CN109919421A - Short-term power load prediction model establishment method based on VMD-PSO-BPNN - Google Patents
Short-term power load prediction model establishment method based on VMD-PSO-BPNN Download PDFInfo
- Publication number
- CN109919421A CN109919421A CN201910060863.0A CN201910060863A CN109919421A CN 109919421 A CN109919421 A CN 109919421A CN 201910060863 A CN201910060863 A CN 201910060863A CN 109919421 A CN109919421 A CN 109919421A
- Authority
- CN
- China
- Prior art keywords
- bpnn
- pso
- vmd
- model
- formula
- 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
Abstract
The invention discloses a short-term power load prediction model establishing method based on VMD-PSO-BPNN, which is applied to power load prediction of paper making enterprises and comprises the following steps: firstly, acquiring total effective load data of a papermaking enterprise; decomposing the preprocessed data sequence by utilizing a VMD decomposition algorithm; selecting an input variable of the model using a hysteresis autocorrelation; modeling the decomposition sequence by utilizing PSO-BPNN; and training the PSO-BPNN model by using the training samples, establishing a prediction model, predicting the electric load of the paper making enterprise, and analyzing the prediction effect. The short-term power load prediction model based on the VMD-PSO-BPNN established by the method has the characteristics of fast convergence and no delay of prediction results.
Description
Technical field
The present invention relates to papermaking enterprise intelligent power technical fields, and in particular to a kind of based on the short-term of VMD-PSO-BPNN
Power load forecasting module method for building up.
Background technique
There are large number of intermittently electrical equipments for pulping and papermaking processes, by optimizing the booting of these interval electrical equipments, shutting down
Plan can effectively realize electricity consumption peak load shifting, reduce production cost.Prediction master of the enterprises in pulp and paper industry to electric power demand side at present
Will be by historical experience power purchase, often there is superfluous or insufficient problem in what this caused to be purchased.Therefore, how negative to business electrical
Lotus is predicted, and then realization Optimal Scheduling and reasonable power purchase are one for improving papermaking enterprise economic benefit enhancing steady production
Important channel.
Over nearly 3 years, main research direction first is that using split reconstruct method electric load is predicted.By
In the electricity consumption aperiodicity of papermaking enterprise, and load fluctuation frequency is high, can be split into high-frequency waveform using VMD several
The characteristics of different and frequency lower waveform.And compared to EMD (classical mode decomposition), VMD algorithm is a kind of more accurate
Mathematical algorithm, it is and all very sensitive to noise and sample frequency, therefore be quite suitable for the fractionation of papermaking enterprise power load.
Neural network model selects to be suitble to solve the problems, such as corresponding structural parameters according to different actual conditions.But work as
When problem to be solved is relatively complicated, the need less than practical application are typically up to basic artificial neural network (ANN)
It asks, the combination algorithm that Neural Network Optimization is carried out based on optimization algorithm is one of solution to the problems described above.
BPNN (BP neural network) is a kind of for non-linear, aperiodic, irregular, amorphousness or half structure data
The most frequently used, the optimal model of effect is modeled, combined data, which is excavated to establish, has the BPNN of time series feature to establish papermaking enterprise
Industry electro-load forecast model is very practical.
PSO (particle swarm optimization algorithm) is the global optimization approach of a kind of probabilistic type.The advantages of non deterministic algorithm, is to calculate
Method can have more chances to solve globally optimal solution.The problems such as predicting there are over-fitting due to BPNN and fall into local optimum, because
This uses the weight and threshold value of PSO algorithm optimization BPNN, and it is very suitable for preventing problem above.
Summary of the invention
The purpose of the present invention is to solve drawbacks described above in the prior art, provide a kind of based on VMD-PSO-BPNN's
Short-term Load Forecasting method for building up decomposes papermaking enterprise power load using VMD decomposition algorithm, and utilization is stagnant
The input of correlation method selection prediction model afterwards, is finally optimized using weight and threshold value of the PSO to BPNN, establishes power load
Lotus prediction model is predicted and is evaluated prediction effect, so as to accurately predict papermaking enterprise power load.
