CN106855957A - Factory's bus load prediction based on similar day and least square method supporting vector machine - Google Patents
Factory's bus load prediction based on similar day and least square method supporting vector machine Download PDFInfo
- Publication number
- CN106855957A CN106855957A CN201510903757.6A CN201510903757A CN106855957A CN 106855957 A CN106855957 A CN 106855957A CN 201510903757 A CN201510903757 A CN 201510903757A CN 106855957 A CN106855957 A CN 106855957A
- Authority
- CN
- China
- Prior art keywords
- particle
- formula
- load
- chaos
- bus load
- 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
- 238000000034 method Methods 0.000 title claims abstract description 71
- 239000013598 vector Substances 0.000 title claims abstract description 60
- 239000002245 particle Substances 0.000 claims abstract description 65
- 238000005457 optimization Methods 0.000 claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 15
- 210000004205 output neuron Anatomy 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 4
- 210000002364 input neuron Anatomy 0.000 claims description 4
- 230000035772 mutation Effects 0.000 claims description 4
- 230000007935 neutral effect Effects 0.000 claims description 4
- 230000007812 deficiency Effects 0.000 claims description 3
- 230000002028 premature Effects 0.000 claims description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 239000006185 dispersion Substances 0.000 claims description 2
- 230000010429 evolutionary process Effects 0.000 claims description 2
- 230000009191 jumping Effects 0.000 claims description 2
- 238000011002 quantification Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims 1
- 238000011160 research Methods 0.000 claims 1
- 230000000739 chaotic effect Effects 0.000 abstract description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 238000002592 echocardiography Methods 0.000 abstract 1
- 230000005611 electricity Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000003723 Smelting Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 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
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of Forecasting Methodology for the special bus load of large size industrial enterprise, select the similar day of day to be predicted using the fuzzy clustering algorithm of SOM networks, similar day load data is decomposed by db4 small echos, after denoising and reconstruction processing as the training sample of later stage forecast model;The punishment parameter and kernel function cover width of least square method supporting vector machine are optimized using Chaos particle swarm optimization algorithm, constructs the bus load forecast model of chaotic particle swarm optimization least square method supporting vector machine.The inventive method is for the special bus load species of large size industrial enterprise is more, skewness, it is regular weak the features such as, carry out the load prediction of bus, bus load precision of prediction can be effectively improved.
Description
Technical field
The present invention relates to a kind of Forecasting Methodology for the special bus load of large size industrial enterprise.
Background technology
To the accuracy of factory's special bus load prediction for improving security and stability, production efficiency and the saving of factory
The aspects such as cost are all significant.The special bus load of factory is similar to the bus load of power network, but its load value is much smaller than and is
System load, therefore prediction radix is smaller, it is regular weak.And special bus load is highly susceptible to the small electricity in power supply area
Source, user, in transformer station the influence such as transformer operation manners, environmental factor and there is data mutation.So relative system load
For prediction, bus load prediction is increasingly difficult, and precision of prediction is relatively low;Current bus load prediction mainly uses bus
The historical data of load, part throttle characteristics and corresponding influence factor directly carry out bus load prediction.Therefore, studying a kind of energy
Forecasting Methodology suitable for the special bus load of large size industrial enterprise has very important significance.
The content of the invention
It is an object of the invention to provide a kind of Forecasting Methodology for the special bus load of large size industrial enterprise.Above-mentioned mesh
Realized by following technical scheme:
(1)Set up bus load prediction similar day Selection Model
The special bus load of large size industrial enterprise have many species, skewness, it is regular weak the features such as, and bus load predict
It is required that promptness, belongs to short-term load forecasting, reasonable selection similar day is the effective way for improving bus load prediction effect.This
The method that invention is combined using SOM self-organizing feature map neural networks with fuzzy clustering algorithm, to bus historical load data
Classified, realized the selection of similar day.The process that the present invention sets up bus load prediction similar day Selection Model is as follows:
1st step:Basic structure feature according to SOM networks(Such as accompanying drawing 1), it is by output layer(That is competition layer)With input layer group
Into;Input layer is made up of n neuron, can be input into the sample with n-dimensional vector;Output layer is made up of a × b neuron
Two dimensional surface battle array;So, can by maximum temperature in historical data, minimum temperature, mean temperature, relative humidity, week type and
The load characteristics such as load data vector set X=[x1, x2, x3... .., xn]T, SOM networks are input into as input vector;
2nd step:Random number is assigned to the weight vector between j-th output neuron and i-th input neuron at the t=0 moment
Wij(0);Wherein Wij(t)= [w1j, w2j, w3j... .., wnj]T(i=1,2,3 ... ..., n;J=1,2, a × b), represent output layer
J-th neuron respectively with the connection weight between i-th input neuron;
