CN108416690A - Load Forecasting based on depth LSTM neural networks - Google Patents
Load Forecasting based on depth LSTM neural networks Download PDFInfo
- Publication number
- CN108416690A CN108416690A CN201810054530.2A CN201810054530A CN108416690A CN 108416690 A CN108416690 A CN 108416690A CN 201810054530 A CN201810054530 A CN 201810054530A CN 108416690 A CN108416690 A CN 108416690A
- Authority
- CN
- China
- Prior art keywords
- lstm
- load
- training sample
- prediction
- data
- 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 36
- 238000012549 training Methods 0.000 claims abstract description 89
- 238000000034 method Methods 0.000 claims description 19
- 238000005457 optimization Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 5
- 238000013277 forecasting method Methods 0.000 description 5
- 238000012216 screening Methods 0.000 description 4
- 240000002853 Nelumbo nucifera Species 0.000 description 3
- 235000006508 Nelumbo nucifera Nutrition 0.000 description 3
- 235000006510 Nelumbo pentapetala Nutrition 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 239000000843 powder Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
-
- 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"
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
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)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of Load Forecastings based on depth LSTM neural networks, can improve precision of prediction, step-length and real-time, include the following steps：Training sample is generated according to input feature vector data and load data, wherein input feature vector data include the weather information of experimental period and whether are workaday time type information；Training sample is handled, and to treated, training sample is trained to obtain LSTM prediction models by LSTM neural networks；By by the weather information of time to be predicted and whether be that workaday time type information inputs LSTM prediction models, being predicted with the network load treated in predicted time to obtain network load prediction result；Network load prediction result is analyzed, and judges whether network load prediction result meets accuracy requirement；If it is determined that being unsatisfactory for accuracy requirement, then new training sample is obtained, and supplementary training is carried out to LSTM prediction models by new training sample, to be updated to LSTM prediction models.
Description
Technical field
The present invention relates to network load electric powder prediction, more particularly to a kind of power grid based on depth LSTM neural networks
Load forecasting method.
Background technology
Prediction for network load is the key point for ensureing power grid security reliability service, reducing electrical network economy loss,
The precision for improving load prediction is always the emphasis of people's research for many years.Due to power grid internal power source wide variety, energy profit
It is also had nothing in common with each other with mode, therefore causes network load data fluctuations, randomness larger, cause load prediction precision relatively low,
It is difficult the distribution of accurate fitting load data.
For load prediction, there are many prediction techniques at present, however as the continuous quickening of power grid intelligence speed, number
Make traditional load forecasting method cannot increasingly meet actually to answer according to the increase of amount and fluctuation, the randomness etc. of data
Demand.The algorithm of machine learning is applied to load by the increasingly maturation of machine Learning Theory and application, many scholars
Among prediction, compared to traditional prediction technique, machine learning method can preferably handle the fluctuation, random of load data
The problems such as property, the precision predicted network load is improved to a certain extent so that people are in the management and scheduling of power grid
Provide certain foundation.
However, current load prediction application, needs before building model, is screened to forecast sample mostly, profit
Training sample is filtered out with similitude, necessarily leads to the loss for not being selected the included information of sample, additionally, due to prediction result
Precision largely influenced by institute's Screening Samples, in this way when Screening Samples inaccuracy, prediction will be directly resulted in
The decline of precision.Also, current load forecasting method is stronger to historical load data dependence, i.e., need to input history in prediction
Load data is suitable for the shorter load prediction of time step and goes predicted time step-length farther out if it is desired to using historical load value
Load value, since load data changes in distribution is very fast, current load forecasting method cannot be satisfied requirement.
Invention content
The present invention is directed to solve the skills such as network load precision of prediction is not high enough, time step is shorter at least to a certain extent
Art problem.For this purpose, an object of the present invention is to provide one kind based on depth LSTM (Long Short-Term Memory,
Shot and long term is remembered) Load Forecasting of neural network, network load precision of prediction can be improved, and can predict bigger
The load data of time step, and the real-time of prediction can be improved.
In order to achieve the above objectives, the Load Forecasting proposed by the present invention based on depth LSTM neural networks, packet
Include following steps：Training sample is generated according to input feature vector data and load data, wherein the input feature vector data include real
It tests the weather information of time and whether is workaday time type information；The training sample is handled, and is passed through
To treated, training sample is trained to obtain LSTM prediction models LSTM neural networks；By by the gas of time to be predicted
Image information and whether be that workaday time type information inputs the LSTM prediction models, in the time to be predicted
Network load is predicted to obtain network load prediction result；The network load prediction result is analyzed, and is judged
Whether the network load prediction result meets accuracy requirement；If it is determined that being unsatisfactory for the accuracy requirement, then obtain new
Training sample, and by the new training sample to the LSTM prediction models carry out supplementary training, with to the LSTM
Prediction model is updated.