The purpose of the present invention can be reached by adopting the following technical scheme that:
A kind of Short-term Load Forecasting method for building up based on VMD-PSO-BPNN, the short-term electric load prediction
Model be applied to papermaking enterprise load forecast, the method for building up the following steps are included:
S1, the electricity consumption data for obtaining papermaking enterprise data;
S2, using VMD decomposition algorithm, sequence decomposition is carried out to pretreated load sequence;
S3, input variable is chosen to each Decomposition Sequence using lag correlation method;
The weight and threshold value of S4, the hidden layer neuron number that BPNN network is set and BPNN network, Decomposition Sequence
Training set inputs in initial BPNN network, using residual error between fitting result and actual result as fitness value, is calculated using PSO
Method updates the size of weight and threshold value, finds optimal fitting result, optimal fitting result is corresponded to BPNN network and is exported,
Decomposition Sequence is predicted using trained BPNN network, the prediction result of all Decomposition Sequences is overlapped, is obtained
Short-term Load Forecasting.
Further, divide varying model optimal solution by search constraint in the step S2 to realize signal adaptive point
Solution, resolves into a series of modal components with sparse characteristic for original loads sequence, i.e., original series is decomposed into different frequencies
The sequence of rate, specifically includes:
S201, for each mode u (t), associated analytic signal, calculation formula are calculated by Hilbert transform
It is as follows:
In formula, H (t) is mode analytic signal, and δ (t) is dirac distribution, and t is sampling time point, and j is imaginary number, and * is indicated
Convolution;
S202, to the centre frequency ω of each mode analytic signal pre-estimationkIt is mixed, by the spectrum modulation of each mode
To corresponding Base Band, formula is as follows:
The gradient square L of Base Band in S203, calculating formula (2)2Norm estimates the bandwidth of each modal components, corresponding
Constraint variation model are as follows:
In formula, f (t)=∑ku(t);
S204, a unconstrained problem is obtained using secondary penalty item and Lagrange multiplier operator, finally solved
The formula of the problem are as follows:
In formula, { uk}={ u1,u2,···,ukRepresent the k IMF component for decomposing and obtaining, { ωk}={ ω1,
ω2,···,ωkIndicate the centre frequency of each component, ∑KIndicate each modal components summation, λ (t) is Lagrange's multiplier, α
It is the balance parameters of data fidelity constraint, f (t) is original signal.
Further, in the step S3, by the lag autocorrelation method power load of finding out over to current power
The influence of load, using the guidance of auto-correlation function alternatively information characteristics subset, i.e., by autocorrelative lag order come
Input variable is chosen, when the absolute value for lagging auto-correlation coefficient is greater than 0.8, is made with this lag moment corresponding effective power
For the input of model, expression formula are as follows:
In formula,It is the average value of all X in given time sequence, X={ Xt: t ∈ T }, it is time series data collection, rk
For the linear dependence of time of measuring t and t-k time series.
Further, the step S4 includes:
S401, the data set that outputs and inputs after data processing is divided into training set and test set;
S402, the weight and threshold value for initializing BPNN, and BPNN parameter is set, wherein BPNN parameter includes: to input layer by layer
Number, the hidden layer number of plies, the number of plies of output layer, frequency of training, training objective and learning rate;
S403, initialization PSO optimization algorithm parameter, PSO optimization algorithm parameter includes: Studying factors c1And c2, population rule
Mould N, inertia weight wmaxAnd wmin, maximum number of iterations and maximum speed Vmax;
S404, setting current iteration number i=1;
S405, training BPNN, and the fitness of each particle is calculated, the size of fitness is ranked up, each grain
Local optimum of the son as current population, is denoted as Ppbest, using the smallest particle of fitness in population as global optimum, it is denoted as
Pgbest;
The speed of S406, more new particle and position;
S407, judge whether to meet the condition terminated, if conditions are not met, the number of iterations i=i+1, and re-execute the steps
S405- step S407, until meeting termination condition;
S408, final weight and threshold value are exported to BPNN, obtains PSO-BPNN model;
The input variable data of the good test set of S409, input processing, are predicted by PSO-BPNN model, and output is pre-
Survey load curve.
Further, each layer of the state only under the influence of one layer of neuron state, wherein hidden layer and output
The excitation function of layer chooses hyperbolic sine Sigmoid function, and learning algorithm selects momentum gradient descent algorithm function traingdm.
