CN107766995A - Power-system short-term load forecasting method based on depth recurrent neural network - Google Patents
Power-system short-term load forecasting method based on depth recurrent neural network Download PDFInfo
- Publication number
- CN107766995A CN107766995A CN201711293583.1A CN201711293583A CN107766995A CN 107766995 A CN107766995 A CN 107766995A CN 201711293583 A CN201711293583 A CN 201711293583A CN 107766995 A CN107766995 A CN 107766995A
- Authority
- CN
- China
- Prior art keywords
- neural network
- recurrent neural
- data
- load
- layer
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 29
- 230000000306 recurrent effect Effects 0.000 title claims abstract description 21
- 238000013277 forecasting method Methods 0.000 title claims abstract description 12
- 230000006870 function Effects 0.000 claims abstract description 14
- 230000002159 abnormal effect Effects 0.000 claims abstract 3
- 238000000034 method Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims 2
- 238000000151 deposition Methods 0.000 claims 1
- 239000002245 particle Substances 0.000 abstract description 41
- 238000005457 optimization Methods 0.000 abstract description 7
- 238000012549 training Methods 0.000 description 9
- 230000006872 improvement Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 230000007547 defect Effects 0.000 description 5
- 210000004027 cell Anatomy 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000008713 feedback mechanism Effects 0.000 description 2
- 230000007087 memory ability Effects 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 230000007935 neutral effect Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 229920006395 saturated elastomer Polymers 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 235000006508 Nelumbo nucifera Nutrition 0.000 description 1
- 240000002853 Nelumbo nucifera Species 0.000 description 1
- 235000006510 Nelumbo pentapetala Nutrition 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003471 mutagenic agent Substances 0.000 description 1
- 231100000707 mutagenic chemical Toxicity 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to power-system short-term load forecasting technical field, discloses the power-system short-term load forecasting method based on depth recurrent neural network, and step is：（1）Collection collects history network load and meteorological data, and it is standby to build storehouse；（2）Removal step（1）Gained abnormal data, to remaining data normalized；（3）It is determined that the model structure with feedforward and feedback function；（4）The DRNN forecast models based on IPSO algorithms are trained using historical data；（5）DRNN forecast models based on IPSO algorithms are used in the prediction of actual load.The technical program depth recurrent neural network short-term load forecasting method sets up associated layers on the basis of the more hidden layer configurations of deep neural network, and to improve particle cluster algorithm as the optimized learning algorithm of network, depth optimization is carried out to model weights space.Error effectively reduces, and can merge feedforward and feedback link, improves network generalization, effectively improves load prediction precision.
Description
Technical field
The present invention relates to power-system short-term load forecasting technical field, more particularly to based on depth recurrent neural network
Power-system short-term load forecasting method.
Background technology
As power system scale and complexity improve constantly, power-system short-term load forecasting it is whether accurate to effective
Reduce generating rate, implementation each department electric power system optimization control has key effect.It is short-term negative compared with long term load forecasting
Lotus prediction is mainly used in arranging generation schedule, ageing highest.Its load change speed is fast, by Mutagens such as the temperature difference, humidity
Influence greatly, to belong to kinematic nonlinearity time series.Due to this category feature of short term, if it is more tired to want to reach precisely prediction
It is difficult.With the implementation that new electricity changes, power sales competition is deepened constantly, and new requirement is proposed to precision of prediction.It is therefore it provides a kind of
The high power-system short-term load forecasting method of accuracy is necessary.
DNN --- deep neural network；
DRNN --- depth recurrent neural network；
PSO (Particle Swarm Optimization) --- particle cluster algorithm；
IPSO --- improve particle cluster algorithm.
The content of the invention
It is an object of the invention to for above-mentioned technical problem, there is provided a kind of power train based on depth recurrent neural network
System short-term load forecasting method.
Step 1：The collection for collecting the data such as historical region network load data, meteorological data is collected with collecting, and is imported
In Excel databases.