3rd step:According to formula(1)Calculate the Euclidean distance of j-th output neuron and input sample:
(1)
After calculating input sample and the distance of all output neurons, it is determined that meeting minimum range i.e. dl=min(dj)Neuron
l;
4th step:If dl<β(β is setting threshold values)Then carry out the study of weights;By formula(2)Amendment neuron weight vector, and hold
The step of row the 5th;
(2)
In formula:T represents the time, and η (t) is learning rate, is typically taken:
(3)
If dl>Then just output neuron is revised as next output neuron to β(j=j+1)And using the input sample as Wij+1
The initial weight vector W of (t)ij+1(0), it is transferred to the 3rd step;
5th step:Next input sample is input into, 3-4 steps is repeated until all training samples train completion;
6th step:Obtain preliminary clusters number c(Wherein c=a × b)With cluster centre M1(0), M2(0) ..., Mc(0);As
The input of FCM clustering algorithms;
7th step:The threshold epsilon of fuzzy clustering algorithm is selected, maximum iteration t is setmax;For input Mk(0) (k=1,
2 ..., c) and t=1,2 ..., tmax;By formula(4)Calculate membership function uij(t);
(4)
Wherein K (t) is a variable control parameter, can be tried to achieve by following formula:
(5)
In formula:K0It is the constant being previously set, there is K0>1;
8th step:Using membership function uij(t) vicarious learning modifying rates weight coefficient, i.e.,
(6)
In formula:NjT () is the field of t output neuron j;Until just terminating, the 7th step is otherwise gone to;
9th step:Output clustering information, that is, obtain similar day information;
(2)Bus load historical data is pre-processed
Bus load data are likely to result in the mutation of bus historical load data, mistake etc. in collection and transmitting procedure and hair occur
Thorn;In order to effectively improve load prediction accuracy, bus load historical data sample is pre-processed using Wavelet Denoising Method;It is right
The historical load data of fixed test point carries out Wavelet Denoising Method treatment;The small echo selection preferable db4 wavelet basis of denoising effect is decomposed,
Decomposition order K is set to 3;Quantification treatment is carried out to the high frequency coefficient threshold values after wavelet decomposition, and using global threshold denoising
Method determines threshold values thresholding T, and threshold values thresholding computing formula is:
(7)
In formula:σ is noise criteria;D is the length of signal;
Noise criteria σ can be determined by the absolute intermediate value of small echo high frequency coefficient after decomposition:
(8)
In formula:Cd is the intermediate value of high frequency series cd [i] (i=1,2 ..., K);
Afterwards, the 3rd layer of low frequency sequence being obtained to wavelet decomposition and by the 1st layer after denoising to the 3rd floor height frequency sequence
Carry out wavelet reconstruction;
(3)Set up the least square method supporting vector machine bus load forecast model using Chaos particle swarm optimization algorithm optimization
Least square method supporting vector machine model needs to determine the cover width of punishment parameter γ and gaussian kernel function in modeling process
δ;Wherein, punishment parameter γ is equilibrium factor, can be come the complexity of decision model and inclined to being fitted according to the characteristic of sample data
Poor punishment degree;The contact that δ is reflected between the correlation degree between supporting vector, the small then supporting vectors of δ is relatively loose, learning machine
Device is relative complex, and Generalization Ability cannot be guaranteed;Influence between the big then supporting vectors of δ is strong, and regression model is difficult to reach essence enough
Degree;In actual applications, γ and δ values are mainly set by rule of thumb, will influence precision of prediction;The present invention is calculated for standard particle group
Method is easily absorbed in local optimum and the deficiency of Premature Convergence occurs, Chaos Variable is introduced, using Chaos Search mechanism, to standard particle
Group's algorithm is improved, and for the cover width of punishment parameter γ and gaussian kernel function in least square method supporting vector machine model
The optimization of δ;Detailed process is as follows:
Step 1:Initialization particle populations and particle rapidity;Particle populations number q, maximum iteration k are setmax, maximum inertia
Weight coefficient wmaxWith minimum inertia weight coefficient wmin, Studying factors c1And c2, the parameter such as particle dimension C;Produce random particles position
Put Xi=(xi1, xi2, xi3..., xiC)TWith speed Vi=(Vi1, Vi2, Vi3..., ViC)T;
Step 2:Calculate the fitness value of each particle, and more new individual extreme value Pi=(Pi1, Pi2, Pi3..., PiC)TWith population pole
Value Pg=(Pg1, Pg2, Pg3..., PgC)T;
Step 3:Using formula(9)And formula(10)The average grain of particle is calculated away from D (t) and fitness variance λ2, judge D (t)
<α or λ2<β (α, β are threshold values), illustrates that particle is precocious if setting up, and carries out Chaos Search, otherwise, performs step 5;
(9)
In formula:L is the diagonal maximum length in search space;Q is population scale;PidIt is i-th d dimensional coordinate values of particle position;
It is the average of the d dimensional coordinate values of all particle positions;Average grain is smaller away from D (t), represents that population gets over concentration;D (t) is bigger, table
Show that population gets over dispersion;
(10)
In formula:λ2It is fitness variance;fiIt is the fitness of individual i;It is the average fitness of current population;F is fixed for normalization
The mark factor, for limiting λ2Size, value is:
(11)
Step 4:Chaos Search Method;Randomly generate Chaos Variable Y0, recycle formula(12)Produce chaos iteration sequence Yh;Profit
Use formula(13)The interval of optimized variable is transformed to, until chaos iteration number of times h>(H is Chaos Search iteration time to H