Load Forecasting according to the ... of the embodiment of the present invention based on depth LSTM neural networks, according to including experiment
The weather information of time and whether be workaday time type information input feature vector data generate training sample, pass through the instruction
Practice sample training and obtain LSTM prediction models, and by by the weather information of time to be predicted and whether be workaday time class
After the type information input LSTM prediction models obtain network load prediction result, network load prediction result can be analyzed,
If being unsatisfactory for accuracy requirement, supplementary training is carried out to LSTM prediction models by new training sample, with pre- to LSTM
Model is surveyed to be updated.It can either ensure training data information without being screened to data in training prediction model as a result,
Integrality, and can avoid because data screening inaccuracy due to load prediction precision is impacted；Due to input feature vector data
Weather information including experimental period and time type information, with the weather information of time to be predicted and time class when subsequent prediction
Type information does not depend on historical load data, can predict the load data of more large time step as input；By pre- to LSTM
Survey model and constantly carry out supplementary training, precision of prediction can either be improved, but can avoid increasing because of re -training calculation amount and
Prediction takes, and substantially increases the real-time of prediction.
In addition, the Load Forecasting based on depth LSTM neural networks proposed according to the above embodiment of the present invention
There can also be following additional technical characteristic：
According to one embodiment of present invention, the weather information includes maximum temperature, minimum temperature, highest relative humidity
And average relative humidity.
Specifically, carrying out processing to the training sample includes：Batch standardization is carried out to the training sample, is made defeated
Going out each dimension of signal, to be distributed in mean value be 0, in the distribution that variance is 1, wherein load data in the training sample is made
For the output signal；The transformation of dimension is carried out to carrying out the training sample after batch standardization, and according to the load number
According to actual value and predictor calculation loss function flost；Based on RMSProp optimization methods to the loss function flostValue into
Row optimization, makes the loss function flostValue it is minimum.
Further, it is calculated by the following formula the loss function flost：
Wherein, t is the sampling time point of each load data, Yt mnFor the actual value of the load data, yt mnIt is described
The predicted value of load data, n are the time step of the LSTM neural networks, and m is the batch number of the input feature vector data.
Further, the LSTM prediction models include LSTM hidden layers, the full articulamentum of input and export full articulamentum,
The network connection weight of LSTM prediction models is mainly made of three parts with biasing, and it is implicit with LSTM respectively to input full articulamentum
[U between layerin,bin], [W in LSTM hidden layersf,bf], [Wi,bi], [Wc,bc], [Wo,bo], LSTM hidden layers and output
[V between full articulamentumout,bout], the state change of each module includes in the LSTM prediction models：
Xt=Uin·Xt+bin
ct=σ (Wf·[ht-1,Xt]+bf)e ct+σ(Wi·[ht,Xt]+bi)e tanh(Wc·[ht-1,Xt]+bc)
ht=σ (Wo·[ht-1,Xt]+bo)e tanh(ct)
yt=Vout·ht+bout
Wherein, XtFor the input feature vector data at t-th of time point of some day, ctFor the network inside the LSTM hidden layers
State, ht-1For the output of the LSTM hidden layers at the t-1 time point, htFor the LSTM hidden layers at t-th of time point
Output, ytFor the output of the full articulamentum of output, tanh and σ are respectively tanh activation primitive and sigmnid activation
Function.
Specifically, the network load prediction result is analyzed, and whether judges the network load prediction result
Meet accuracy requirement, including：Obtain the predicted value and actual value of the load data of multiple sampled points in a predetermined period
Deviation average and deviation maximum value；The deviation average is compared with first threshold, and by the deviation maximum value
It is compared with second threshold；If the deviation average is more than or equal to the first threshold or the deviation maximum value is more than
Equal to the second threshold, then judge to be unsatisfactory for the accuracy requirement.
Further, the deviation average is：
The deviation maximum value is：
Wherein, k is the number of sampled point in a predetermined period, YtFor the actual value of the load data, ytIt is described negative
The predicted value of lotus data.
According to one embodiment of present invention, if it is determined that meeting the accuracy requirement, then the LSTM is not predicted
Model is updated.
Description of the drawings
Fig. 1 is the flow according to the Load Forecasting based on depth LSTM neural networks of the embodiment of the present invention
Figure；
Fig. 2 is the structural schematic diagram according to the LSTM prediction models of one embodiment of the invention；
Fig. 3 is according to the prediction error of one embodiment of the invention and the graph of relation of learning rate；
Fig. 4 is according to the prediction error of one embodiment of the invention and the graph of relation of frequency of training；
Fig. 5 is the network load prediction result schematic diagram according to one embodiment of the invention；
Fig. 6 is to predict error schematic diagram according to the network load of one embodiment of the invention；
Fig. 7 is to predict error schematic diagram according to the network load of another embodiment of the present invention；
Fig. 8 is to predict error schematic diagram according to the network load of another embodiment of the invention；
Fig. 9 is to be shown according to the box of the network load forecasting system based on depth LSTM neural networks of the embodiment of the present invention
It is intended to.
Specific implementation mode
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to for explaining the present invention, and is not considered as limiting the invention.
