimal solution of bat population,
fc=fmin+(fmax-fmin)×β (13)
In formula：fcIt is the Echolocation frequency of the c bat, and fc∈[fmin, fmax]；β is generally evenly distributed on [0,1]
Random number；Bat c is respectively indicated in the speed at+1 moment of t ' moment and t ',Respectively indicate bat
C is in the position at+1 moment of t ' moment and t '；
S2.3.5：Generate a several rand1 at random in [0,1] range, ifThen own currently
Select an individual for global optimum body position P in individualbest, using formula (16) in PbestAn office is nearby randomly generated
Portion's individual calculates the fitness value F of this part individualnew,
In formula：θ is the random number on [0,1], PnewIndicate the new position of bat, PoldIndicate the old position of bat,Table
Show the average value of the pulse loudness of a sound of all bats in current bat population；
S2.3.6:Generate a several rand2 at random in [0,1] range, ifAnd Fnew< Fbest, then
Receiving new explanation is current optimal global individual, updates bat history optimal solution and globally optimal solution, saves its fitness value, utilizes
Formula (17), formula (18) reduce pulse loudness of a sound and increase impulse ejection rate,
In formula, α is pulse loudness of a sound attenuation coefficient, and μ is that impulse ejection rate increases coefficient；Respectively indicate bat c
In the pulse loudness of a sound at+1 moment of t ' moment and t '；For the maximum impulse emissivity of the c bat,For bat c t '+
The impulse ejection rate at 1 moment；
S2.3.7:Update the number of iterations z=z+1, return step S2.3.3；
S2.3.8:Using bat algorithm globally optimal solution as the weight matrix W of RBF neural hidden layer to output layer,
The RBF neural prediction model of bat algorithm optimization is constructed,
S3:Using the RBF neural network model of bat algorithm optimization to prediction day load forecast
By the electric load influence factor and the load data at the previous day t-1, t, t+1 moment prediction day of predicting day by formula
(7) as prediction input vector after normalizing, which is input in trained RBF neural, through public affairs
Formula (19) obtains the output valve of the load forecast of prediction day t moment,
In formula：waτFor the connection weight of hidden layer to output layer, haFor the output of each neuron of hidden layer, a=1 ...,
Nh,NhFor node in hidden layer, N is output layer number of nodes；For the τ output node of corresponding with input sample network
Real output value,
Obtained output valve is obtained into the load forecast of prediction day t moment through formula (20) anti-normalization processing again
Value,
In formula：Electric Load Forecasting measured value after indicating anti-normalization processing, yτ, maxAnd yτ, minRespectively normalized
The maximum value and minimum value of τ-component in preceding original output vector.
The value that the L depends on the data volume that system is possessed is 3.
The beneficial effects of the present invention are embodied in：
(1) present invention is associated using distance similarity and shape proximity on the basis of conventional gray association analysis
Synthetical grey relation degree choose the similar day of higher similarity, compensate for conventional gray correlation fractal dimension in selection similar day
When, only consider the geometric similarity degree between data sequence and ignore the deficiency of numerical value degree of closeness, improves precision of prediction；
(2) present invention utilizes the weight of bat algorithm optimization radial basis function (RBF) neural network, and RBF can be overcome refreshing
It is easily trapped into the defect of local optimum through network, speeds whole network convergence rate, operation efficiency improves；
(3) scientific and reasonable, strong applicability of the invention, effect are good.
Detailed description of the invention
Fig. 1 is that similar day collection chooses flow chart；
Fig. 2 is the flow chart of bat algorithm optimization RBF model weight.
Specific embodiment
The technical issues of solved in order to further illustrate the present invention, carries out the present invention below in conjunction with attached drawing and table detailed
Explanation.