Number), it is optimal feasible solution, replace a particle at random using this optimal solution;
(12)
(13)
In formula:C represents particle dimension;Pgd(d=1,2,…,C)Represent the optimal location of particle d dimensions; ad, bdBecome for d is tieed up
The span of amount;
Step 5:Using formula(14)With(15)Update particle rapidity and position;
(14)
(15)
Wherein d=1,2 ..., C;I=1,2 ..., q;K is current iteration number of times;W is inertia weight coefficient; r1And r2It is [0, l]
Between random number;
Be conducive to jumping out local minimum point when w is larger, algorithmic statement is conducive to when smaller and the precision of optimizing solution is improved;
Therefore inertia weight coefficient linear decrease strategy amendment w values can be utilized, computing formula is as follows:
(16)
In formula, wmaxIt is maximum inertia weight, typically takes 0.9;wminIt is minimum inertia weight, typically takes 0.4;kmaxIt is setting
Maximum iteration;
Step 6:Judge the condition of convergence;If meeting the condition of convergence, step 7 is performed;Otherwise, step 2 is gone to;
Step 7:Evolutionary process terminates, and returns to globally optimal solution;
Step 8:The global optimum that Chaos particle swarm optimization algorithm is returned assigns the penalty coefficient of least square method supporting vector machine
With the cover width of gaussian kernel function, Optimized Least Square Support Vector model.
Beneficial effect
1. the present invention is analyzed and researched for the system of selection of similar day in bus load prediction, is overcome with SOM god
The shortcoming of accurate clustering information after prediction similar day cannot not provide classification is chosen through network;
2. the method that the utilization Wavelet Denoising Method that the present invention is mentioned is pre-processed to bus load historical data, can effectively reduce number
There is burr according to the mutation of bus historical load data, mistake caused in collection and transmitting procedure etc., reduce bus load number
According to noise pollution.So as to effectively improve later stage load prediction accuracy;
3. the present invention is easily absorbed in local optimum and the deficiency of Premature Convergence occurs for standard particle group's algorithm, introduces chaos and becomes
Amount, using Chaos Search mechanism, is improved to standard particle group's algorithm, and punishing for least square method supporting vector machine model
The cover width δ optimizations of penalty parameter γ and gaussian kernel function, effectively raise the precision of bus load prediction.
Brief description of the drawings
Fig. 1 is the basic block diagram of SOM networks
Fig. 2 is that the historical load in certain large-scale metallurgical factory special 35Kv buses May carries out the result figure of Wavelet Denoising Method treatment
Fig. 3 is May 31 respectively with least square method supporting vector machine, particle group optimizing least square method supporting vector machine, chaos
The load value figure of particle group optimizing least square method supporting vector machine prediction
Fig. 4 is May 31 respectively with least square method supporting vector machine, particle group optimizing least square method supporting vector machine, chaos
Each future position relative error figure when particle group optimizing least square method supporting vector machine is predicted
Fig. 5 is the load value figure that continuous 3 days of June was predicted with chaotic particle swarm optimization least square method supporting vector machine.
Specific embodiment
Embodiment 1:
(1)Choose certain large-scale metallurgical enterprise-specific 35kV bus load May(31 days)Data sample, included most in sample
High-temperature, minimum temperature, mean temperature, relative humidity, date type(Working day, weekend)And load data(Daily 24
Point);It was prediction day with 31 days, ten is chosen from other 30 days data using the fuzzy clustering algorithm model and artificial experience of SOM
Individual similar day, as a result as shown in table 1;
Artificial and SOM the fuzzy clustering algorithm of table 1 chooses similar day
Used as training sample, day is predicted in the conduct of May 31 to 10 similar days that two methods are selected, using least square branch
Hold vector machine model to be predicted, obtain result such as table 2;
Table 2 is artificial and SOM fuzzy clustering algorithms are chosen similar day and are used for
From table 2 it can be seen that choosing the similar day of prediction day using SOM fuzzy clustering algorithms, make average daily load prediction accuracy rate
3 percentage points are improve, average relative error reduces nearly 4 percentage points.So choosing similar day using SOM fuzzy clustering algorithms
Choose more scientific with respect to artificial experience, the precision of later stage load prediction can be effectively improved;
(2)Historical load to certain large-scale metallurgical enterprise-specific 35kV bus May carries out Wavelet Denoising Method pretreatment, pretreatment knot
Really(Such as accompanying drawing 2)It is shown;
S subgraphs are original loads data in accompanying drawing 2, and ca3 subgraphs are the low frequency part after load sequence is decomposed, and it is load that can regard as
Primary signal, do not make denoising.Cd1, cd2, cd3 subgraph are the HFS of load sequence, and load variations acutely, embody
Relation between short-term load (duty) and enchancement factor.Sd is signal after the denoising that reconstruct is obtained, it can be seen that most of noise is
It is eliminated;
(3)Choose May before 24 days historical load through Wavelet Denoising Method pre-process after data as training sample, with least square branch
Vector machine is held for forecast model, and by the punishment parameter γ of the least square method supporting vector machine and cover width δ of gaussian kernel function
1 is taken, continuous prediction 7 days(25 to 31)Load, predict the outcome as shown in table 3;