The network load prediction side based on depth LSTM neural networks of the embodiment of the present invention described below in conjunction with the accompanying drawings
Method.
Fig. 1 is the flow according to the Load Forecasting based on depth LSTM neural networks of the embodiment of the present invention
Figure.
As shown in Figure 1, the Load Forecasting based on depth LSTM neural networks of the embodiment of the present invention, including with
Lower step：
S1 generates training sample, wherein when input feature vector data include experiment according to input feature vector data and load data
Between weather information and whether be workaday time type information.
In one embodiment of the invention, training sample data={ X, Y } may include input feature vector data X and load number
According to Y.Wherein, weather information may include in a period of time such as maximum temperature, minimum temperature, the highest relative humidity peace in one day
Equal relative humidity, time type information include working day or festivals or holidays type, and X can be expressed as to X=(x accordingly1,x2,x3,x4,
x5).Load data can be by being sampled to obtain, for example, a sampled point, sampling can be used as every 5 minutes every preset time
Obtained load data Y is represented by Y=(y1,y2,y3,...,y288)。
S2 handles training sample, and by LSTM neural networks to treated training sample is trained with
Obtain LSTM prediction models.
In one embodiment of the invention, training sample is handled, is specifically included：Batch mark is carried out to training sample
Quasi-ization processing, it is 0 so that each dimension of output signal is distributed in mean value, in the distribution that variance is 1, wherein in training sample
Load data is as output signal；The transformation of dimension is carried out to carrying out the training sample after batch standardization, and according to load
The actual value and predictor calculation loss function f of datalost；Based on RMSProp optimization methods to loss function flostValue carry out
Optimization, makes loss function flostValue it is minimum.
Specifically, criticizing the training sample after standardization is：
Data '=(data-data_mean)/data_std, (1)
Wherein, data_mean is the average value of training sample data, and data_std is the variance of training sample data.
When the time step of LSTM neural networks is n, and the batch number of input feature vector data is m, the input of p-th of batch
Characteristic is represented by：
Correspondingly, it is the follow-up deviation being convenient between the actual value and predicted value of calculated load data, it can be by load data
It is expressed as：
Then, specific public using the quadratic sum counting loss function of the actual value of load data and the difference of predicted value
Formula is as follows：
Wherein, t is the sampling time point of each load data, Yt mnFor the actual value of the load data, yt mnIt is described
The predicted value of load data.
Finally, RMSProp optimization methods are based on to above-mentioned flostValue optimize, make its minimum.The optimization sides RMSProp
Method relatively batch gradient declines the optimization methods difference such as (BGD), stochastic gradient descent (SGD), passes through the fetching portion from training sample
Small lot sample carries out gradient calculating, and then saves a large amount of LSTM model trainings time, improves prediction model and is handling
Real-time when mass data.
S3, by by the weather information of time to be predicted and whether be that workaday time type information inputs LSTM predictions
Model is predicted with the network load treated in predicted time to obtain network load prediction result.
In one embodiment of the invention, time to be predicted and experimental period can day be unit, specific time length
It can be set according to actual needs.If the time to be predicted includes a couple of days, it will can daily be used as a predetermined period.
Wherein, the weather information of time to be predicted can be obtained according to weather forecast.
In one embodiment of the invention, the structure of LSTM prediction models can be as shown in Figure 2.Wherein, LSTM predicts mould
Type includes LSTM hidden layers, the full articulamentum of input and the full articulamentum of output, network connection weight and the biasing of LSTM prediction models
It is mainly made of three parts, respectively inputs the [U between full articulamentum and LSTM hidden layersin,bin], in LSTM hidden layers
[Wf,bf], [Wi,bi], [Wc,bc], [Wo,bo], [V between LSTM hidden layers and the full articulamentum of outputout,bout], LSTM predictions
The state change of each module includes in model：
Xt=Uin·Xt+bin, (5)
ct=σ (Wf·[ht-1,Xt]+bf)e ct+σ(Wi·[ht,Xt]+bi)e tanh(Wc·[ht-1,Xt]+bc), (6)
ht=σ (Wo·[ht-1,Xt]+bo)e tanh(ct), (7)
yt=Vout·ht+bout, (8)
Wherein, XtFor the input feature vector data at t-th of time point of some day, ctFor the network state inside LSTM hidden layers,
ht-1For the output of the LSTM hidden layers at the t-1 time point, htFor the output of the LSTM hidden layers at t-th of time point, ytIt is defeated
Go out the output of full articulamentum, tanh and σ are respectively tanh activation primitive and sigmnid activation primitives.
By above-mentioned LSTM prediction models, it is pre- that network load can be obtained according to the input feature vector data including weather information
Survey result.
S4 analyzes network load prediction result, and judges whether network load prediction result meets accuracy and want
It asks.