Further, the speed of the n-th dimension of kth time iteration particle i and the update of position are become in the step S406
It is as follows to change formula, wherein 1≤n≤N:
In formula:It is the n-th dimension component of kth time iteration particle i position vector, xi=(xi1, xi2, xiN), model
It encloses and is limited to [Xmin,n, Xmax,n] in,It is the n-th dimension component of kth time iteration particle i flight velocity vector, Vi=(vi1,
vi2,…,viN), range is limited to [- Vmax,n, Vmax,n] in, pbestinRepresent the n-th dimension component itself warp of particle i position vector
The desired positions gone through, particle i itself undergo best position to be denoted as, pbesti=(pbesti1,pbesti2,…,
pbestin), gbestnThe desired positions that all particles of the n-th dimension component of group live through are represented, all particles of group pass through
Desired positions be denoted as, gbest=(gbest1, gbest2, gbestn), c1、c2It is acceleration constant, adjusts study
Maximum step-length, rand () are random functions, and value range [0,1], to increase search randomness, ω is inertial factor, non-negative
Number.
Further, in the step S405 Select Error as fitness.
Further, the method for building up further includes model evaluation step:
S5, prediction effect analysis is carried out to the prediction result of prediction model, process is as follows:
S501, the predicted value for the power load that prediction model obtains and actual value are compared;
S502, the analysis that forecast result of model is carried out according to the evaluation index of prediction model prediction effect;
Wherein, the evaluation index of forecast result of model includes root-mean-square error quadratic sum RMSE, average absolute percentage error
MAPE and relative error percentage RE, the expression formula of the root-mean-square error quadratic sum RMSE are as follows:
In formula, ypiFor predicted value, yoiFor actual value, i indicates that sampled point, n indicate sampling sum;
The expression formula of the average absolute percentage error MAPE are as follows:
The expression formula of the relative error percentage are as follows:
RE=yoi-yPi/yoi× 100% (10).
The present invention has the following advantages and effects with respect to the prior art:
(1) series of steps such as the data prediction that the present invention takes and modeling are ten divisions for universal papermaking enterprise
Suitable method;
(2) VMD decomposition algorithm can decompose the high-frequency power load of papermaking enterprise in method for establishing model of the present invention
For more gentle load sequence;
(3) present invention selects input variable using auto-correlation lag order, always effectively negative with papermaking enterprise compared to selection
The electricity consumption data of the high electrical equipment of the lotus degree of correlation reduces data acquisition, pretreated workload, and institute as input variable
Select influence of the variable gone to total Payload bigger;
(4) the VMD-PSO-BPNN model established in method for establishing model of the present invention, prediction accuracy meet technique requirement,
And feature of the prediction without lag, be conducive to the timely Arrange Production Schedule of papermaking enterprise and normal operation.
Detailed description of the invention
Fig. 1 is the Short-term Load Forecasting method for building up process disclosed by the invention based on VMD-PSO-BPNN
Figure;
Fig. 2 is the power load figure in the embodiment of the present invention after data prediction;
Fig. 3 is the Decomposition Sequence figure after being decomposed in the embodiment of the present invention by VMD decomposition model, wherein Fig. 3 (a) is logical
First Decomposition Sequence schematic diagram that VMD decomposition model obtains is crossed, Fig. 3 (b) is the second point obtained by VMD decomposition model
Solution sequence schematic diagram, Fig. 3 (c) are the third Decomposition Sequence schematic diagrames obtained by VMD decomposition model, and Fig. 3 (d) is to pass through
The 4th Decomposition Sequence schematic diagram that VMD decomposition model obtains;
Fig. 4 is that VMD-PSO-BPNN prediction model analyzes the prediction effect of each Decomposition Sequence in the embodiment of the present invention
Figure, wherein Fig. 4 (a) is the corresponding prediction effect analysis chart of first Decomposition Sequence, and Fig. 4 (b) is that second Decomposition Sequence is corresponding
Prediction effect analysis chart, Fig. 4 (c) is the corresponding prediction effect analysis chart of third Decomposition Sequence, and Fig. 4 (d) is four point
The corresponding prediction effect analysis chart of solution sequence;
Fig. 5 is final prediction effect analysis chart in the embodiment of the present invention;
Fig. 6 is the relative error chart of percentage comparison of VMD-PSO-BPNN prediction model in the embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Embodiment one
Present embodiment discloses a kind of Short-term Load Forecasting Model method for building up based on VMD-PSO-BPNN, using fractionation
It reconstructs and the power load trend of following 1 hour is predicted, as shown in Figure 1, should the short-term electricity based on VMD-PSO-BPNN
Power load forecasting model method for building up, includes the following steps:
S1, the electricity consumption data for obtaining papermaking enterprise.
The history electricity consumption data saved using papermaking enterprise energy management system historical data base obtains the two of them moon
Power load data.
S2, using VMD (variation mode decomposition) decomposition algorithm, sequence decomposition is carried out to pretreated load sequence.