Step 2：Data prediction, to avoid the generation of neuron saturated conditions, it is necessary to be carried out to original loads data pre-
Processing, the convergence of training process is so beneficial to, improves precision of prediction.Main pretreatment mode is, to training sample
The historical load data of concentration, its maximum and minimum value are counted, load data is normalized to [- 1,1] section, can be made at data
In same number of levels, accelerate neutral net convergence.
Step 3：Determine model structure.
DNN (deep neural network) has more hidden layer configurations, and Multiple Training is carried out repeatedly to carry to the input vector of network
Rise the accuracy of classification or prediction.DNN forecast models are made up of input layer, more hidden layers and output layer.With conventional feed forward nerve
Network is compared, and DNN has more hidden layer configurations.X is network inputs, is the column vector for including m dimensions；(W, B) be each hidden layer it
Between weight matrix and threshold matrix.DNN each hidden layer obtains input vector from its preceding layer, utilizes the hidden layer
Activation primitive carries out nonlinear transformation, then is transmitted to next layer of neuron using obtained vector as input, successively reciprocal iteration, most
Network output y is passed to eventually.Compared with BP networks, the DNN training depth for having multiple hidden layers substantially increases, and learning ability significantly increases
By force, the defects of traditional BP neural network can be overcome.
Though DNN learning abilities are stronger, its essence is still static network, can not be portrayed comprehensively with characterizing load dynamic sequence
The rule of row.Construct the DRNN forecast models with feedback mechanism.DRNN is by input layer, n-layer hidden layer, associated layers and output layer
Deng composition.DRNN associated layers contain the memory cell for storing historical information, and memory cell is by the current time of storage
Historical information is as input of the feedback for subsequent time first layer hidden layer.DRNN is reconstructed network internal structure and state,
The output for making network final is not only information-related with current time, also closely related with the historical information at each moment, makes its tool
There is preferable dynamic memory ability.
X=[x1,x2,…xm] be DRNN input vector.In DRNN n hidden layer, every layer of node number difference
Represented with l1, l2 ..., ln, output layer node number is 1.DRNN first layer hidden layer can describe in the output vector of t
For
R1(t)=f (W1·[X(t),z(t)]+B1) (1)
In formula：R1 (t) represents the output of first layer hidden layer；W1, B1 represent the power between input layer and first layer hidden layer
Matrix；Z (t) is represented in t associated layers to the feed back input between first layer hidden layer.
Retardation z is as feedback term, in t (t>0, t ∈ Z) input at moment should be history corresponding to the output layer t-1 moment
Information.Z stores the historical information of last moment, and is used for the prediction at current time as the input quantity of hidden layer.Therefore,
Feedback term z (t) should meet
Each layer output of other hidden layers of the DRNN in t in addition to first layer can be described as
RL(t)=f (WL·RL-1(t)+BL)
(3) in formula：RL (t) represents the output vector of L layer hidden layers；WL、
BL represents the weight matrix between L-1 layers hidden layer and L layer hidden layers；F is the nonlinear activation function of DRNN hidden layers.
DRNN can be described as in the output y (t) of t output layer
Y (t)=g (Wn+1·Rn(t)+Bn+1)
(4) in formula：Wn+1, Bn+1 be n-th layer hidden layer and output layer it
Between weight matrix；G is the nonlinear activation function of DRNN output layers.
Activation primitive f, g are used uniformly sigmoid functions.The advantages of function, is to be restricted to output area rationally
Section, data are not easy to dissipate during transmission, can preferably leverage linear and it is non-linear between behavior.Sigmoid letters
Number is expressed as
DRNN has merged feedforward and feedback link, network is possessed dynamic property, can Efficient Characterization load dynamic sequence
Inherent law, excavate dynamic load characteristic.
Step 4：The DRNN forecast models based on IPSO algorithms are trained using historical data, determine model parameter
And weights.