Specifically, the deviation of the predicted value and actual value of the load data of multiple sampled points in a predetermined period can be obtained
Deviation average, is then compared by average value and deviation maximum value with first threshold, and by deviation maximum value and the second threshold
Value is compared.If deviation average is more than or equal to first threshold or deviation maximum value is more than or equal to second threshold, judge
It is unsatisfactory for accuracy requirement.
In one embodiment of the invention, for the accuracy estimating of network load prediction result, mainly from two sides
Face considers, by taking predetermined period is one day as an example, first, and the predicted value and actual value of prediction 288 sampled point network loads of the same day
Deviation average is answered as small as possible；Second, prediction the same day in each sampled point load data predicted value and actual value it is inclined
Difference is answered as small as possible.According to mentioned above principle, deviation average can be obtained：
And deviation maximum value：
Wherein, k is the number of sampled point in a predetermined period, YtFor the actual value of load data, ytFor load data
Predicted value.
Respectively the predicted value and actual value of load data set corresponding threshold value, i.e. first threshold E simultaneouslymean, second
Threshold value Emax.Then the predicted value of load data and actual value are compared with corresponding threshold value respectively.If emax≥EmaxOr
Person emean≥Emean, then judge to be unsatisfactory for accuracy requirement.
S5 then obtains new training sample, and pass through new training sample pair if it is determined that being unsatisfactory for accuracy requirement
LSTM prediction models carry out supplementary training, to be updated to LSTM prediction models.
I.e. if there is emax≥EmaxOr emean≥Emean, then be directly based upon new training sample to LSTM prediction models into
Row supplementary training obtains new LSTM prediction models, and replaces original LSTM prediction models.
The return to step S3 after executing step S5, it is negative to treat the power grid in predicted time based on new LSTM prediction models
Lotus is predicted.
And if judging to meet accuracy requirement in step s 4, LSTM prediction models are not updated, i.e., are not held
Row step S5.
Load Forecasting according to the ... of the embodiment of the present invention based on depth LSTM neural networks, according to including experiment
The weather information of time and whether be workaday time type information input feature vector data generate training sample, pass through the instruction
Practice sample training and obtain LSTM prediction models, and by by the weather information of time to be predicted and whether be workaday time class
After the type information input LSTM prediction models obtain network load prediction result, network load prediction result can be analyzed,
If being unsatisfactory for accuracy requirement, supplementary training is carried out to LSTM prediction models by new training sample, with pre- to LSTM
Model is surveyed to be updated.It can either ensure training data information without being screened to data in training prediction model as a result,
Integrality, and can avoid because data screening inaccuracy due to load prediction precision is impacted；Due to input feature vector data
Weather information including experimental period and time type information, with the weather information of time to be predicted and time class when subsequent prediction
Type information does not depend on historical load data, can predict the load data of more large time step as input；By pre- to LSTM
Survey model and constantly carry out supplementary training, precision of prediction can either be improved, but can avoid increasing because of re -training calculation amount and
Prediction takes, and substantially increases the real-time of prediction.
Further supplementary explanation is carried out to above-described embodiment below by specific embodiment and the advantages of to above-described embodiment
It is verified.
In one particular embodiment of the present invention, the meteorological letter on June 30,1 day to 2016 January in 2014 can be obtained
Breath and load data, after rejecting missing data, the load data that remaining data is 900 days, sample frequency is sampling in 5 minutes
Point, i.e., one day totally 288 load datas choose preceding 600 groups of data as training data, and rear 300 groups of data are as test data.
Influence in view of learning rate to LSTM prediction model performances, excessive learning rate are easy algorithm is made to skip optimal
Point influences training precision, and too small learning rate can be absorbed in local optimum and generate over-fitting.Really for learning rate
It is fixed, at present still without specific theory support.It is similar with learning rate, the shadow of the frequency of training of LSTM neural networks to model performance
It rings equally very greatly, the determination of frequency of training is also without specific theories integration.Pass through fixed training time in the embodiment of the present invention
Influence of the number observational learning rate to prediction result, and influence of the fixed learning rate observation frequency of training to estimated performance, obtain
Result as shown in Figure 3 and Figure 4.It can be seen from Fig. 3 and Fig. 4 when learning rate lr is less than 0.0001, due to learning rate mistake
Small, model is easily trapped into local optimum；When lr is more than 0.01, model accuracy is relatively low, can not preferably be fitted load data
Variation.For frequency of training, when frequency of training is 800 or so, precision of prediction highest, and frequency of training is relatively
The rare training speed for being conducive to accelerate model, meets the requirement of real-time.It, can be by learning rate and frequency of training based on above-mentioned conclusion
It is set to respectively 0.001 and 800 time.
The weather information of time to be predicted and time type information are inputted into above-mentioned LSTM prediction models, obtained power grid is negative
Lotus prediction result is as shown in Figure 5.Then can statistical forecast error, be calculated daily mean error average value be 0.078, in advance
Survey daily worst error average value be 0.278, it can be seen that when model to step-length load is predicted farther out when, equally have
Good prediction effect.Include a large amount of distributed energies additionally, due to power grid inside, sudden load change phenomenon is more serious, by upper figure
It can be seen that load data often mutates in some time, it is based on this characteristic, it is prominent that LSTM can accurately chase load
Become data, ensures the accuracy of network load prediction.