Wherein, VMD decomposition algorithm can divide varying model optimal solution by search constraint to realize that signal adaptive decomposes, and incite somebody to action
Original loads sequence resolves into a series of modal components with sparse characteristic, i.e., original series are decomposed into the sequence of different frequency
Column, algorithm steps are as follows:
S201, for each mode u (t), associated analytic signal, calculation formula are calculated by Hilbert transform
It is as follows:
In formula, H (t) is mode analytic signal, and δ (t) is dirac distribution, and t is sampling time point, and j is imaginary number, and * is indicated
Convolution;
S202, to the centre frequency ω of each mode analytic signal pre-estimationkIt is mixed, by the spectrum modulation of each mode
To corresponding Base Band, formula is as follows:
The gradient square L of Base Band in S203, calculating formula (2)2Norm estimates the bandwidth of each modal components, corresponding
Constraint variation model are as follows:
In formula, f (t)=∑ku(t);
S204, a unconstrained problem is obtained using secondary penalty item and Lagrange multiplier operator, finally solved
The formula of the problem are as follows:
In formula, { uk}={ u1,u2,···,ukRepresent the k IMF component for decomposing and obtaining, { ωk}={ ω1,
ω2,···,ωkIndicate the centre frequency of each component, ∑KIndicate each modal components summation, λ (t) is Lagrange's multiplier, α
It is the balance parameters of data fidelity constraint, f (t) is original signal.
S3, input variable is chosen to each Decomposition Sequence using lag correlation method.
Influence by the lag autocorrelation method power load of finding out over to current power load, uses auto-correlation herein
The guidance of function (ACF) alternatively information characteristics subset, i.e., choose input variable by autocorrelative lag order.When stagnant
When the absolute value of auto-correlation coefficient is greater than 0.8 afterwards, use this lag moment corresponding effective power as the input of model.It is counted
Learn expression formula are as follows:
In formula,It is the average value of all X in given time sequence, X={ Xt: t ∈ T }, it is time series data collection, rk
For the linear dependence of time of measuring t and t-k time series.
The weight and threshold value of S4, the hidden layer neuron number that BPNN network is set and BPNN network, Decomposition Sequence
Training set inputs in initial BPNN network, using residual error between fitting result and actual result as fitness value, is calculated using PSO
Method updates the size of weight and threshold value, finds optimal fitting result, optimal fitting result is corresponded to BPNN network and is exported,
Decomposition Sequence is predicted using trained BPNN network, the prediction result of all Decomposition Sequences is overlapped, is obtained
Short-term Load Forecasting.
It is optimized by weight and threshold value of PSO (population) optimization algorithm to BPNN, prevents single BPNN from occurring
The case where over-fitting, specific algorithm steps are as follows:
S401, the data set that outputs and inputs after data processing is divided into training set and test set;
S402, the weight and threshold value for initializing BPNN, and it (includes: the input layer number of plies, the hidden layer number of plies, defeated that parameter, which is arranged,
The number of plies of layer, frequency of training, training objective, learning rate out);
S403, the parameter for initializing PSO optimization algorithm (include: c1、c2Equal Studying factors;Population scale N;Inertia weight
wmaxAnd wmin;Maximum number of iterations;Maximum speed Vmax);
S404, setting current iteration number i=1
S405, training BPNN, and the fitness (this paper Select Error is as fitness) of each particle is calculated, to fitness
Size be ranked up, using each particle as the local optimum of current population, be denoted as Ppbest, fitness in population is minimum
Particle as global optimum, be denoted as Pgbest;
S406, speed and position according to formula (6) and formula (7) more new particle;
S407, judge whether to meet the condition terminated, if conditions are not met, the number of iterations i=i+1, and re-execute the steps
S405- step S407, until meeting termination condition;
S408, final weight and threshold value are exported to BPNN, obtains PSO-BPNN model;
The input variable data of the good test set of S409, input processing, are predicted by PSO-BPNN model, and output is pre-
Survey load curve.
Wherein each layer of BPNN of state only under the influence of one layer of neuron state, the wherein excitation of hidden layer and output layer
Function all chooses hyperbolic sine Sigmoid function, and learning algorithm selects momentum gradient descent algorithm function traingdm.