Particle cluster algorithm (Particle Swarm Optimization, PSO) lacks speed dynamic regulation, particle is existed
Local extremum is easily trapped into during renewal oneself state.The defects of to overcome PSO, following improvement is done to it, and using improvement
IPSO afterwards is trained to DRNN.IPSO improvement is mainly：
A：In d ties up solution space, v is used in the speed of i-th particle and position respectivelyidAnd xidRepresent；Its individual extreme value and kind
Group's global extremum is pid, sd.Each particle is ranked up from big to small by fitness, by the status information of preceding n particle
Current particle state is modified.Revised speed formula is
In formula：For i-th of particle kth generation and K+1 generations speed；In K generations, the position of i-th of particle
Used respectively with individual optimal valueWithRepresent；ω is inertia weight；C1, c2 for Studying factors；r1、r2jBetween (0,1)
Random number；The position of j-th of particle is used in the particle that kth generation is selectedRepresent.
B：If particle individual extreme value no longer changes by successive ignition, random operator is introduced to change current particle
State, avoid particle from being absorbed in local extremum, can be described as
In formula：It is i-th of particle in the position in the generation of kth+1；For in kth for Selected Particles in the last grain
Sub- position；r3The random number between (0,1).
Weight between each hierarchical structure in DRNN and threshold value is corresponding with IPSO particle states, optimized using IPSO
DRNN weighs threshold value.The fitness function of particle is formula (8)
In formula：N represents test sample size；Y (i) represents the predicted value and actual value of load respectively.
Said process Chinese style (4) is depth recurrent neural networks model, and formula (5), (6), (7) and (8) is to determine weights
Process.
Step 5：DRNN forecast models based on IPSO algorithms are used in the prediction of actual load, draw prediction load
Value.
The technical program depth recurrent neural network short-term load forecasting method is in the more hidden layer configurations of deep neural network
On the basis of set up associated layers, and to improve particle cluster algorithm as the optimized learning algorithm of network, model weights space is carried out
Depth optimization.Error effectively reduces, and can merge feedforward and feedback link, improves network generalization, it is pre- to effectively improve load
Survey precision.
Brief description of the drawings
Fig. 1：DNN structure charts in the present invention；
Fig. 2：DRNN structure charts in the present invention；
Fig. 3：Weights-selected Algorithm flow chart in IPSO optimizations DRNN forecast models of the present invention.
Embodiment
With reference to accompanying drawing 1-3, the technical program is described further.
Step 1：The collection for collecting the data such as historical region network load data, meteorological data is collected with collecting, and is imported
In Excel databases.
Step 2：Data prediction, to avoid the generation of neuron saturated conditions, it is necessary to be carried out to original loads data pre-
Processing, the convergence of training process is so beneficial to, improves precision of prediction.Main pretreatment mode is, to training sample
The historical load data of concentration, its maximum and minimum value are counted, load data is normalized to [- 1,1] section, can be made at data
In same number of levels, accelerate neutral net convergence.
Step 3：Determine model structure.
DNN (deep neural network) has more hidden layer configurations, and Multiple Training is carried out repeatedly to carry to the input vector of network
Rise the accuracy of classification or prediction.DNN forecast models are made up of input layer, more hidden layers and output layer.With conventional feed forward nerve
Network is compared, and DNN has more hidden layer configurations.X is network inputs, is the column vector for including m dimensions；(W, B) be each hidden layer it
Between weight matrix and threshold matrix.DNN each hidden layer obtains input vector from its preceding layer, utilizes the hidden layer
Activation primitive carries out nonlinear transformation, then is transmitted to next layer of neuron using obtained vector as input, successively reciprocal iteration, most
Network output y is passed to eventually.Compared with BP networks, the DNN training depth for having multiple hidden layers substantially increases, and learning ability significantly increases
By force, the defects of traditional BP neural network can be overcome.