A kind of Short-Term Load Forecasting Method of the invention, includes the following steps：
S1:Similar day collection is chosen, is described in detail in conjunction with Fig. 1 method chosen to similar day collection of the present invention,
S1.1:Choose similar day rough set
The influence factor of temperature, weather condition, date type as similar day rough set is chosen, is selected 60 days a few days ago to be predicted
Range of each 30 day data as data sample before and after data, preceding L same date, L chooses the time limit of sample data, depending on being
The possessed data volume of system, value range 2-6 generally take 3 to be advisable；
S1.2:Quantization influence factor
It because the order of magnitude and unit of temperature, weather condition and date type are different from, cannot further calculate, need
Following quantification treatment is carried out to prediction day and each influence factor of similar day rough set：
S1.2.1:The quantization of temperature
Daily maximum temperature, the lowest temperature and mean temperature are quantified, quantized value is shown in Table 1；
Table 1 is temperature quantized value：
S1.2.2:The quantization of weather condition
Weather condition be broadly divided into fine, cloudy, negative, mist, haze, sand, light rain, shower, thunder shower, drizzle or moderate rain, moderate rain,
Moderate rain or heavy rain, heavy rain, heavy or torrential rain, heavy rain, rain and snow mixed, slight snow, light to moderate snow, moderate snow, moderate or heavy snow, heavy snow, heavy to torrential snow
And severe snow, quantized value are shown in Table 2；
Table 2 is weather condition quantized value：
| Weather situation | Quantify value | Weather situation | Quantify value |
| It is fine | 0.1 | Heavy rain | 1.3 |
| It is cloudy | 0.2 | Heavy or torrential rain | 1.4 |
| Yin | 0.3 | Heavy rain | 1.5 |
| Mist | 0.4 | Rain and snow mixed | 1.6 |
| Haze | 0.5 | Slight snow | 1.7 |
| Sand | 0.6 | Light to moderate snow | 1.8 |
| Light rain | 0.7 | Moderate snow | 1.9 |
| Shower | 0.8 | Moderate or heavy snow | 2.0 |
| Thunder shower | 0.9 | Heavy snow | 2.1 |
| Drizzle or moderate rain | 1.0 | Heavy to torrential snow | 2.2 |
| Moderate rain | 1.1 | Severe snow | 2.3 |
| Moderate rain or heavy rain | 1.2 |
S1.2.3:The quantization of date type
Date type can be divided into working day (Mon-Fri), day off (Saturday to Sunday) and festivals or holidays, quantized value and see
Table 3,
Table 3 is date type quantized value：
| Date type | Quantify value |
| Monday | 0.35 |
| Tuesday | 0.2 |
| Wednesday | 0.15 |
| Thursday | 0.2 |
| Friday | 0.35 |
| Saturday | 0.6 |
| Sunday | 0.7 |
| Festivals or holidays | 1.0 |
S1.3:Similar day collection is chosen using improved grey model correlating method
The present invention provides a kind of method for merging poor mode and the improved grey relational analysis except mode, from shape and distance
Upper definition synthetical grey relation degree, specific step is as follows：
S1.3.1:Construction feature matrix
By each influence factor value composition characteristic vector X after step S1.2 quantization, wherein prediction day character vector X0=
[x0(1), x0(2) ..., x0(m)] it indicates, m is the dimension of feature vector, every day character vector X in similar day rough set1,
X2..., XnIt indicates, n is the number of sample in similar day rough set, the eigenmatrix of this n+1 Sequence composition one m × (n+1) is such as
Shown in formula (1),
In formula：x0It (k) is k-th of feature of prediction day, xiIt (k) is k-th of feature of i-th of sample in similar day rough set；
S1.3.2:Calculate shape similarity grey relational grade
The difference of each component of sample sequence in prediction day sequence and similar day rough set is calculated, matrix of differences, such as formula are formed
(2) shown in：
In formula：For k-th of feature of i-th of sample in the value and similar day rough set of k-th of feature of prediction day
Value difference, willIt introduces formula (3) and calculates its shape similarity grey relational grade：
In formula：γ1(x0(k), xiIt (k)) is the kth of i-th of sample in k-th of the feature and similar day rough set of prediction day
The shape similarity grey relational grade of a feature；
S1.3.3:Calculate closely located property grey relational grade
The quotient of each component of sample sequence in prediction day sequence and similar day rough set is calculated, quotient's matrix is formed, such as formula (4) institute
Show：
In formula：K-th for i-th of sample in the value and similar day rough set of k-th of feature of prediction day is special
The quotient of the value of sign, willIt introduces formula (5) and calculates its closely located property grey relational grade：
In formula：γ2(x0(k), xiIt (k)) is the kth of i-th of sample in k-th of the feature and similar day rough set of prediction day
The closely located property grey relational grade of a feature；
S1.3.4:Calculate synthetical grey relation degree
The synthetical grey relation degree for calculating each history day and prediction day in similar day rough set, as shown in formula (6)：
In formula：For the synthetical grey relation degree of i-th of sample in prediction day and similar day rough set；
S1.3.5:Choose similar day
All similar day rough set samples of the synthetical grey relation degree greater than 0.85 in step S1.3.4 are selected to form similar day
Sample set,
S2:Construct the RBF neural prediction model of bat algorithm optimization
S2.1:Data prediction
S2.1.1:Confirm the input quantity and output quantity of load forecasting model
The input vector includes the load data at t-1, t, t+1 moment and the loading effects of similar day on the day before similar day
Factor, the output vector are the load data of similar day t moment,
S2.1.2:Input vector obtained by step S2.1.1 and output vector are normalized by formula (7), formula (8)
Processing obtains normalization input vector and normalized output vector, wherein normalization formula is：
In formula：M is input layer number, and N is output layer number of nodes, dτAnd yτRespectively it is originally inputted before normalized
τ-component in the original output vector of vector sum, dτ, maxAnd dτ, minτ points are respectively originally inputted in vector before normalized