The training sample of table 3 is continuous after being pre-processed through Wavelet Denoising Method
The MRE values and A of prediction 7 daysLValue
From table 3 it can be seen that without the Wavelet Denoising Method before processing average daily load prediction accuracy rate (A of continuous seven daysL) average value
It is 74.97%, the average value of average relative error (MRE) is 22.75%;The A predicted after being pre-processed through Wavelet Denoising MethodLAverage value
The average value for reaching 89.22%, MRE is greatly reduced, and only 8.67%;So by historical data before the prediction by Wavelet Denoising Method at
Reason can effectively improve load prediction precision;
(2)It it is respectively prediction day with May 29,30 days, 31 days;It is sample 1 to remove day to be predicted remaining historical load;Profit
Ten similar days of day to be predicted are chosen with the fuzzy clustering algorithm of SOM(As shown in table 1)As sample 2;By 2 points of sample 1, sample
As the training sample of least square method supporting vector machine model, punishment parameter γ and Gaussian kernel letter after not pre-processed through Wavelet Denoising Method
Several cover width δ still take 1;Predict the outcome as shown in table 4;
The fuzzy clustering algorithm of the SOM of table 4 chooses similar day and after Wavelet Denoising Method
The MRE values and A of continuous prediction 3 daysLValue
From table 4, it can be seen that when not choosing similar day as training sample, although number of training is bigger, continuously predicting 3 days
ALIt is 8.63% that average value only has the average value of 88.63%, MRE;Day to be predicted is chosen using the fuzzy clustering algorithm of SOM neutral nets
Similar day as training sample after, number of training is reduced, but A for three days on endLAverage value reaches the average of 97.66%, MRE
Value is reduced to 2.22%, and the average relative error of each prediction day is both less than 5%, fully meets precision of prediction demand;So
Similar day is chosen using SOM neutral nets, bus load precision of prediction can be improved;
(3)It it is respectively prediction day with May 31,30 days, 29 days, 28 days, 27 days.Using the fuzzy clustering algorithm of SOM neutral nets
The similar day of day to be predicted is chosen from May remaining number of days, and carries out Wavelet Denoising Method pretreatment;Forecast model is chosen most respectively
A young waiter in a wineshop or an inn multiplies SVMs, particle cluster algorithm Optimized Least Square Support Vector, Chaos particle swarm optimization algorithm optimization least square
The punishment parameter γ of SVMs, wherein least square method supporting vector machine model and the cover width σ of gaussian kernel function take
It is 1, particle cluster algorithm Optimized Least Square Support Vector model, Chaos particle swarm optimization algorithm Optimized Least Square Support Vector
Two kinds of particle cluster algorithm hunting zones of model are set as:γ ∈ [0.1,10], δ ∈ [0.1,1].Population q takes 40, maximum
Iterations kmaxTake 200, c1And c22 are, inertia weight w maximums take 0.9, and minimum value takes 0.4;Three kinds of models are continuously predicted
As shown in table 5, accompanying drawing 3 is the prediction load value of 31 days, each future position phase to the average relative error and root mean square relative error of 5 days
To error as shown in Figure 4;
5 three kinds of models of table continuously predict MRE values and the RMSE value of 5 days
As can be seen from Table 5, the least square method supporting vector machine model middle MRE that predicts the outcome of continuous 5 days is close or larger than 2%,
RMSE is all higher than 3%;Using the prediction after particle cluster algorithm and Chaos particle swarm optimization algorithm Optimized Least Square Support Vector parameter
MRE values and RMSE value decrease in model prediction result, so using optimized algorithm Optimized Least Square Support Vector
Punishment parameter and kernel function cover width can improve precision of prediction of the least square method supporting vector machine to bus load;Using mixed
During ignorant particle cluster algorithm Optimized Least Square Support Vector model prediction, relative error is less than 1.8%, root mean square relative error
About 2%, show that the optimizing result of Chaos particle swarm optimization algorithm is more accurate, it has been likely to occur grain during standard particle group's algorithm optimizing
Sub- precocity phenomenon, causes item forecast resultant error larger;
From accompanying drawing 3, accompanying drawing 4 as can be seen that three kinds of predicted values of model 0 ~ 9 point, 18 ~ 22 time period future position and reality
Actual value is closer to, and 10 ~ 17 points of predicted value and the error of actual value are larger;Reason is 10 points to 17 time periods, in enterprise
In addition to main smelting furnace, many other mini electrical equipments(Such as dust removal machine, water treatment facilities, annular furnace combustion-supporting machine)Input
Causing electricity consumption increases;Bus load species increases, and regularity dies down, and have impact on prediction effect;
Embodiment 2:
Here it is prediction day with 30 days of June, 29 days, 28 days.Ten similar days of selection, pre-process by Wavelet Denoising Method, then
With Chaos particle swarm optimization algorithm Optimized Least Square Support Vector model prediction;Predict load value as shown in Figure 5, MRE values,
RMSE value and ALValue is as shown in table 6;
The Chaos particle swarm optimization algorithm Optimized Least Square Support Vector of bus 2 of table 6
MRE values, RMSE value and A that model continuously predicts 3LValue
From accompanying drawing 5 and table 6 as can be seen that in 72 time points of continuous prediction 3 days, MRE values are less than 2%, ALValue is more than 97%, table
Load prediction of the bright Chaos particle swarm optimization algorithm Optimized Least Square Support Vector model to bus 2 meets precision of prediction requirement.