Though DNN learning abilities are stronger, its essence is still static network, can not be portrayed comprehensively with characterizing load dynamic sequence
The rule of row.Construct the DRNN forecast models with feedback mechanism.DRNN is by input layer, n-layer hidden layer, associated layers and output layer
Deng composition.DRNN associated layers contain the memory cell for storing historical information, and memory cell is by the current time of storage
Historical information is as input of the feedback for subsequent time first layer hidden layer.DRNN is reconstructed network internal structure and state,
The output for making network final is not only information-related with current time, also closely related with the historical information at each moment, makes its tool
There is preferable dynamic memory ability.
X=[x1,x2,…xm] be DRNN input vector.In DRNN n hidden layer, every layer of node number difference
Represented with l1, l2 ..., ln, output layer node number is 1.DRNN first layer hidden layer can describe in the output vector of t
For
R1(t)=f (W1·[X(t),z(t)]+B1) (1)
In formula：R1 (t) represents the output of first layer hidden layer；W1, B1 represent the power between input layer and first layer hidden layer
Matrix；Z (t) is represented in t associated layers to the feed back input between first layer hidden layer.
Retardation z is as feedback term, in t (t>0, t ∈ Z) input at moment should be history corresponding to the output layer t-1 moment
Information.Z stores the historical information of last moment, and is used for the prediction at current time as the input quantity of hidden layer.Therefore,
Feedback term z (t) should meet
Each layer output of other hidden layers of the DRNN in t in addition to first layer can be described as
RL(t)=f (WL·RL-1(t)+BL)
(3) in formula：RL (t) represents the output vector of L layer hidden layers；WL、BL
Represent the weight matrix between L-1 layers hidden layer and L layer hidden layers；F is the nonlinear activation function of DRNN hidden layers.
DRNN can be described as in the output y (t) of t output layer
Y (t)=g (Wn+1·Rn(t)+Bn+1)
(4) in formula：Wn+1, Bn+1 are between n-th layer hidden layer and output layer
Weight matrix；G is the nonlinear activation function of DRNN output layers.
Activation primitive f, g are used uniformly sigmoid functions.The advantages of function, is to be restricted to output area rationally
Section, data are not easy to dissipate during transmission, can preferably leverage linear and it is non-linear between behavior.Sigmoid letters
Number is expressed as
DRNN has merged feedforward and feedback link, network is possessed dynamic property, can Efficient Characterization load dynamic sequence
Inherent law, excavate dynamic load characteristic.
Step 4：The DRNN forecast models based on IPSO algorithms are trained using historical data, determine model parameter
And weights.
Particle cluster algorithm (Particle Swarm Optimization, PSO) lacks speed dynamic regulation, particle is existed
Local extremum is easily trapped into during renewal oneself state.The defects of to overcome PSO, following improvement is done to it, and using improvement
IPSO afterwards is trained to DRNN.IPSO improvement is mainly：
A：In d ties up solution space, v is used in the speed of i-th particle and position respectivelyidAnd xidRepresent；Its individual extreme value and kind
Group's global extremum is pid, sd.Each particle is ranked up from big to small by fitness, by the status information of preceding n particle
Current particle state is modified.Revised speed formula is
In formula：For i-th of particle kth generation and K+1 generations speed；In K generations, the position of i-th of particle
Used respectively with individual optimal valueWithRepresent；ω is inertia weight；C1, c2 for Studying factors；r1、r2jBetween (0,1)
Random number；The position of j-th of particle is used in the particle that kth generation is selectedRepresent.
B：If particle individual extreme value no longer changes by successive ignition, random operator is introduced to change current particle
State, avoid particle from being absorbed in local extremum, can be described as
In formula：It is i-th of particle in the position in the generation of kth+1；For in kth for Selected Particles in the last grain
Sub- position；r3The random number between (0,1).
Weight between each hierarchical structure in DRNN and threshold value is corresponding with IPSO particle states, optimized using IPSO
DRNN weighs threshold value.The fitness function of particle is formula (8)
In formula：N represents test sample size；Y (i) represents the predicted value and actual value of load respectively.