With the propulsion of time, when model, which cannot be satisfied prediction, to be required, as shown in fig. 6, the predicted value of network load
Error is larger, can be by continuing to carry out supplementary training to model interpolation data, to make when error is higher than set threshold value
Model is fitted power load distributing again, continues to train by the way that the data of a preceding training data backward are added, obtains new prediction mould
Type.The network load obtained by new prediction model is as shown in Figure 7.
As seen from Figure 7, after carrying out supplementary training to original model, power load distributing equally can be preferably fitted.
In addition, to save the time, practicability is improved, the embodiment of the present invention is when load error is unsatisfactory for precision of prediction, just to model
Supplementary training is carried out, this avoid having new data that will recall original model every time to carry out supplementary training, to subtract significantly
Computation burden is lacked.
The last embodiment of the present invention by being compared using other wide prediction techniques in load prediction, and
And by utilizing the method that T is examined to prediction result, it was demonstrated that the Load Forecasting of the embodiment of the present invention and other predictions
The significance difference of algorithm is anisotropic.The method of common load prediction mainly has support vector machines (SVM), linear extrapolation, grey pre-
(GM) method of survey etc., respectively by comparing precision and calculating time, practicality of the prominent LSTM algorithms in terms of network load prediction
Property.Specifically precision evaluation index is respectively：Root-mean-square error (RMSE), absolute value error (MAE) and relative error
(MAPE), the calculation formula of each index is as follows：
Be 5 minutes data using above-mentioned sample frequency, carry out the verification of comparison algorithm, gained prediction result such as Fig. 8 and
Shown in table 1.
Table 1
As seen from Figure 8, when load data is in rising or decline since data are smoother, general pre- measuring and calculating
Method can be fitted this variation tendency well, obtain relatively good effect, but when load data is at inflection point, prediction essence
Degree is substantially reduced, and especially when handling this sudden load change, general prediction algorithm is difficult this variation of tracking, therefore is directed to this
The case where kind mutation, to load forecasting method, more stringent requirements are proposed.Precision of prediction obtained by LSTM is above other three kinds
Prediction technique, especially when load data mutates, prediction algorithm equally can be compared with subject to used in the embodiment of the present invention
This variation tendency of true tracking, achieves good prediction effect.In addition as can be seen from Table 1, although with other three kinds
Algorithm can be more than the training time of other models compared to the time of originally trained LSTM prediction models, but when load data becomes
When changing existing model and cannot be satisfied prediction and require, other prediction techniques that compare carry out the re -training of model, and the present invention is real
The prediction technique for carrying out supplementary training on the basis of having model that example is proposed is applied, model will be greatly improved in processing data
The fast applicable ability of magnanimity, pace of change.
95% confidence interval is set based on the statistical that T is examined using SPSS software platforms, it was demonstrated that the present invention is real
Apply the otherness of the prediction technique and other three kinds of prediction algorithms of example.The results are shown in Table 2 for gained.
Table 2
As can be seen from Table 2, LSTM (the i.e. network load predictions based on depth LSTM neural networks of the embodiment of the present invention
Method) and GM obtained by prediction result and actual value between difference, first P (sig) value is 0.002, is less than 0.05, illustrates vacation
If variance is equal invalid, second P (sig) value is 0, similarly less than 0.05, illustrates the prediction result number obtained by LSTM and GM
There are the differences of conspicuousness between.Similarly, it can be deduced that between prediction results obtained by LSTM and other two kinds of algorithms there is also
Significant difference.
To realize that the Load Forecasting based on depth LSTM neural networks of above-described embodiment, the present invention also propose
A kind of network load forecasting system based on depth LSTM neural networks.
As shown in figure 9, the network load forecasting system based on depth LSTM neural networks of the embodiment of the present invention includes：Sample
This generation module 10, model construction module 20, prediction module 30, interpretation of result module 40 and supplementary training module 50.
Wherein, sample generation module 10 is used to generate training sample according to input feature vector data and load data, wherein defeated
Enter the weather information and whether be workaday time type information that characteristic includes experimental period；Model construction module 20 is used
It is handled in training sample, and to treated, that training sample is trained is pre- to obtain LSTM by LSTM neural networks
Survey model；Whether prediction module 30 is used for by by the weather information of time to be predicted and be that workaday time type information is defeated
Enter LSTM prediction models, is predicted with the network load treated in predicted time to obtain network load prediction result；As a result
Analysis module 40 judges whether network load prediction result meets accuracy for analyzing network load prediction result
It is required that；Supplementary training module 50 is used to be unsatisfactory for accuracy requirement in the judgement of interpretation of result module 40, obtains new training sample,
And supplementary training is carried out to LSTM prediction models by new training sample, to be updated to LSTM prediction models.
More specific embodiment can refer to above-described embodiment, and to avoid redundancy, details are not described herein.