Claims (3)
1. it is a kind of based on similar day selection and Chaos particle swarm optimization algorithm Optimized Least Square Support Vector large size industrial enterprise
Special bus load Forecasting Methodology, it is characterised in that:When carrying out bus load and predicting, there is provided the similar day to predicting day
System of selection and load data early stage preprocess method, set up load forecasting model on this basis, carry out bus load prediction;
The method comprises the following steps:
(1)Set up the similar day Selection Model that bus load predicts day
The special bus load of large size industrial enterprise have many species, skewness, it is regular weak the features such as, and require bus load
Prediction promptness is strong, belongs to short-term load forecasting;In order to carry out load prediction, can be entered by selecting the similar day of load prediction day
Row research, reasonable selection similar day is the effective way for improving bus load precision of prediction;The present invention uses SOM self-organizing features
Map neural network is combined with fuzzy clustering algorithm, and bus historical load data is classified, and realizes the selection of similar day;This
The process that bus load prediction similar day Selection Model is set up in invention is as follows:
1st step:Basic structure feature according to SOM networks(Such as accompanying drawing 1), it is by output layer(That is competition layer)With input layer group
Into;Input layer is made up of n neuron, can be input into the sample with n-dimensional vector;Output layer is made up of a × b neuron
Two dimensional surface battle array;So, can by maximum temperature in historical data, minimum temperature, mean temperature, relative humidity, week type and
The load characteristics such as load data vector set X=[x1, x2, x3... .., xn]T, SOM networks are input into as input vector;
2nd step:Random number is assigned to the weight vector between j-th output neuron and i-th input neuron at the t=0 moment
Wij(0);Wherein Wij(t)= [w1j, w2j, w3j... .., wnj]T(i=1,2,3 ... ..., n;J=1,2, a × b), represent output layer
J-th neuron respectively with the connection weight between i-th input neuron;
3rd step:According to formula(1)Calculate the Euclidean distance of j-th output neuron and input sample:
(1)
After calculating input sample and the distance of all output neurons, it is determined that meeting minimum range i.e. dl=min(dj)Neuron
l;
4th step:If dl<β(β is setting threshold values)Then carry out the study of weights;By formula(2)Amendment neuron weight vector, and perform
5th step;
(2)
In formula:T represents the time, and η (t) is learning rate, is typically taken:
(3)
If dl>Then just output neuron is revised as next output neuron to β(j=j+1)And using the input sample as Wij+1
The initial weight vector W of (t)ij+1(0), it is transferred to the 3rd step;
5th step:Next input sample is input into, 3-4 steps is repeated until all training samples train completion;
6th step:Obtain preliminary clusters number c(Wherein c=a × b)With cluster centre M1(0), M2(0) ..., Mc(0);As
The input of FCM clustering algorithms;
7th step:The threshold epsilon of fuzzy clustering algorithm is selected, maximum iteration t is setmax;For input Mk(0) (k=1,2 ...,
And t=1,2 ..., t c)max;By formula(4)Calculate membership function uij(t);
(4)
Wherein K (t) is a variable control parameter, can be tried to achieve by following formula:
(5)
In formula:K0It is the constant being previously set, there is K0>1;
8th step:Using membership function uij(t) vicarious learning modifying rates weight coefficient, i.e.,
(6)
In formula:NjT () is the field of t output neuron j;Until just terminating, the 7th step is otherwise gone to;
9th step:Output clustering information, that is, obtain similar day information;
(2)Bus load historical data is pre-processed
Bus load data are likely to result in the mutation of bus historical load data, mistake etc. in collection and transmitting procedure and hair occur
Thorn;In order to effectively improve load prediction accuracy, bus load historical data sample is pre-processed using Wavelet Denoising Method;It is right
The historical load data of fixed test point carries out Wavelet Denoising Method treatment;The small echo selection preferable db4 wavelet basis of denoising effect is decomposed,
Decomposition order K is set to 3;Quantification treatment is carried out to the high frequency coefficient threshold values after wavelet decomposition, and using global threshold denoising
Method determines threshold values thresholding T, and threshold values thresholding computing formula is:
(7)
In formula:σ is noise criteria;D is the length of signal;
Noise criteria σ can be determined by the absolute intermediate value of small echo high frequency coefficient after decomposition:
(8)