Said process Chinese style (4) is depth recurrent neural networks model, and formula (5), (6), (7) and (8) is to determine weights
Process.
Step 5：DRNN forecast models based on IPSO algorithms are used in the prediction of actual load, draw prediction load
Value.
Tectonic model 1 (the BP-NN forecast models for using traditional BP algorithm), model 2 (IPSO-DNN forecast models) and mould
Type 3 (IPSO-DRNN forecast models), comparative study is carried out to it.
The power network actual load of one week in somewhere is predicted with model 1, model 2, model 3, error is as shown in table 1.
1 three kinds of forecast model prediction application conditions of table
Data in table 1 are compared, the error of model 1 of BP-NN constructions is maximum, and effect is worst；The model 2 of DNN constructions
Because training depth substantially increases, learning ability enhancing, the defects of overcoming BP networks, compared with model 1, mean absolute error
Decreased with maximum relative error, prediction effect is better than BP-NN；The model 3 of DRNN constructions sets up feedback associated layers, makes net
Network possesses good dynamic property, can characterize the inherent law of load dynamic sequence, the mean absolute error compared with model 2
Decreased with maximum relative error, prediction effect is better than DNN.DRNN has merged feedforward and feedback link, and it is dynamic to enhance it
State property energy, improve the precision of prediction of model.
By being predicted emulation to somewhere power network actual load, the results showed that：With BP networks, deep neural network phase
Than the prediction error of depth recurrent neural network effectively reduces respectively, and depth recurrent neural network can merge feedforward and feedback
Connection, network generalization is improved, effectively improves load prediction precision.
Embodiment only illustrates technical scheme, rather than carries out any restrictions to it；Although with reference to the foregoing embodiments
The present invention is described in detail, for the person of ordinary skill of the art, still can be to previous embodiment institute
The technical scheme of record is modified, or carries out equivalent substitution to which part technical characteristic；And these modifications or substitutions, and
The essence of appropriate technical solution is not set to depart from the spirit and scope of claimed technical solution of the invention.
Claims (9)
1. the power-system short-term load forecasting method based on depth recurrent neural network, it is characterised in that comprise the following steps：
(1) collect and collect history network load and meteorological data, it is standby to build storehouse；
(2) abnormal data obtained by removal step (1), to remaining data normalized；
(3) model structure with feedforward and feedback function is determined；
(4) the DRNN forecast models based on IPSO algorithms are trained using historical data, determine model parameter and weights；
(5) the DRNN forecast models based on IPSO algorithms are used in the prediction of actual load, draw prediction load value.
2. the power-system short-term load forecasting method according to claim 1 based on depth recurrent neural network, it is special
Sign is：The model that step (3) determines includes input layer, multiple hidden layers, associated layers and output layer, and associated layers contain for depositing
Store up the memory cell of historical information, memory cell using the historical information at the current time of storage as feedback for subsequent time the
The input of one layer of hidden layer.
3. the power-system short-term load forecasting method according to claim 2 based on depth recurrent neural network, it is special
Sign is：The model that step (3) determines is y (t)=g (Wn+1·Rn(t)+Bn+1)
In formula：The weight matrix of Wn+1, Bn+1 between n-th layer hidden layer and output layer；G is the nonlinear activation of DRNN output layers
Function；Activation primitive f, g are sigmoid functions,
4. according to any described power-system short-term load forecasting sides based on depth recurrent neural network of claim 1-3
Method, it is characterised in that：After step (2) normalized, Load Characteristic Analysis is carried out to data.
5. the power-system short-term load forecasting method according to claim 4 based on depth recurrent neural network, it is special
Sign is：Load Characteristic Analysis is to use statistical technique analysis load data characteristic, determines load data relevant feature.
6. according to any described power-system short-term load forecasting sides based on depth recurrent neural network of claim 1-3
Method, it is characterised in that：History network load and meteorological data in step (1) are first 3 years.