PSO algorithm therein is the fitness situation based on group to environment, the individual in group is moved to best
A kind of algorithm in region.It regards each individual as a particle (no volume) in search space (being assumed to be N-dimensional),
It is flown in search space with certain speed (speed is adjusted according to own experience and social experience), until finding most
Good flight path terminates.It is as follows to the speed and change in location formula of the n-th dimension (1≤n≤N) of kth time iteration particle i:
In formula:It is the n-th dimension component of kth time iteration particle i position vector, xi=(xi1, xi2, xiN), model
It encloses and is limited to [Xmin,n, Xmax,n] in,It is the n-th dimension component of kth time iteration particle i flight velocity vector, Vi=(vi1,
vi2,…,viN), range is limited to [- Vmax,n, Vmax,n] in, pbestinRepresent the n-th dimension component itself warp of particle i position vector
The desired positions gone through, particle i itself undergo best position to be denoted as, pbesti=(pbesti1,pbesti2,…,
pbestin), gbestnThe desired positions that all particles of the n-th dimension component of group live through are represented, all particles of group pass through
Desired positions be denoted as, gbest=(gbest1, gbest2, gbestn), c1、c2It is acceleration constant, adjusts study
Maximum step-length, rand () are random functions, and value range [0,1], to increase search randomness, ω is inertial factor, non-negative
Number.
S5, prediction effect analysis is carried out to the prediction result of prediction model.
S501, the predicted value for the power load that prediction model obtains and actual value are compared;
S502, the analysis that forecast result of model is carried out according to the evaluation index of VMD-PSO-BPNN forecast result of model;
Wherein, the evaluation index of forecast result of model includes root-mean-square error quadratic sum RMSE, average absolute percentage error
MAPE and relative error percentage RE;
The expression formula of root-mean-square error quadratic sum RMSE is as follows:
In formula, ypiFor predicted value, yoiFor actual value;I indicates that sampled point, n indicate sampling sum
The expression formula of average absolute percentage error MAPE are as follows:
The expression formula of relative error percentage are as follows:
RE=yoi-yPi/yoi× 100% (10).
Embodiment two
A kind of method for building up of the papermaking enterprise Short-term Load Forecasting based on VMD-PSO-BPNN, comprising following
Modeling and model evaluation step:
1, papermaking enterprise electricity consumption data is obtained by the EMS (energy management system) from certain papermaking enterprise, including
All data in March, 2018~2018 year May, frequency acquisition 1h, as shown in Fig. 2, using before data 75% as instruction
Practice collection, rear 25% is used as test set;
2, the data of training set are split by VMD decomposition model, wherein modal characteristics select 3, i.e., electricity consumption
Load decomposition is 3 sequences, and the 4th sequence is by the noise having in power load, and decomposition result is respectively such as Fig. 3 (a)~Fig. 3
(d) shown in.
3, input variable is chosen by lag autocorrelation method, the input variable of each Decomposition Sequence is as shown in table 1.
The each mode input argument table for decomposing load sequence of table 1.
In table, It-i,jIt is the load of the preceding i sampled point of t-th of sampling time point of j-th of Decomposition Sequence;
4, Decomposition Sequence is trained and is predicted by PSO-BPNN algorithm, wherein the selection of model parameter are as follows: study
Factor c1、c2=2;Maximum speed vmax=0.5;Population number N=30;The number of iterations=100;Hiding the number of plies is 2n+1, and (n is defeated
Enter variables number);Exporting the number of plies is 1;The input number of plies is n.The prediction result of 4 Decomposition Sequences is respectively such as Fig. 4 (a)~Fig. 4 (d)
It is shown.
5, the prediction result of each Decomposition Sequence is overlapped, obtains final prediction result as shown in figure 5, average exhausted
To percent error as shown in fig. 6, prediction result table is as shown in table 2.
2. prediction result of table
Evaluation index | MAPE (%) | RMSE(kWh) |
Case study on implementation result | 2.62 | 366.45 |
The present invention occurs the feelings of unstable (non-programmed halt) according to papermaking enterprise aperiodicity, there are steady production and easily
Condition is proposed after being decomposed using VMD decomposition algorithm, and the method for establishing prediction model by the BPNN that PSO optimizes is extracted different
With electrical feature, predicted respectively, finally carry out prediction superposition, the relative error of 95% or more discovery [- 10%, 10%] with
It is interior, the problem of meeting industrial requirements, and lagged without prediction.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (8)
1. a kind of Short-term Load Forecasting method for building up based on VMD-PSO-BPNN, the short-term electric load prediction mould
Type is applied to the load forecast of papermaking enterprise, which is characterized in that the method for building up the following steps are included:
S1, the electricity consumption data for obtaining papermaking enterprise data;
S2, using VMD decomposition algorithm, sequence decomposition is carried out to pretreated load sequence;
S3, input variable is chosen to each Decomposition Sequence using lag correlation method;
The weight and threshold value of S4, the hidden layer neuron number that BPNN network is set and BPNN network, the training of Decomposition Sequence
Collection inputs in initial BPNN network, using residual error between fitting result and actual result as fitness value, more using PSO algorithm
The size of new weight and threshold value, finds optimal fitting result, optimal fitting result is corresponded to BPNN network and is exported, and utilizes
Trained BPNN network predicts Decomposition Sequence, and the prediction result of all Decomposition Sequences is overlapped, and obtains short-term
Power load forecasting module.