The maximum value and minimum value of amount, yτ, maxAnd yτ, minRespectively before normalized in original output vector τ-component maximum value
And minimum value,Withτ-component in the original output vector of vector sum is originally inputted respectively after normalized；
S2.2:Initialize RBF network model
Using after the obtained normalization of step S2.1.2 input vector and output vector as training sample building RBF mind
Through Network Prediction Model, RBF neural is three-decker, including input layer, hidden layer and output layer, wherein hidden layer node
Number is determined according to the empirical equation (9) of Hecht-Nielsen：
In formula：Input layer number M is 8, and output layer number of nodes N is 1, and formula (9) resulting value rounds up number, therefore by
Formula (9) determines node in hidden layer NhIt is 17,
In RBF neural prediction model, the output of each neuron of hidden layer is：
In formula：For the training sample of input, B is number of samples；| | | | it is European norm；εa
The center vector for indicating the α neuronal kernel function of hidden layer, is determined by K-means algorithm；σaFor the α neuron of hidden layer
Extension constant；G () takes Gaussian function, i.e.,：
S2.3:It is described in detail in conjunction with weight method of the Fig. 2 to bat algorithm optimization RBF model of the invention,
S2.3.1:Bat population is initialized, initializes the bat individual amount S of entire population, every bat individual is most
Big pulse loudness of a sound A0With maximum impulse emissivity R0, the c bat is generated at random in the position at the t ' momentSpeedArteries and veins
Rush loudness of a soundWith impulse ejection rateC=1,2 ..., S；Initialize the maximum frequency f of bat population echolocationmaxIt is 1, most
Small frequency fminIt is 0；Pulse loudness of a sound attenuation coefficient α is 0.9, and it is 0.9 that impulse ejection rate, which increases coefficient μ, entire population greatest iteration
Number Z is 5000；Current iteration number z, z is 1 when primary iteration；Search precision θ is 0.001；
S2.3.2:Current iteration number is recorded, calculates the c bat ideal adaptation angle value Fitness according to formula (12)
(c)：
In formula：B is that training sample concentrates number of samples, and N is output layer number of nodes；Q-th for c-th of sample is defeated
The predicted value of egress,For the corresponding actual value of q-th of output node of c-th of sample；Similarly found out using formula (12)
The fitness of remaining S-1 bat, then the current adaptive optimal control of the smallest conduct of fitness value is found out in all fitness values
Spend Fbest, corresponding position is current optimal solution, is labeled as Pbest；
S2.3.3:Judge whether current iteration number z reaches maximum number of iterations Z or current adaptive optimal control degree is less than search
Precision θ enters step S2.3.8 if meeting above-mentioned termination condition；Otherwise S2.3.4 is entered step,
S2.3.4:Speed and the position of every bat are updated according to formula (13), formula (14), formula (15), and more
Position after new calculates the fitness of new explanation according to formula (12), updates the history optimal solution of every bat as current location
With the globally optimal solution of bat population,
fc=fmin+(fmax-fmin)×β (13)
In formula：fcIt is the Echolocation frequency of the c bat, and fc∈[fmin, fmax]；β is generally evenly distributed on [0,1]
Random number；Bat c is respectively indicated in the speed at+1 moment of t ' moment and t ',Respectively indicate bat
C is in the position at+1 moment of t ' moment and t '；
S2.3.5：Generate a several rand1 at random in [0,1] range, ifThen own currently
Select an individual for global optimum body position P in individualbest, using formula (16) in PbestAn office is nearby randomly generated
Portion's individual calculates the fitness value F of this part individualnew,
In formula：θ is the random number on [0,1], PnewIndicate the new position of bat, PoldIndicate the old position of bat,Table
Show the average value of the pulse loudness of a sound of all bats in current bat population；
S2.3.6:Generate a several rand2 at random in [0,1] range, ifAnd Fnew< Fbest, then
Receiving new explanation is current optimal global individual, updates bat history optimal solution and globally optimal solution, saves its fitness value, utilizes
Formula (17), formula (18) reduce pulse loudness of a sound and increase impulse ejection rate,
In formula, α is pulse loudness of a sound attenuation coefficient, and μ is that impulse ejection rate increases coefficient；Respectively indicate bat c
In the pulse loudness of a sound at+1 moment of t ' moment and t '；For the maximum impulse emissivity of the c bat,For bat c t '+
The impulse ejection rate at 1 moment；
S2.3.7:Update the number of iterations z=z+1, return step S2.3.3；
S2.3.8:Using bat algorithm globally optimal solution as the weight matrix W of RBF neural hidden layer to output layer,
The RBF neural prediction model of bat algorithm optimization is constructed,
S3:Using the RBF neural network model of bat algorithm optimization to prediction day load forecast
By the electric load influence factor and the load data at the previous day t-1, t, t+1 moment prediction day of predicting day by formula
(7) as prediction input vector after normalizing, which is input in trained RBF neural, through public affairs
Formula (19) obtains the output valve of the load forecast of prediction day t moment,
In formula：waτFor the connection weight of hidden layer to output layer, haFor the output of each neuron of hidden layer, a=1 ...,
Nh,NhFor node in hidden layer, N is output layer number of nodes；For the τ output node of corresponding with input sample network
Real output value,
Obtained output valve is obtained into the load forecast of prediction day t moment through formula (20) anti-normalization processing again
Value,
In formula：Electric Load Forecasting measured value after indicating anti-normalization processing, yτ, maxAnd yτ, minRespectively normalized
The maximum value and minimum value of τ-component in preceding original output vector.