7. according to any described power-system short-term load forecasting sides based on depth recurrent neural network of claim 1-3
Method, it is characterised in that：Abnormal data in step (2) is zero data or maximum or minimum value.
8. according to any described power-system short-term load forecasting sides based on depth recurrent neural network of claim 1-3
Method, it is characterised in that：Normalization in step (2) is by data normalization to [- 1,1] section.
9. according to any described power-system short-term load forecasting sides based on depth recurrent neural network of claim 1-3
Method, it is characterised in that：The value obtained by forecast model in step (5), renormalization, you can the prediction for obtaining actual number magnitude is born
Charge values.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711293583.1A CN107766995A (en) | 2017-12-08 | 2017-12-08 | Power-system short-term load forecasting method based on depth recurrent neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711293583.1A CN107766995A (en) | 2017-12-08 | 2017-12-08 | Power-system short-term load forecasting method based on depth recurrent neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107766995A true CN107766995A (en) | 2018-03-06 |
Family
ID=61278006
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711293583.1A Pending CN107766995A (en) | 2017-12-08 | 2017-12-08 | Power-system short-term load forecasting method based on depth recurrent neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107766995A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108459933A (en) * | 2018-03-21 | 2018-08-28 | 哈工大大数据（哈尔滨）智能科技有限公司 | A kind of big data computer system fault detection method based on depth Recursive Networks |
| CN109062962A (en) * | 2018-06-27 | 2018-12-21 | 华南理工大学 | A kind of gating cycle neural network point of interest recommended method merging Weather information |
| CN109615124A (en) * | 2018-11-29 | 2019-04-12 | 中国铁路总公司 | A kind of SCADA main website load forecasting method based on deep learning |
| CN110110930A (en) * | 2019-05-08 | 2019-08-09 | 西南交通大学 | A kind of Recognition with Recurrent Neural Network Short-Term Load Forecasting Method improving whale algorithm |
| CN110222887A (en) * | 2019-05-27 | 2019-09-10 | 国网安徽省电力有限公司 | Prediction technique based on VMD and DNN and the application in short-term load forecasting |
| CN110222727A (en) * | 2019-05-15 | 2019-09-10 | 广东电网有限责任公司电力调度控制中心 | A kind of short-term load forecasting method and device based on deep neural network |
| CN110245786A (en) * | 2019-05-23 | 2019-09-17 | 上海交通大学 | House power load forecasting method based on neural network model |
| CN110516736A (en) * | 2019-06-04 | 2019-11-29 | 沈阳瑞初科技有限公司 | The visual multi-source heterogeneous data multilayer DRNN depth integration method of multidimensional |
| CN110751318A (en) * | 2019-09-26 | 2020-02-04 | 上海电力大学 | IPSO-LSTM-based ultra-short-term power load prediction method |
| CN111931973A (en) * | 2020-06-16 | 2020-11-13 | 广东电网有限责任公司 | Method for improving short-time load prediction precision of cable line |
| CN112308342A (en) * | 2020-11-25 | 2021-02-02 | 广西电网有限责任公司北海供电局 | Daily load prediction method based on deep time decoupling and application |
| CN113169579A (en) * | 2018-06-22 | 2021-07-23 | 莫伊沙能源控股有限公司 | System for machine learning, local multi-asset flexibility optimization and management of distributed energy storage resources |
| CN113205207A (en) * | 2021-04-19 | 2021-08-03 | 深圳供电局有限公司 | XGboost algorithm-based short-term power consumption load fluctuation prediction method and system |
| CN113759225A (en) * | 2020-06-01 | 2021-12-07 | 上海交通大学 | IGBT residual life prediction and state evaluation realization method |
| CN114715331A (en) * | 2022-06-02 | 2022-07-08 | 中国海洋大学 | Floating ocean platform power positioning control method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103729695A (en) * | 2014-01-06 | 2014-04-16 | 国家电网公司 | Short-term power load forecasting method based on particle swarm and BP neural network |