In formula:Cd is the intermediate value of high frequency series cd [i] (i=1,2 ..., K);
Afterwards, the 3rd layer of low frequency sequence being obtained to wavelet decomposition and by the 1st layer after denoising to the 3rd floor height frequency sequence
Carry out wavelet reconstruction;
(3)Set up the least square method supporting vector machine bus load forecast model using Chaos particle swarm optimization algorithm optimization
Least square method supporting vector machine model needs to determine the cover width of punishment parameter γ and gaussian kernel function in modeling process
δ;Wherein, punishment parameter γ is equilibrium factor, can be come the complexity of decision model and inclined to being fitted according to the characteristic of sample data
Poor punishment degree;The contact that δ is reflected between the correlation degree between supporting vector, the small then supporting vectors of δ is relatively loose, learning machine
Device is relative complex, and Generalization Ability cannot be guaranteed;Influence between the big then supporting vectors of δ is strong, and regression model is difficult to reach essence enough
Degree;In actual applications, γ and δ values are mainly set by rule of thumb, will influence precision of prediction;The present invention is calculated for standard particle group
Method is easily absorbed in local optimum and the deficiency of Premature Convergence occurs, Chaos Variable is introduced, using Chaos Search mechanism, to standard particle
Group's algorithm is improved, and for the cover width of punishment parameter γ and gaussian kernel function in least square method supporting vector machine model
The optimization of δ;Detailed process is as follows:
Step 1:Initialization particle populations and particle rapidity;Particle populations number q, maximum iteration k are setmax, maximum inertia
Weight coefficient wmaxWith minimum inertia weight coefficient wmin, Studying factors c1And c2, the parameter such as particle dimension C;Produce random particles position
Put Xi=(xi1, xi2, xi3..., xiC)TWith speed Vi=(Vi1, Vi2, Vi3..., ViC)T;
Step 2:Calculate the fitness value of each particle, and more new individual extreme value Pi=(Pi1, Pi2, Pi3..., PiC)TWith population extreme value
Pg=(Pg1, Pg2, Pg3..., PgC)T;
Step 3:Using formula(9)And formula(10)The average grain of particle is calculated away from D (t) and fitness variance λ2, judge D (t)
<α or λ2<β (α, β are threshold values), illustrates that particle is precocious if setting up, and carries out Chaos Search, otherwise, performs step 5;
(9)
In formula:L is the diagonal maximum length in search space;Q is population scale;PidIt is i-th d dimensional coordinate values of particle position;
It is the average of the d dimensional coordinate values of all particle positions;Average grain is smaller away from D (t), represents that population gets over concentration;D (t) is bigger, table
Show that population gets over dispersion;
(10)
In formula:λ2It is fitness variance;fiIt is the fitness of individual i;It is the average fitness of current population;F is echo cancellation
The factor, for limiting λ2Size, value is:
(11)
Step 4:Chaos Search Method;Randomly generate Chaos Variable Y0, recycle formula(12)Produce chaos iteration sequence Yh;Profit
Use formula(13)The interval of optimized variable is transformed to, until chaos iteration number of times h>(H is Chaos Search iteration time to H
Number), it is optimal feasible solution, replace a particle at random using this optimal solution;
(12)
(13)
In formula:C represents particle dimension;Pgd(d=1,2,…,C)Represent the optimal location of particle d dimensions; ad, bdFor d ties up variable
Span;
Step 5:Using formula(14)With(15)Update particle rapidity and position;
(14)
(15)
Wherein d=1,2 ..., C;I=1,2 ..., q;K is current iteration number of times;W is inertia weight coefficient; r1And r2Be [0, l] it
Between random number;
Be conducive to jumping out local minimum point when w is larger, algorithmic statement is conducive to when smaller and the precision of optimizing solution is improved;
Therefore inertia weight coefficient linear decrease strategy amendment w values can be utilized, computing formula is as follows:
(16)
In formula, wmaxIt is maximum inertia weight, typically takes 0.9;wminIt is minimum inertia weight, typically takes 0.4;kmaxIt is setting
Maximum iteration;
Step 6:Judge the condition of convergence;If meeting the condition of convergence, step 7 is performed;Otherwise, step 2 is gone to;
Step 7:Evolutionary process terminates, and returns to globally optimal solution;
Step 8:The global optimum that Chaos particle swarm optimization algorithm is returned assigns the penalty coefficient of least square method supporting vector machine
With the cover width of gaussian kernel function, Optimized Least Square Support Vector model.