2. the Short-term Load Forecasting method for building up according to claim 1 based on VMD-PSO-BPNN, special
Sign is, divides varying model optimal solution to realize that signal adaptive decomposes, by original minus by search constraint in the step S2
Lotus sequence resolves into a series of modal components with sparse characteristic, i.e., original series are decomposed into the sequence of different frequency, tool
Body includes:
S201, for each mode u (t), associated analytic signal is calculated by Hilbert transform, calculation formula is such as
Under:
In formula, H (t) is mode analytic signal, and δ (t) is dirac distribution, and t is sampling time point, and j is imaginary number, and * indicates convolution;
S202, to the centre frequency ω of each mode analytic signal pre-estimationkIt is mixed, by the spectrum modulation of each mode to phase
The Base Band answered, formula are as follows:
The gradient square L of Base Band in S203, calculating formula (2)2Norm, estimates the bandwidth of each modal components, and corresponding constraint becomes
Sub-model are as follows:
In formula, f (t)=∑ku(t);
S204, a unconstrained problem is obtained using secondary penalty item and Lagrange multiplier operator, finally solves this and asks
The formula of topic are as follows:
In formula, { uk}={ u1,u2,···,ukRepresent the k IMF component for decomposing and obtaining, { ωk}={ ω1,ω2,···,
ωkIndicate the centre frequency of each component, ∑KIndicate each modal components summation, λ (t) is Lagrange's multiplier, and α is data fidelity
The balance parameters of constraint, f (t) are original signal.
3. the Short-term Load Forecasting method for building up according to claim 1 based on VMD-PSO-BPNN, special
Sign is, in the step S3, by lagging influence of the autocorrelation method power load of finding out over to current power load,
Using the guidance of auto-correlation function alternatively information characteristics subset, i.e., input is chosen by autocorrelative lag order and is become
Amount uses this lag moment corresponding effective power as the defeated of model when the absolute value for lagging auto-correlation coefficient is greater than 0.8
Enter, expression formula are as follows:
In formula,It is the average value of all X in given time sequence, X={ Xt: t ∈ T }, it is time series data collection, rkFor measurement
The linear dependence of time t and t-k time series.
4. the Short-term Load Forecasting method for building up according to claim 1 based on VMD-PSO-BPNN, special
Sign is that the step S4 includes:
S401, the data set that outputs and inputs after data processing is divided into training set and test set;
S402, the weight and threshold value for initializing BPNN, and are arranged BPNN parameter, wherein BPNN parameter include: the input layer number of plies, it is hidden
Hide number, the number of plies of output layer, frequency of training, training objective and learning rate layer by layer;
S403, initialization PSO optimization algorithm parameter, PSO optimization algorithm parameter includes: Studying factors c1And c2, population scale N,
Inertia weight wmaxAnd wmin, maximum number of iterations and maximum speed Vmax;
S404, setting current iteration number i=1;
S405, training BPNN, and the fitness of each particle is calculated, the size of fitness is ranked up, each particle is made
For the local optimum of current population, it is denoted as Ppbest, using the smallest particle of fitness in population as global optimum, it is denoted as
Pgbest;
The speed of S406, more new particle and position;
S407, judge whether to meet the condition terminated, if conditions are not met, the number of iterations i=i+1, and it re-execute the steps S405-
Step S406, until meeting termination condition;
S408, final weight and threshold value are exported to BPNN, obtains PSO-BPNN model;
The input variable data of the good test set of S409, input processing, are predicted by PSO-BPNN model, and output prediction is negative
Lotus curve.
5. the Short-term Load Forecasting method for building up according to claim 4 based on VMD-PSO-BPNN, special
Sign is, in the BPNN each layer of state only under the influence of one layer of neuron state, wherein hidden layer and output layer
Excitation function chooses hyperbolic sine Sigmoid function, and learning algorithm selects momentum gradient descent algorithm function traingdm.