| CN105631517A (en) * | 2015-12-17 | 2016-06-01 | 河海大学 | Photovoltaic power generation power short term prediction method based on mind evolution Elman neural network |
| WO2016106373A1 (en) * | 2014-12-22 | 2016-06-30 | Robert Bosch Gmbh | Method for adaptive demand charge reduction |
| CN106022549A (en) * | 2016-07-28 | 2016-10-12 | 兰州理工大学 | Short term load predication method based on neural network and thinking evolutionary search |
| CN107423839A (en) * | 2017-04-17 | 2017-12-01 | 湘潭大学 | A kind of method of the intelligent building microgrid load prediction based on deep learning |
-
2017
- 2017-12-08 CN CN201711293583.1A patent/CN107766995A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103729695A (en) * | 2014-01-06 | 2014-04-16 | 国家电网公司 | Short-term power load forecasting method based on particle swarm and BP neural network |
| WO2016106373A1 (en) * | 2014-12-22 | 2016-06-30 | Robert Bosch Gmbh | Method for adaptive demand charge reduction |
| CN105631517A (en) * | 2015-12-17 | 2016-06-01 | 河海大学 | Photovoltaic power generation power short term prediction method based on mind evolution Elman neural network |
| CN106022549A (en) * | 2016-07-28 | 2016-10-12 | 兰州理工大学 | Short term load predication method based on neural network and thinking evolutionary search |
| CN107423839A (en) * | 2017-04-17 | 2017-12-01 | 湘潭大学 | A kind of method of the intelligent building microgrid load prediction based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 刘荣: "基于Elman神经网络的短期负荷预测", 《中国优秀硕士学位论文全文数据库工程科技II辑》 * |
| 王晓霞，等: "基于改进粒子群优化的神经网络及应用", 《华北电力大学学报》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108459933B (en) * | 2018-03-21 | 2021-10-22 | 哈工大大数据（哈尔滨）智能科技有限公司 | Big data computer system fault detection method based on deep recursion network |
| CN108459933A (en) * | 2018-03-21 | 2018-08-28 | 哈工大大数据（哈尔滨）智能科技有限公司 | A kind of big data computer system fault detection method based on depth Recursive Networks |
| CN113169579A (en) * | 2018-06-22 | 2021-07-23 | 莫伊沙能源控股有限公司 | System for machine learning, local multi-asset flexibility optimization and management of distributed energy storage resources |
| CN109062962A (en) * | 2018-06-27 | 2018-12-21 | 华南理工大学 | A kind of gating cycle neural network point of interest recommended method merging Weather information |
| CN109062962B (en) * | 2018-06-27 | 2022-03-29 | 华南理工大学 | Weather information fused gated cyclic neural network interest point recommendation method |
| CN109615124A (en) * | 2018-11-29 | 2019-04-12 | 中国铁路总公司 | A kind of SCADA main website load forecasting method based on deep learning |
| CN109615124B (en) * | 2018-11-29 | 2022-12-30 | 中国铁路总公司 | SCADA master station load prediction method based on deep learning |
| CN110110930A (en) * | 2019-05-08 | 2019-08-09 | 西南交通大学 | A kind of Recognition with Recurrent Neural Network Short-Term Load Forecasting Method improving whale algorithm |
| CN110110930B (en) * | 2019-05-08 | 2022-03-25 | 西南交通大学 | Recurrent neural network short-term power load prediction method for improving whale algorithm |
| CN110222727A (en) * | 2019-05-15 | 2019-09-10 | 广东电网有限责任公司电力调度控制中心 | A kind of short-term load forecasting method and device based on deep neural network |
| CN110245786A (en) * | 2019-05-23 | 2019-09-17 | 上海交通大学 | House power load forecasting method based on neural network model |
| CN110222887A (en) * | 2019-05-27 | 2019-09-10 | 国网安徽省电力有限公司 | Prediction technique based on VMD and DNN and the application in short-term load forecasting |
| CN110516736B (en) * | 2019-06-04 | 2022-02-22 | 沈阳瑞初科技有限公司 | Multi-dimensional visual multi-source heterogeneous data multi-layer DRNN depth fusion method |
| CN110516736A (en) * | 2019-06-04 | 2019-11-29 | 沈阳瑞初科技有限公司 | The visual multi-source heterogeneous data multilayer DRNN depth integration method of multidimensional |