Network load forecasting system according to the ... of the embodiment of the present invention based on depth LSTM neural networks, can ensure to train
The integrality of data information improves precision of prediction, and can predict the load data of more large time step, and can improve pre-
The real-time of survey.
In the description of the present invention, it is to be understood that, term "center", " longitudinal direction ", " transverse direction ", " length ", " width ",
" thickness ", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outside", " up time
The orientation or positional relationship of the instructions such as needle ", " counterclockwise ", " axial direction ", " radial direction ", " circumferential direction " be orientation based on ... shown in the drawings or
Position relationship is merely for convenience of description of the present invention and simplification of the description, and does not indicate or imply the indicated device or element must
There must be specific orientation, with specific azimuth configuration and operation, therefore be not considered as limiting the invention.
In addition, term " first ", " second " are used for description purposes only, it is not understood to indicate or imply relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include one or more this feature.In the description of the present invention, the meaning of " plurality " is two or more,
Unless otherwise specifically defined.
In the present invention unless specifically defined or limited otherwise, term " installation ", " connected ", " connection ", " fixation " etc.
Term shall be understood in a broad sense, for example, it may be being fixedly connected, may be a detachable connection, or integral；Can be that machinery connects
It connects, can also be electrical connection；It can be directly connected, can also can be indirectly connected through an intermediary in two elements
The interaction relationship of the connection in portion or two elements.It for the ordinary skill in the art, can be according to specific feelings
Condition understands the concrete meaning of above-mentioned term in the present invention.
In the present invention unless specifically defined or limited otherwise, fisrt feature can be with "above" or "below" second feature
It is that the first and second features are in direct contact or the first and second features pass through intermediary mediate contact.Moreover, fisrt feature exists
Second feature " on ", " top " and " above " but fisrt feature be directly above or diagonally above the second feature, or be merely representative of
Fisrt feature level height is higher than second feature.Fisrt feature second feature " under ", " lower section " and " below " can be
One feature is directly under or diagonally below the second feature, or is merely representative of fisrt feature level height and is less than second feature.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiments or example.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, changes, replacing and modification.
Claims (8)
1. a kind of Load Forecasting based on depth LSTM neural networks, which is characterized in that include the following steps：
Training sample is generated according to input feature vector data and load data, wherein the input feature vector data include experimental period
Weather information and whether be workaday time type information；
The training sample is handled, and to treated, training sample is trained to obtain by LSTM neural networks
LSTM prediction models；
By by the weather information of time to be predicted and whether be that workaday time type information inputs the LSTM and predicts mould
Type, to be predicted the network load in the time to be predicted to obtain network load prediction result；
The network load prediction result is analyzed, and judges whether the network load prediction result meets accuracy and want
It asks；
If it is determined that being unsatisfactory for the accuracy requirement, then new training sample is obtained, and pass through the new training sample pair
The LSTM prediction models carry out supplementary training, to be updated to the LSTM prediction models.
2. the Load Forecasting according to claim 1 based on depth LSTM neural networks, which is characterized in that institute
It includes maximum temperature, minimum temperature, highest relative humidity and average relative humidity to state weather information.
3. the Load Forecasting according to claim 2 based on depth LSTM neural networks, which is characterized in that right
The training sample is handled, and is specifically included：
Batch standardization is carried out to the training sample, it is 0 so that each dimension of output signal is distributed in mean value, variance 1
Distribution in, wherein the load data in the training sample is as the output signal；
Carry out the transformation of dimension to carrying out the training sample after batch standardization, and according to the actual value of the load data and
Predictor calculation loss function flost；
Based on RMSProp optimization methods to the loss function flostValue optimize, make the loss function flostValue most
It is small.
4. the Load Forecasting according to claim 3 based on depth LSTM neural networks, which is characterized in that logical
It crosses following formula and calculates the loss function flost：
Wherein, t is the sampling time point of each load data, Yt mnFor the actual value of the load data, yt mnFor the load
The predicted value of data, n are the time step of the LSTM neural networks, and m is the batch number of the input feature vector data.
5. the Load Forecasting according to claim 4 based on depth LSTM neural networks, which is characterized in that institute
It includes LSTM hidden layers, the full articulamentum of input and the full articulamentum of output to state LSTM prediction models, and the network of LSTM prediction models connects
It connects weight to be mainly made of three parts with biasing, respectively inputs the [U between full articulamentum and LSTM hidden layersin,bin],
[W in LSTM hidden layersf,bf], [Wi,bi], [Wc,bc], [Wo,bo], between LSTM hidden layers and the full articulamentum of output
[Vout,bout], the state change of each module includes in the LSTM prediction models：
Xt=Uin·Xt+bin
ct=σ (Wf·[ht-1,Xt]+bf)e ct+σ(Wi·[ht,Xt]+bi)e tanh(Wc·[ht-1,Xt]+bc)
ht=σ (Wo·[ht-1,Xt]+bo)e tanh(ct)
yt=Vout·ht+bout
Wherein, XtFor the input feature vector data at t-th of time point of some day, ctFor the network state inside the LSTM hidden layers,
ht-1For the output of the LSTM hidden layers at the t-1 time point, htFor t-th of time point the LSTM hidden layers it is defeated
Go out, ytFor the output of the full articulamentum of output, tanh and σ are respectively tanh activation primitive and sigmnid activation primitives.