The embodiment of the present invention is only used for that the present invention is further illustrated, not exhaustive, does not constitute to claim
The restriction of protection scope, the enlightenment that those skilled in the art obtain according to embodiments of the present invention, without creative work energy
Enough expect other substantially equivalent substitutions, all falls in the scope of protection of the present invention.
Claims (2)
1. a kind of Short-Term Load Forecasting Method, characterized in that it includes the following steps：
S1:Choose similar day collection
S1.1:Choose similar day rough set
The influence factor of temperature, weather condition, date type as similar day rough set is chosen, 60 number of days a few days ago to be predicted is selected
Range according to 30 day data each before and after, preceding L same date as data sample, L indicate the time limit of sample data, depending on system
The data volume possessed, value range 2-6；
S1.2:Quantization influence factor
It because the order of magnitude and unit of temperature, weather condition and date type are different from, cannot further calculate, need pair
Prediction day and each influence factor of similar day rough set are quantified,
S1.2.1:Quantization for temperature：When the highest temperature >=40 DEG C, quantization value is 2.0；At 33~39 DEG C of the highest temperature,
Quantifying value is 1.9；At 26~32 DEG C of the highest temperature, quantization value is 1.8；At 19~25 DEG C of the highest temperature, quantization value is
1.7；At 12~18 DEG C of the highest temperature, quantization value is 1.6；At 5~11 DEG C of the highest temperature, quantization value is 1.5；Highest gas
At -2~4 DEG C of temperature, quantization value is 1.4；When the highest temperature≤- 3 DEG C, quantization value is 1.3；When the lowest temperature≤- 25 DEG C, amount
Changing value is 2.0；When the lowest temperature -24~-18 DEG C, quantization value is 1.9；When the lowest temperature -17~-11 DEG C, quantify value
It is 1.8；When the lowest temperature -10~-4 DEG C, quantization value is 1.7；When the lowest temperature -3~3 DEG C, quantization value is 1.6；It is minimum
At 4~10 DEG C of temperature, quantization value is 1.5；At 11~17 DEG C of the lowest temperature, quantization value is 1.4；The lowest temperature >=18 DEG C
When, quantization value is 1.3；When temperature on average≤- 20 DEG C, quantization value is 2.0；When temperature on average -19~-12 DEG C, quantization is taken
Value is 1.9；When temperature on average -11~-4 DEG C, quantization value is 1.8；When temperature on average -3~4 DEG C, quantization value is 1.7；It is flat
At equal 5~12 DEG C of temperature, quantization value is 1.6；At 13~20 DEG C of temperature on average, quantization value is 1.5；Temperature on average 21~28
DEG C when, quantization value be 1.4；When temperature on average >=29 DEG C, quantization value is 1.3；
Quantization of the S1.2.2 for weather condition：When fine, quantify value 0.1；When cloudy, quantify value 0.2；When negative, quantization is taken
Value 0.3；When mist, quantify value 0.4；When haze, quantify value 0.5；When sand, quantify value 0.6；When light rain, quantify value
0.7；When shower, quantify value 0.8；When thunder shower, quantify value 0.9；When drizzle or moderate rain, quantify value 1.0；When moderate rain, amount
Change value 1.1；When moderate rain or heavy rain, quantify value 1.2；When raining heavyly, quantify value 1.3；When heavy or torrential rain, quantify value 1.4；Cruelly
When rain, quantify value 1.5；When rain and snow mixed, quantify value 1.6；When slight snow, quantify value 1.7；When light to moderate snow, quantify value
1.8；When moderate snow, quantify value 1.9；When moderate or heavy snow, quantify value 2.0；When heavy snow, quantify value 2.1；When heavy to torrential snow,
Quantify value 2.2；When severe snow, quantify value 2.3；
Quantization of the S1.2.3 for date type：When Monday, quantify value 0.35；When Tuesday, quantify value 0.2；When Wednesday, amount
Change value 0.15；When Thursday, quantify value 0.2；When Friday, quantify value 0.35；When Saturday, quantify value 0.6；When Sunday,
Quantify value 0.7；When festivals or holidays, quantify value 1.0,
S1.3:Similar day collection is chosen using improved grey model correlating method
S1.3.1:Construction feature matrix
By each influence factor value composition characteristic vector X after step S1.2 quantization, wherein prediction day character vector X0=[x0
(1),x0(2),…,x0(m)] it indicates, m is the dimension of feature vector, every day character vector X in similar day rough set1,