2. the bus load as described in claim 1 predicts similar day Selection Model, it is characterised in that SOM networks do not need priori
The supervision of knowledge and learning process, can automatically find out the degree of approximation between sample, so as to realize the classification to sample data;But
SOM neutral nets can not provide accurate clustering information after classification, so, can be accurate by fuzzy clustering algorithm and SOM network integrations
Extract with the common substantive characteristics of type load, improve classifying quality.
3. the least square method supporting vector machine bus load prediction of the utilization Chaos particle swarm optimization algorithm optimization as described in claim 1
Model, it is characterised in that local optimum is easily sunk into the search for overcoming standard particle group's algorithm, and particle is absorbed in the shortcoming of precocity;
And Chaos Search is utilized, guiding particle quickly jumps out local optimum, accelerates convergence rate;
The global search of particle is realized using the ergodic of chaos, particle globally optimal solution is found;The optimal solution for obtaining can use
Punishment parameter and gaussian kernel function cover width in Optimized Least Square Support Vector, improve least square supporting vector
The precision of prediction of machine.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510903757.6A CN106855957A (en) | 2015-12-09 | 2015-12-09 | Factory's bus load prediction based on similar day and least square method supporting vector machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510903757.6A CN106855957A (en) | 2015-12-09 | 2015-12-09 | Factory's bus load prediction based on similar day and least square method supporting vector machine |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106855957A true CN106855957A (en) | 2017-06-16 |
Family
ID=59131437
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510903757.6A Pending CN106855957A (en) | 2015-12-09 | 2015-12-09 | Factory's bus load prediction based on similar day and least square method supporting vector machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106855957A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107506859A (en) * | 2017-08-17 | 2017-12-22 | 国网重庆市电力公司电力科学研究院 | A kind of SF6The Forecasting Methodology and system of decomposition gas concentration-time sequence |
CN108038271A (en) * | 2017-11-22 | 2018-05-15 | 华中科技大学 | A kind of abrasion wear prediction method of milling cutter and the method for state recognition |
CN108446429A (en) * | 2018-02-05 | 2018-08-24 | 电子科技大学 | A kind of particle stress finite element solving algorithm applied to PIC static electric charge models |
CN109583635A (en) * | 2018-11-16 | 2019-04-05 | 贵州电网有限责任公司 | A kind of short-term load forecasting modeling method towards operational reliability |
CN109685277A (en) * | 2018-12-28 | 2019-04-26 | 国网冀北电力有限公司经济技术研究院 | Electricity demand forecasting method and device |
CN109726862A (en) * | 2018-12-24 | 2019-05-07 | 深圳供电局有限公司 | A kind of user's daily electricity mode prediction method |
CN109871977A (en) * | 2018-12-25 | 2019-06-11 | 广东电网有限责任公司 | Load forecasting method based on wavelet transform and the minimum support vector machines of optimization |
CN110083079A (en) * | 2018-01-26 | 2019-08-02 | 阿里巴巴集团控股有限公司 | Technological parameter determines method, apparatus and system |
CN110489893A (en) * | 2019-08-23 | 2019-11-22 | 国网山东省电力公司莱芜供电公司 | A kind of bus load prediction technique and system based on variable weight |
CN110807508A (en) * | 2019-10-31 | 2020-02-18 | 国网辽宁省电力有限公司经济技术研究院 | Bus peak load prediction method considering complex meteorological influence |
CN110909912A (en) * | 2019-10-08 | 2020-03-24 | 西安交通大学 | Park electric power system net load combination prediction method based on self-adaptive error feedback |
CN111027760A (en) * | 2019-11-28 | 2020-04-17 | 广东工业大学 | Power load prediction method based on least square vector machine |
CN112773365A (en) * | 2019-10-22 | 2021-05-11 | 上海交通大学 | System for monitoring mental load of underwater vehicle during underwater operation |
CN112949207A (en) * | 2021-03-25 | 2021-06-11 | 上海电机学院 | Short-term load prediction method based on improved least square support vector machine |
CN113325696A (en) * | 2021-06-01 | 2021-08-31 | 吉林大学 | Hybrid control method for combining single neuron PID and model prediction applied to crosslinked cable production equipment |
-
2015
- 2015-12-09 CN CN201510903757.6A patent/CN106855957A/en active Pending
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107506859A (en) * | 2017-08-17 | 2017-12-22 | 国网重庆市电力公司电力科学研究院 | A kind of SF6The Forecasting Methodology and system of decomposition gas concentration-time sequence |
CN108038271B (en) * | 2017-11-22 | 2020-05-19 | 华中科技大学 | Wear prediction method and state recognition method for milling cutter |
CN108038271A (en) * | 2017-11-22 | 2018-05-15 | 华中科技大学 | A kind of abrasion wear prediction method of milling cutter and the method for state recognition |