6. the Load Forecasting according to claim 5 based on depth LSTM neural networks, which is characterized in that right
The network load prediction result is analyzed, and judges whether the network load prediction result meets accuracy requirement, tool
Body includes：
Obtain the predicted value of load data of multiple sampled points and the deviation average of actual value and deviation in a predetermined period
Maximum value；
The deviation average is compared with first threshold, and the deviation maximum value is compared with second threshold；
If the deviation average is more than or equal to the first threshold or the deviation maximum value is more than or equal to second threshold
Value then judges to be unsatisfactory for the accuracy requirement.
7. the Load Forecasting according to claim 6 based on depth LSTM neural networks, which is characterized in that
The deviation average is：
The deviation maximum value is：
Wherein, k is the number of sampled point in a predetermined period, YtFor the actual value of the load data, ytFor the load number
According to predicted value.
8. the Load Forecasting according to claim 1 or 5 based on depth LSTM neural networks, feature exist
In if it is determined that meeting the accuracy requirement, not then being updated to the LSTM prediction models.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810054530.2A CN108416690A (en) | 2018-01-19 | 2018-01-19 | Load Forecasting based on depth LSTM neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810054530.2A CN108416690A (en) | 2018-01-19 | 2018-01-19 | Load Forecasting based on depth LSTM neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108416690A true CN108416690A (en) | 2018-08-17 |
Family
ID=63125901
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810054530.2A Pending CN108416690A (en) | 2018-01-19 | 2018-01-19 | Load Forecasting based on depth LSTM neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108416690A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214610A (en) * | 2018-11-16 | 2019-01-15 | 国网江苏省电力有限公司经济技术研究院 | A kind of saturation Methods of electric load forecasting based on shot and long term Memory Neural Networks |
| CN109242236A (en) * | 2018-07-05 | 2019-01-18 | 西安图迹信息科技有限公司 | Based on the PMU primary frequency modulation load forecasting method for being associated with full Connection Neural Network and LSTM |
| CN109376896A (en) * | 2018-08-29 | 2019-02-22 | 国网重庆市电力公司南岸供电分公司 | A kind of term load forecasting for distribution based on multimodality fusion |
| CN109583346A (en) * | 2018-11-21 | 2019-04-05 | 齐鲁工业大学 | EEG feature extraction and classifying identification method based on LSTM-FC |
| CN109615226A (en) * | 2018-12-12 | 2019-04-12 | 焦点科技股份有限公司 | A kind of operation indicator method for monitoring abnormality |
| CN109754113A (en) * | 2018-11-29 | 2019-05-14 | 南京邮电大学 | Load forecasting method based on dynamic time warping Yu length time memory |
| CN109829587A (en) * | 2019-02-12 | 2019-05-31 | 国网山东省电力公司电力科学研究院 | Zonule grade ultra-short term and method for visualizing based on depth LSTM network |
| CN109934392A (en) * | 2019-02-28 | 2019-06-25 | 武汉大学 | A kind of micro-capacitance sensor short-term load forecasting method based on deep learning |
| CN109948845A (en) * | 2019-03-15 | 2019-06-28 | 国网江苏省电力有限公司经济技术研究院 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
| CN111104298A (en) * | 2019-11-22 | 2020-05-05 | 国网黑龙江省电力有限公司信息通信公司 | LSTM-based power grid server running state prediction device |
| CN111489036A (en) * | 2020-04-14 | 2020-08-04 | 天津相和电气科技有限公司 | Resident load prediction method and device based on electrical appliance load characteristics and deep learning |
| CN111598289A (en) * | 2020-03-30 | 2020-08-28 | 国网河北省电力有限公司 | Distributed optimization method of integrated energy system considering LSTM photovoltaic output prediction |
| CN111832796A (en) * | 2020-02-29 | 2020-10-27 | 上海电力大学 | Fine classification and prediction method and system for residential electricity load mode |
| CN112036658A (en) * | 2020-09-09 | 2020-12-04 | 华东交通大学 | Wind power real-time prediction method based on random forgetting long-term and short-term memory |
| 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 |
| CN113408795A (en) * | 2021-06-03 | 2021-09-17 | 国网河北省电力有限公司高邑县供电分公司 | Power load prediction system and method based on grey theory |
| CN113537578A (en) * | 2021-06-25 | 2021-10-22 | 国网江苏省电力有限公司信息通信分公司 | Power consumer behavior prediction method |