X2..., XnIt indicates, n is the number of sample in similar day rough set, the eigenmatrix of this n+1 Sequence composition one m × (n+1) is such as
Shown in formula (1),
In formula：x0It (k) is k-th of feature of prediction day, xiIt (k) is k-th of feature of i-th of sample in similar day rough set；
S1.3.2:Calculate shape similarity grey relational grade
The difference of each component of sample sequence in prediction day sequence and similar day rough set is calculated, matrix of differences is formed, such as formula (2) institute
Show：
In formula：For the value of k-th of feature of i-th of sample in the value and similar day rough set of k-th of feature of prediction day
Difference, willIt introduces formula (3) and calculates its shape similarity grey relational grade：
In formula：γ1(x0(k),xi(k)) k-th for i-th of sample in k-th of the feature and similar day rough set of prediction day is special
The shape similarity grey relational grade of sign；
S1.3.3:Calculate closely located property grey relational grade
The quotient of each component of sample sequence in prediction day sequence and similar day rough set is calculated, quotient's matrix is formed, as shown in formula (4)：
In formula：For the value of k-th of feature of i-th of sample in the value and similar day rough set of k-th of feature of prediction day
Quotient, willIt introduces formula (5) and calculates its closely located property grey relational grade：
In formula：γ2(x0(k),xi(k)) k-th for i-th of sample in k-th of the feature and similar day rough set of prediction day is special
The closely located property grey relational grade of sign；
S1.3.4:Calculate synthetical grey relation degree
The synthetical grey relation degree for calculating each history day and prediction day in similar day rough set, as shown in formula (6)：
In formula：For the synthetical grey relation degree of i-th of sample in prediction day and similar day rough set；
S1.3.5:Choose similar day
All similar day rough set samples of the synthetical grey relation degree greater than 0.85 in step S1.3.4 are selected to form similar day sample
Collection,
S2:Construct the RBF neural prediction model of bat algorithm optimization
S2.1:Data prediction
S2.1.1:Confirm the input quantity and output quantity of load forecasting model
The input vector include the load data and similar day at t-1, t, t+1 moment on the day before similar day loading effects because
Element, the output vector are the load data of similar day t moment,
S2.1.2:Input vector obtained by step S2.1.1 and output vector are normalized by formula (7), formula (8),
Normalization input vector and normalized output vector are obtained, wherein normalization formula is：
In formula：M is input layer number, and N is output layer number of nodes, dτAnd yτRespectively vector is originally inputted before normalized
With τ-component, d in original output vectorτ,maxAnd dτ,minτ-component in vector is respectively originally inputted before normalized
Maximum value and minimum value, yτ,maxAnd yτ,minThe maximum value of τ-component and most in original output vector respectively before normalized
Small value,Withτ-component in the original output vector of vector sum is originally inputted respectively after normalized；
S2.2:Initialize RBF network model
Using after the obtained normalization of step S2.1.2 input vector and output vector as training sample construct RBF nerve net
Network prediction model, RBF neural are three-decker, including input layer, hidden layer and output layer, and wherein hidden layer node is several
It is determined according to the empirical equation (9) of Hecht-Nielsen：
In formula：NhFor node in hidden layer, M is input layer number, and N is output layer number of nodes, and formula (9) resulting value takes upwards
Integer,
In RBF neural prediction model, the output of each neuron of hidden layer is：
In formula：For the training sample of input, B is number of samples；‖ ‖ is European norm；εaIndicate hidden
The center vector of a-th of the neuronal kernel function containing layer, is determined by K-means algorithm；σaFor the extension of a-th of neuron of hidden layer
Constant；G () takes Gaussian function, i.e.,：
S2.3:The weight of bat algorithm optimization RBF model
S2.3.1:Bat population is initialized, the bat individual amount S of entire population, the maximum arteries and veins of every bat individual are initialized
Rush loudness of a sound A0With maximum impulse emissivity R0；The c bat is generated at random in the position at the t ' momentSpeedPulse sound