6. the Short-term Load Forecasting method for building up according to claim 4 based on VMD-PSO-BPNN, special
Sign is, as follows to the speed of the n-th dimension of kth time iteration particle i and the more new change formula of position in the step S406,
Wherein, 1≤n≤N:
In formula:It is the n-th dimension component of kth time iteration particle i position vector, xi=(xi1, xi2, xiN), range limit
It is scheduled on [Xmin,n, Xmax,n] in,It is the n-th dimension component of kth time iteration particle i flight velocity vector, Vi=(vi1,vi2,…,
viN), range is limited to [- Vmax,n, Vmax,n] in, pbestinIt represents the n-th of particle i position vector and ties up what component itself lived through
Desired positions, particle i itself undergo best position to be denoted as, pbesti=(pbesti1,pbesti2,…,pbestin), gbestn
The desired positions that all particles of the n-th dimension component of group live through are represented, the desired positions that all particles of group pass through are denoted as,
Gbest=(gbest1, gbest2, gbestn), c1、c2It is acceleration constant, adjusts study maximum step-length, rand ()
It is random function, value range [0,1], to increase search randomness, ω is inertial factor, nonnegative number.
7. the Short-term Load Forecasting method for building up according to claim 4 based on VMD-PSO-BPNN, special
Sign is that Select Error is as fitness in the step S405.
8. the Short-term Load Forecasting method for building up according to claim 1 based on VMD-PSO-BPNN, special
Sign is that the method for building up further includes model evaluation step:
S5, prediction effect analysis is carried out to the prediction result of prediction model, process is as follows:
S501, the predicted value for the power load that prediction model obtains and actual value are compared;
S502, the analysis that forecast result of model is carried out according to the evaluation index of prediction model prediction effect;
Wherein, the evaluation index of forecast result of model includes root-mean-square error quadratic sum RMSE, average absolute percentage error MAPE
And relative error percentage RE, the expression formula of the root-mean-square error quadratic sum RMSE are as follows:
In formula, ypiFor predicted value, yoiFor actual value, i indicates that sampled point, n indicate sampling sum;
The expression formula of the average absolute percentage error MAPE are as follows:
The expression formula of the relative error percentage are as follows:
RE=yoi-yPi/yoi× 100% (10).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910060863.0A CN109919421A (en) | 2019-01-23 | 2019-01-23 | Short-term power load prediction model establishment method based on VMD-PSO-BPNN |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910060863.0A CN109919421A (en) | 2019-01-23 | 2019-01-23 | Short-term power load prediction model establishment method based on VMD-PSO-BPNN |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109919421A true CN109919421A (en) | 2019-06-21 |
Family
ID=66960476
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910060863.0A Pending CN109919421A (en) | 2019-01-23 | 2019-01-23 | Short-term power load prediction model establishment method based on VMD-PSO-BPNN |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109919421A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110929809A (en) * | 2019-12-14 | 2020-03-27 | 北京工业大学 | Soft measurement method for key water quality index of sewage by using characteristic self-enhanced circulating neural network |
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 |
CN111476402A (en) * | 2020-03-16 | 2020-07-31 | 云南电网有限责任公司 | Wind power generation capacity prediction method coupling meteorological information and EMD technology |
CN112348271A (en) * | 2020-11-12 | 2021-02-09 | 华北电力大学 | Short-term photovoltaic power prediction method based on VMD-IPSO-GRU |
CN112883886A (en) * | 2021-03-01 | 2021-06-01 | 兰州理工大学 | Parallel accelerated VMD-SVPSO-BP neural network fault diagnosis method |
CN113723752A (en) * | 2021-07-26 | 2021-11-30 | 国网江苏省电力有限公司江阴市供电分公司 | Decomposition algorithm performance evaluation method in combined wind power prediction model |
CN114492938A (en) * | 2021-12-29 | 2022-05-13 | 中国大唐集团科学技术研究总院有限公司华东电力试验研究院 | Coal mill fault early warning method and system based on BPNN model and adaptive threshold |
CN116663863A (en) * | 2023-07-28 | 2023-08-29 | 石家庄科林电气股份有限公司 | Virtual power plant load prediction method based on scheduling plan |
-
2019
- 2019-01-23 CN CN201910060863.0A patent/CN109919421A/en active Pending
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN110929809A (en) * | 2019-12-14 | 2020-03-27 | 北京工业大学 | Soft measurement method for key water quality index of sewage by using characteristic self-enhanced circulating neural network |