CN110083079A (en) * | 2018-01-26 | 2019-08-02 | 阿里巴巴集团控股有限公司 | Technological parameter determines method, apparatus and system |
CN108446429A (en) * | 2018-02-05 | 2018-08-24 | 电子科技大学 | A kind of particle stress finite element solving algorithm applied to PIC static electric charge models |
CN109583635A (en) * | 2018-11-16 | 2019-04-05 | 贵州电网有限责任公司 | A kind of short-term load forecasting modeling method towards operational reliability |
CN109583635B (en) * | 2018-11-16 | 2022-04-22 | 贵州电网有限责任公司 | Short-term load prediction modeling method for operational reliability |
CN109726862A (en) * | 2018-12-24 | 2019-05-07 | 深圳供电局有限公司 | A kind of user's daily electricity mode prediction method |
CN109871977A (en) * | 2018-12-25 | 2019-06-11 | 广东电网有限责任公司 | Load forecasting method based on wavelet transform and the minimum support vector machines of optimization |
CN109685277A (en) * | 2018-12-28 | 2019-04-26 | 国网冀北电力有限公司经济技术研究院 | Electricity demand forecasting method and device |
CN110489893A (en) * | 2019-08-23 | 2019-11-22 | 国网山东省电力公司莱芜供电公司 | A kind of bus load prediction technique and system based on variable weight |
CN110909912A (en) * | 2019-10-08 | 2020-03-24 | 西安交通大学 | Park electric power system net load combination prediction method based on self-adaptive error feedback |
CN110909912B (en) * | 2019-10-08 | 2022-08-09 | 西安交通大学 | Park electric power system net load combination prediction method based on self-adaptive error feedback |
CN112773365A (en) * | 2019-10-22 | 2021-05-11 | 上海交通大学 | System for monitoring mental load of underwater vehicle during underwater operation |
CN110807508A (en) * | 2019-10-31 | 2020-02-18 | 国网辽宁省电力有限公司经济技术研究院 | Bus peak load prediction method considering complex meteorological influence |
CN110807508B (en) * | 2019-10-31 | 2023-06-09 | 国网辽宁省电力有限公司经济技术研究院 | Bus peak load prediction method considering complex weather influence |
CN111027760A (en) * | 2019-11-28 | 2020-04-17 | 广东工业大学 | Power load prediction method based on least square vector machine |
CN112949207A (en) * | 2021-03-25 | 2021-06-11 | 上海电机学院 | Short-term load prediction method based on improved least square support vector machine |
CN113325696A (en) * | 2021-06-01 | 2021-08-31 | 吉林大学 | Hybrid control method for combining single neuron PID and model prediction applied to crosslinked cable production equipment |
CN113325696B (en) * | 2021-06-01 | 2022-07-19 | 吉林大学 | Single neuron PID and model prediction combined hybrid control method applied to crosslinked cable production equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106855957A (en) | Factory's bus load prediction based on similar day and least square method supporting vector machine | |
He et al. | A discrete multi-objective fireworks algorithm for flowshop scheduling with sequence-dependent setup times | |
Kuo et al. | Integration of particle swarm optimization and genetic algorithm for dynamic clustering | |
CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
CN104636985A (en) | Method for predicting radio disturbance of electric transmission line by using improved BP (back propagation) neural network | |
CN111612125B (en) | Novel HTM time pool method and system for online learning | |
CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
Li et al. | Building’s hourly electrical load prediction based on data clustering and ensemble learning strategy | |
Abdechiri et al. | Adaptive imperialist competitive algorithm (AICA) | |
CN110163444A (en) | A kind of water demand prediction method based on GASA-SVR | |
CN109002928A (en) | A kind of electric load peak value prediction technique and device based on Bayesian network model | |
Tian et al. | An adaptive ensemble predictive strategy for multiple scale electrical energy usages forecasting | |
CN114119273A (en) | Park comprehensive energy system non-invasive load decomposition method and system | |
CN108985323A (en) | A kind of short term prediction method of photovoltaic power | |
Goh et al. | Hybrid SDS and WPT-IBBO-DNM Based Model for Ultra-short Term Photovoltaic Prediction | |
Zhong et al. | PV power short-term forecasting model based on the data gathered from monitoring network | |
Wang et al. | Electricity load forecasting based on support vector machines and simulated annealing particle swarm optimization algorithm | |
CN109408896A (en) | A kind of anerobic sowage processing gas production multi-element intelligent method for real-time monitoring | |
Liu et al. | A survey of artificial bee colony algorithm | |
CN111310974A (en) | Short-term water demand prediction method based on GA-ELM | |
CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
Tang | Optimal scheduling method of transport path in coastal port international logistics park | |
Jahan et al. | Intelligent system for power load forecasting in off-grid platform | |
CN109359671A (en) | A kind of classification intelligent extract method of Hydropower Station Reservoir Dispatching rule | |
CN111401638B (en) | Spatial load prediction method based on extreme learning machine and load density index method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170616 |
|
WD01 | Invention patent application deemed withdrawn after publication |