By forceWith impulse ejection rateC=1,2, S；Initialize the maximum frequency f of bat population echolocationmaxAnd minimum
Frequency fmin, pulse loudness of a sound attenuation coefficient α, impulse ejection rate increase coefficient μ, entire population maximum number of iterations Z, current iteration
Number z, search precision θ；
S2.3.2:Current iteration number is recorded, calculates the c bat ideal adaptation angle value Fitness (c) according to formula (12)：
In formula：B is that training sample concentrates number of samples, and N is output layer number of nodes；It is saved for q-th of output of c-th of sample
The predicted value of point,For the corresponding actual value of q-th of output node of c-th of sample；It is similarly found out using formula (12) remaining
The fitness of S-1 bat, then the current adaptive optimal control degree of the smallest conduct of fitness value is found out in all fitness values
Fbest, corresponding position is current optimal solution, is labeled as Pbest；
S2.3.3:Judge whether current iteration number z reaches maximum number of iterations Z or current adaptive optimal control degree is less than search precision
θ enters step S2.3.8 if meeting above-mentioned termination condition；Otherwise S2.3.4 is entered step,
S2.3.4:Speed and the position of every bat are updated according to formula (13), formula (14), formula (15), and after updating
Position as current location, the fitness of new explanation is calculated according to formula (12), updates history optimal solution and the bat of every bat
The globally optimal solution of bat population,
fc=fmin+(fmax-fmin)×β (13)
In formula：fcIt is the Echolocation frequency of the c bat, and fc∈[fmin,fmax]；β be generally evenly distributed on [0,1] with
Machine number；Bat c is respectively indicated in the speed at+1 moment of t ' moment and t ',Bat c is respectively indicated to exist
The position at+1 moment of t ' moment and t '；
S2.3.5：Generate a several rand1 at random in [0,1] range, ifThen in current all individuals
Select an individual for global optimum body position Pbest, using formula (16) in PbestA part individual is nearby randomly generated,
Calculate the fitness value F of this part individualnew,
In formula：θ is the random number on [0,1], PnewIndicate the new position of bat, PoldIndicate the old position of bat,Expression is worked as
The average value of the pulse loudness of a sound of all bats in preceding bat population；
S2.3.6:Generate a several rand2 at random in [0,1] range, ifAnd Fnew<Fbest, then receive new
Solution is current optimal global individual, updates bat history optimal solution and globally optimal solution, saves its fitness value, utilize formula
(17), formula (18) reduces pulse loudness of a sound and increases impulse ejection rate,
In formula, α is pulse loudness of a sound attenuation coefficient, and μ is that impulse ejection rate increases coefficient；Bat c is respectively indicated in t '
The pulse loudness of a sound at+1 moment of moment and t '；For the maximum impulse emissivity of the c bat,It is bat c at+1 moment of t '
Impulse ejection rate；
S2.3.7:Update the number of iterations z=z+1, return step S2.3.3；
S2.3.8:Using bat algorithm globally optimal solution as the weight matrix W of RBF neural hidden layer to output layer, building
The RBF neural prediction model of bat algorithm optimization,
S3:Using the RBF neural network model of bat algorithm optimization to prediction day load forecast
By the electric load influence factor and the load data at the previous day t-1, t, t+1 moment prediction day of predicting day by formula (7)
As prediction input vector after normalization, which is input in trained RBF neural, through formula
(19) output valve of the load forecast of prediction day t moment is obtained,
In formula：waτFor the connection weight of hidden layer to output layer, haFor the output of each neuron of hidden layer, a=1,
Nh,NhFor node in hidden layer, N is output layer number of nodes；For the τ output node of corresponding with input sample network
Real output value,
Obtained output valve is obtained into the Electric Load Forecasting measured value of prediction day t moment through formula (20) anti-normalization processing again,
In formula：Electric Load Forecasting measured value after indicating anti-normalization processing, yτ, maxAnd yτ, minIt is former respectively before normalized
The maximum value and minimum value of τ-component in beginning output vector.