CN110929809B (en) * | 2019-12-14 | 2023-05-02 | 北京工业大学 | Sewage key water quality index soft measurement method of characteristic self-enhanced cyclic neural network |
CN111476402A (en) * | 2020-03-16 | 2020-07-31 | 云南电网有限责任公司 | Wind power generation capacity prediction method coupling meteorological information and EMD technology |
CN112348271A (en) * | 2020-11-12 | 2021-02-09 | 华北电力大学 | Short-term photovoltaic power prediction method based on VMD-IPSO-GRU |
CN112348271B (en) * | 2020-11-12 | 2024-01-30 | 华北电力大学 | Short-term photovoltaic power prediction method based on VMD-IPSO-GRU |
CN112883886A (en) * | 2021-03-01 | 2021-06-01 | 兰州理工大学 | Parallel accelerated VMD-SVPSO-BP neural network fault diagnosis method |
CN113723752A (en) * | 2021-07-26 | 2021-11-30 | 国网江苏省电力有限公司江阴市供电分公司 | Decomposition algorithm performance evaluation method in combined wind power prediction model |
CN114492938A (en) * | 2021-12-29 | 2022-05-13 | 中国大唐集团科学技术研究总院有限公司华东电力试验研究院 | Coal mill fault early warning method and system based on BPNN model and adaptive threshold |
CN116663863A (en) * | 2023-07-28 | 2023-08-29 | 石家庄科林电气股份有限公司 | Virtual power plant load prediction method based on scheduling plan |
CN116663863B (en) * | 2023-07-28 | 2023-10-20 | 石家庄科林电气股份有限公司 | Virtual power plant load prediction method based on scheduling plan |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109919421A (en) | Short-term power load prediction model establishment method based on VMD-PSO-BPNN | |
Lin et al. | An improved moth-flame optimization algorithm for support vector machine prediction of photovoltaic power generation | |
Behera et al. | Solar photovoltaic power forecasting using optimized modified extreme learning machine technique | |
Ren et al. | Optimal parameters selection for BP neural network based on particle swarm optimization: A case study of wind speed forecasting | |
Aly | An intelligent hybrid model of neuro Wavelet, time series and Recurrent Kalman Filter for wind speed forecasting | |
Rahman et al. | Machine learning with big data an efficient electricity generation forecasting system | |
CN109657882A (en) | Short-term power load prediction model establishment method based on VMD-PSO-LSSVM | |
Shao et al. | Nickel price forecast based on the LSTM neural network optimized by the improved PSO algorithm | |
CN108229754A (en) | Short-term load forecasting method based on similar day segmentation and LM-BP networks | |
CN105631483A (en) | Method and device for predicting short-term power load | |
CN110009125A (en) | Short-term power load prediction model establishment method based on EMD-VMD-PSO-BPNN | |
CN109711755A (en) | Short-term power load prediction model establishment method based on EMD-VMD-PSO-LSSVM | |
CN110443417A (en) | Multiple-model integration load forecasting method based on wavelet transformation | |
CN110309603A (en) | A kind of short-term wind speed forecasting method and system based on wind speed characteristics | |
CN108090563A (en) | A kind of electric flux Forecasting Methodology based on BP neural network | |
Rizwan et al. | Artificial intelligence based approach for short term load forecasting for selected feeders at madina saudi arabia | |
CN110119846A (en) | A kind of multiple target deepness belief network for mid-term electric load forecasting | |
Koo et al. | Comparative study of short-term electric load forecasting | |
CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
CN115099511A (en) | Photovoltaic power probability estimation method and system based on optimized copula | |
Tavares et al. | Comparison of PV power generation forecasting in a residential building using ANN and DNN | |
Putri et al. | Short-term forecasting of electricity consumption revenue on Java-Bali electricity system using Jordan recurrent neural network | |
Xu et al. | Accelerated and adaptive power scheduling for more electric aircraft via hybrid learning | |
Wang et al. | Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises | |
CN116307111A (en) | Reactive load prediction method based on K-means clustering and random forest algorithm |
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 | ||
TA01 | Transfer of patent application right | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20220509 Address after: 510000 room 1021, No.2, Tengfei 1st Street, Zhongxin Guangzhou Knowledge City, Huangpu District, Guangzhou City, Guangdong Province Applicant after: GUANGZHOU POI INTELLIGENT INFORMATION TECHNOLOGY CO.,LTD. Address before: 510640 No. five, 381 mountain road, Guangzhou, Guangdong, Tianhe District Applicant before: SOUTH CHINA University OF TECHNOLOGY |