| CN110751318A (en) * | 2019-09-26 | 2020-02-04 | 上海电力大学 | IPSO-LSTM-based ultra-short-term power load prediction method |
| CN110751318B (en) * | 2019-09-26 | 2024-01-30 | 上海电力大学 | Ultra-short-term power load prediction method based on IPSO-LSTM |
| CN113759225A (en) * | 2020-06-01 | 2021-12-07 | 上海交通大学 | IGBT residual life prediction and state evaluation realization method |
| CN113759225B (en) * | 2020-06-01 | 2022-07-19 | 上海交通大学 | IGBT residual life prediction and state evaluation realization method |
| CN111931973A (en) * | 2020-06-16 | 2020-11-13 | 广东电网有限责任公司 | Method for improving short-time load prediction precision of cable line |
| CN112308342A (en) * | 2020-11-25 | 2021-02-02 | 广西电网有限责任公司北海供电局 | Daily load prediction method based on deep time decoupling and application |
| CN113205207A (en) * | 2021-04-19 | 2021-08-03 | 深圳供电局有限公司 | XGboost algorithm-based short-term power consumption load fluctuation prediction method and system |
| CN114715331A (en) * | 2022-06-02 | 2022-07-08 | 中国海洋大学 | Floating ocean platform power positioning control method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107766995A (en) | Power-system short-term load forecasting method based on depth recurrent neural network | |
| Li et al. | A hybrid model based on synchronous optimisation for multi-step short-term wind speed forecasting | |
| CN108510074A (en) | A kind of implementation method for improving GWO algorithms | |
| CN110517482B (en) | Short-term traffic flow prediction method based on 3D convolutional neural network | |
| CN105844371A (en) | Electricity customer short-term load demand forecasting method and device | |
| CN111861013B (en) | Power load prediction method and device | |
| CN102831489B (en) | Power matching network builds material requirements Forecasting Methodology and device | |
| CN105303262A (en) | Short period load prediction method based on kernel principle component analysis and random forest | |
| CN106296044B (en) | Power system risk scheduling method and system | |
| CN116340006B (en) | Computing power resource idle prediction method based on deep learning and storage medium | |
| CN112215428B (en) | Photovoltaic power generation power prediction method and system based on error correction and fuzzy logic | |
| CN105930916A (en) | Parallel modular neural network-based byproduct gas real-time prediction method | |
| CN111710376B (en) | Block calculation load balancing method and system for macromolecules and cluster systems | |
| CN109214565A (en) | A kind of subregion system loading prediction technique suitable for the scheduling of bulk power grid subregion | |
| Fan et al. | Multi-objective LSTM ensemble model for household short-term load forecasting | |
| Aishwarya et al. | Prediction of time series data using GA-BPNN based hybrid ANN model | |
| Al-Hajj et al. | Multi-level stacking of long short term memory recurrent models for time series forecasting of solar radiation | |
| CN114611757A (en) | Electric power system short-term load prediction method based on genetic algorithm and improved depth residual error network | |
| CN115528750B (en) | Power grid safety and stability oriented data model hybrid drive unit combination method | |
| CN112288187A (en) | Big data-based electricity sales amount prediction method | |
| CN113283638A (en) | Load extreme curve prediction method and system based on fusion model | |
| CN112465195A (en) | Bus load prediction method and system considering high-proportion distributed photovoltaic access | |
| Choudhary et al. | Estimation of wind power using different soft computing methods | |
| Zhao et al. | Study of short-term load forecasting in big data environment | |
| Liu et al. | Ultra-short-term wind power forecasting based on stacking model |
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: 20180306 |