2. a kind of Short-Term Load Forecasting Method according to claim 1, characterized in that the L depends on system institute
The value of the data volume possessed is 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810613378.7A CN108830418A (en) | 2018-06-14 | 2018-06-14 | A kind of Short-Term Load Forecasting Method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810613378.7A CN108830418A (en) | 2018-06-14 | 2018-06-14 | A kind of Short-Term Load Forecasting Method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108830418A true CN108830418A (en) | 2018-11-16 |
Family
ID=64141774
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810613378.7A Pending CN108830418A (en) | 2018-06-14 | 2018-06-14 | A kind of Short-Term Load Forecasting Method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108830418A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109376971A (en) * | 2018-12-29 | 2019-02-22 | 北京中电普华信息技术有限公司 | A kind of load curve forecasting method and system towards power consumer |
| CN109887035A (en) * | 2018-12-27 | 2019-06-14 | 哈尔滨理工大学 | Based on bat algorithm optimization BP neural network binocular vision calibration |
| CN110309988A (en) * | 2019-07-12 | 2019-10-08 | 广东电网有限责任公司 | A kind of Methods of electric load forecasting based on grey relational grade and support vector machines |
| CN110728401A (en) * | 2019-10-10 | 2020-01-24 | 郑州轻工业学院 | Short-term power load prediction method of neural network based on squirrel and weed hybrid algorithm |
| CN110967471A (en) * | 2019-11-14 | 2020-04-07 | 广东电网有限责任公司 | Method for predicting concentration of dissolved gas in transformer oil |
| CN111340273A (en) * | 2020-02-17 | 2020-06-26 | 南京邮电大学 | Short-term load prediction method for power system based on GEP parameter optimization XGboost |
| CN111612227A (en) * | 2020-05-12 | 2020-09-01 | 国网河北省电力有限公司电力科学研究院 | Load prediction method based on K-means clustering and bat optimization neural network |
| CN112052913A (en) * | 2020-09-27 | 2020-12-08 | 国网江苏省电力有限公司南京供电分公司 | Distributed photovoltaic power station power data virtual acquisition method |
| CN112200391A (en) * | 2020-11-17 | 2021-01-08 | 国网陕西省电力公司经济技术研究院 | Power distribution network edge side load prediction method based on k-nearest neighbor mutual information characteristic simplification |
| CN112529262A (en) * | 2020-11-27 | 2021-03-19 | 北京京能高安屯燃气热电有限责任公司 | Short-term power prediction method, device, computer equipment and storage medium |
| CN116610911A (en) * | 2023-07-19 | 2023-08-18 | 南昌工程学院 | Power consumption data restoration method and system based on Bayesian Gaussian tensor decomposition model |
| CN118134292A (en) * | 2024-05-07 | 2024-06-04 | 湖北华中电力科技开发有限责任公司 | Regional power flexible load prediction method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105243428A (en) * | 2015-09-07 | 2016-01-13 | 天津市市政工程设计研究院 | Bus arrival time prediction method through optimizing support vector machine based on bat algorithm |
| CN105955032A (en) * | 2016-06-23 | 2016-09-21 | 上海电机学院 | Inverter control method for optimization of extreme learning machine on the basis of bat algorithm |
| CN106384153A (en) * | 2016-09-18 | 2017-02-08 | 河海大学常州校区 | WSAN actuator task distribution method based on BA-BPNN data fusion |
| CN106570581A (en) * | 2016-10-26 | 2017-04-19 | 东北电力大学 | Attribute association based load prediction system and method in energy Internet environment |
| CN107688862A (en) * | 2017-10-12 | 2018-02-13 | 电子科技大学 | Insulator equivalent salt density accumulation rate Forecasting Methodology based on BA GRNN |
| CN108121999A (en) * | 2017-12-10 | 2018-06-05 | 北京工业大学 | Support vector machines parameter selection method based on mixing bat algorithm |
-
2018
- 2018-06-14 CN CN201810613378.7A patent/CN108830418A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105243428A (en) * | 2015-09-07 | 2016-01-13 | 天津市市政工程设计研究院 | Bus arrival time prediction method through optimizing support vector machine based on bat algorithm |
| CN105955032A (en) * | 2016-06-23 | 2016-09-21 | 上海电机学院 | Inverter control method for optimization of extreme learning machine on the basis of bat algorithm |
| CN106384153A (en) * | 2016-09-18 | 2017-02-08 | 河海大学常州校区 | WSAN actuator task distribution method based on BA-BPNN data fusion |
| CN106570581A (en) * | 2016-10-26 | 2017-04-19 | 东北电力大学 | Attribute association based load prediction system and method in energy Internet environment |
| CN107688862A (en) * | 2017-10-12 | 2018-02-13 | 电子科技大学 | Insulator equivalent salt density accumulation rate Forecasting Methodology based on BA GRNN |
| CN108121999A (en) * | 2017-12-10 | 2018-06-05 | 北京工业大学 | Support vector machines parameter selection method based on mixing bat algorithm |
Non-Patent Citations (3)
| Title |
|---|
| 姚陶: "群智能算法在短期电力负荷预测中的研究及应用", 《万方学位论文在线数据库》 * |
| 宁小磊，等: "一种改进的灰色关联模型验证方法研究", 《计算机仿真》 * |
| 王晓东，等: "WNBA－RBF算法在企业经营状况评价中的应用", 《纺织高校基础科学学报》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109887035A (en) * | 2018-12-27 | 2019-06-14 | 哈尔滨理工大学 | Based on bat algorithm optimization BP neural network binocular vision calibration |
| CN109376971B (en) * | 2018-12-29 | 2022-04-26 | 北京中电普华信息技术有限公司 | Load curve prediction method and system for power consumers |
| CN109376971A (en) * | 2018-12-29 | 2019-02-22 | 北京中电普华信息技术有限公司 | A kind of load curve forecasting method and system towards power consumer |
| CN110309988A (en) * | 2019-07-12 | 2019-10-08 | 广东电网有限责任公司 | A kind of Methods of electric load forecasting based on grey relational grade and support vector machines |
| CN110728401A (en) * | 2019-10-10 | 2020-01-24 | 郑州轻工业学院 | Short-term power load prediction method of neural network based on squirrel and weed hybrid algorithm |
| CN110967471A (en) * | 2019-11-14 | 2020-04-07 | 广东电网有限责任公司 | Method for predicting concentration of dissolved gas in transformer oil |
| CN111340273B (en) * | 2020-02-17 | 2022-08-26 | 南京邮电大学 | Short-term load prediction method for power system based on GEP parameter optimization XGboost |
| CN111340273A (en) * | 2020-02-17 | 2020-06-26 | 南京邮电大学 | Short-term load prediction method for power system based on GEP parameter optimization XGboost |
| CN111612227A (en) * | 2020-05-12 | 2020-09-01 | 国网河北省电力有限公司电力科学研究院 | Load prediction method based on K-means clustering and bat optimization neural network |
| CN112052913A (en) * | 2020-09-27 | 2020-12-08 | 国网江苏省电力有限公司南京供电分公司 | Distributed photovoltaic power station power data virtual acquisition method |
| CN112052913B (en) * | 2020-09-27 | 2022-08-23 | 国网江苏省电力有限公司南京供电分公司 | Distributed photovoltaic power station power data virtual acquisition method |
| CN112200391A (en) * | 2020-11-17 | 2021-01-08 | 国网陕西省电力公司经济技术研究院 | Power distribution network edge side load prediction method based on k-nearest neighbor mutual information characteristic simplification |
| CN112200391B (en) * | 2020-11-17 | 2023-07-04 | 国网陕西省电力公司经济技术研究院 | Power distribution network edge side load prediction method based on k-nearest neighbor mutual information feature simplification |
| CN112529262A (en) * | 2020-11-27 | 2021-03-19 | 北京京能高安屯燃气热电有限责任公司 | Short-term power prediction method, device, computer equipment and storage medium |
| CN116610911A (en) * | 2023-07-19 | 2023-08-18 | 南昌工程学院 | Power consumption data restoration method and system based on Bayesian Gaussian tensor decomposition model |
| CN116610911B (en) * | 2023-07-19 | 2023-09-19 | 南昌工程学院 | Power consumption data restoration method and system based on Bayesian Gaussian tensor decomposition model |
| CN118134292A (en) * | 2024-05-07 | 2024-06-04 | 湖北华中电力科技开发有限责任公司 | Regional power flexible load prediction method and system |
| CN118134292B (en) * | 2024-05-07 | 2024-07-23 | 湖北华中电力科技开发有限责任公司 | Regional power flexible load prediction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830418A (en) | A kind of Short-Term Load Forecasting Method | |
| CN110730046B (en) | Cross-frequency-band spectrum prediction method based on deep migration learning | |
| CN108564192B (en) | Short-term photovoltaic power prediction method based on meteorological factor weight similarity day | |
| CN106485262B (en) | Bus load prediction method | |
| CN106096767A (en) | A kind of link travel time prediction method based on LSTM | |
| CN103730006B (en) | A kind of combination forecasting method of Short-Term Traffic Flow | |
| CN109886387B (en) | Traffic time sequence prediction method based on gating network and gradient lifting regression | |
| CN109063911A (en) | A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks | |
| CN110942205B (en) | Short-term photovoltaic power generation power prediction method based on HIMVO-SVM | |
| CN105976051A (en) | Wavelet transformation and improved firefly-optimized extreme learning machine-based short-term load prediction method | |
| CN114897227B (en) | Multi-steel mechanical property forecasting method based on improved random forest algorithm | |
| CN109671272B (en) | Expressway traffic flow prediction method | |
| CN105678397A (en) | Short-term photovoltaic power prediction method based on improved EMD algorithm and Elman algorithm | |
| CN112330487A (en) | Photovoltaic power generation short-term power prediction method | |
| CN112561178A (en) | Distributed photovoltaic output prediction method considering microclimate factors | |
| CN108460486A (en) | A kind of voltage deviation prediction technique based on improvement clustering algorithm and neural network | |
| CN111160650B (en) | Adaboost algorithm-based traffic flow characteristic analysis and prediction method | |
| CN108280998A (en) | Short-time Traffic Flow Forecasting Methods based on historical data dynamic select | |
| CN110222908A (en) | A kind of short-term electricity demand forecasting method based on data-driven | |
| CN112507613B (en) | Second-level ultra-short-term photovoltaic power prediction method | |
| CN109961190B (en) | Photovoltaic power station day-ahead power prediction method based on learning day fitting degree clustering | |
| CN110322075A (en) | A kind of scenic spot passenger flow forecast method and system based on hybrid optimization RBF neural | |
| CN111582567B (en) | Wind power probability prediction method based on hierarchical integration | |
| CN117077671B (en) | Interactive data generation method and system | |
| Ghofrani et al. | Hybrid clustering-time series-bayesian neural network short-term load forecasting method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181116 |