| CN113632131A (en) * | 2019-04-10 | 2021-11-09 | 三菱电机株式会社 | Power management device |
| WO2023284731A1 (en) * | 2021-07-13 | 2023-01-19 | 中兴通讯股份有限公司 | Parameter setting method and apparatus, and electronic device and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952181A (en) * | 2017-03-08 | 2017-07-14 | 深圳市景程信息科技有限公司 | Electric Load Prediction System based on long Memory Neural Networks in short-term |
| CN106960252A (en) * | 2017-03-08 | 2017-07-18 | 深圳市景程信息科技有限公司 | Methods of electric load forecasting based on long Memory Neural Networks in short-term |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| CN107590567A (en) * | 2017-09-13 | 2018-01-16 | 南京航空航天大学 | A kind of Recognition with Recurrent Neural Network short-term load forecasting method based on comentropy cluster and notice mechanism |
-
2018
- 2018-01-19 CN CN201810054530.2A patent/CN108416690A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952181A (en) * | 2017-03-08 | 2017-07-14 | 深圳市景程信息科技有限公司 | Electric Load Prediction System based on long Memory Neural Networks in short-term |
| CN106960252A (en) * | 2017-03-08 | 2017-07-18 | 深圳市景程信息科技有限公司 | Methods of electric load forecasting based on long Memory Neural Networks in short-term |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| CN107590567A (en) * | 2017-09-13 | 2018-01-16 | 南京航空航天大学 | A kind of Recognition with Recurrent Neural Network short-term load forecasting method based on comentropy cluster and notice mechanism |
Non-Patent Citations (2)
| Title |
|---|
| 张宇航 等: "一种基于LSTM神经网络的短期用电负荷预测方法", 《电力信息与通信技术》 * |
| 朱海兵等: "基于改进型BP神经网络的电网负荷预测", 《现代电子技术》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109242236B (en) * | 2018-07-05 | 2021-04-20 | 西安图迹信息科技有限公司 | PMU primary frequency modulation load prediction method based on correlation full-connection neural network and LSTM |
| CN109242236A (en) * | 2018-07-05 | 2019-01-18 | 西安图迹信息科技有限公司 | Based on the PMU primary frequency modulation load forecasting method for being associated with full Connection Neural Network and LSTM |
| CN109376896A (en) * | 2018-08-29 | 2019-02-22 | 国网重庆市电力公司南岸供电分公司 | A kind of term load forecasting for distribution based on multimodality fusion |
| CN109214610B (en) * | 2018-11-16 | 2022-05-31 | 国网江苏省电力有限公司经济技术研究院 | Saturated power load prediction method based on long-term and short-term memory neural network |
| CN109214610A (en) * | 2018-11-16 | 2019-01-15 | 国网江苏省电力有限公司经济技术研究院 | A kind of saturation Methods of electric load forecasting based on shot and long term Memory Neural Networks |
| CN109583346A (en) * | 2018-11-21 | 2019-04-05 | 齐鲁工业大学 | EEG feature extraction and classifying identification method based on LSTM-FC |
| CN109754113A (en) * | 2018-11-29 | 2019-05-14 | 南京邮电大学 | Load forecasting method based on dynamic time warping Yu length time memory |
| CN109754113B (en) * | 2018-11-29 | 2022-08-30 | 南京邮电大学 | Load prediction method based on dynamic time warping and long-and-short time memory |
| CN109615226A (en) * | 2018-12-12 | 2019-04-12 | 焦点科技股份有限公司 | A kind of operation indicator method for monitoring abnormality |
| CN109829587A (en) * | 2019-02-12 | 2019-05-31 | 国网山东省电力公司电力科学研究院 | Zonule grade ultra-short term and method for visualizing based on depth LSTM network |
| CN109934392A (en) * | 2019-02-28 | 2019-06-25 | 武汉大学 | A kind of micro-capacitance sensor short-term load forecasting method based on deep learning |
| CN109948845A (en) * | 2019-03-15 | 2019-06-28 | 国网江苏省电力有限公司经济技术研究院 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
| CN113632131A (en) * | 2019-04-10 | 2021-11-09 | 三菱电机株式会社 | Power management device |
| US11954744B2 (en) | 2019-04-10 | 2024-04-09 | Mitsubishi Electric Corporation | Electric power management station |
| CN111104298A (en) * | 2019-11-22 | 2020-05-05 | 国网黑龙江省电力有限公司信息通信公司 | LSTM-based power grid server running state prediction device |
| CN111832796A (en) * | 2020-02-29 | 2020-10-27 | 上海电力大学 | Fine classification and prediction method and system for residential electricity load mode |
| CN111832796B (en) * | 2020-02-29 | 2022-12-23 | 上海电力大学 | Fine classification and prediction method and system for residential electricity load mode |
| CN111598289A (en) * | 2020-03-30 | 2020-08-28 | 国网河北省电力有限公司 | Distributed optimization method of integrated energy system considering LSTM photovoltaic output prediction |
| CN111489036A (en) * | 2020-04-14 | 2020-08-04 | 天津相和电气科技有限公司 | Resident load prediction method and device based on electrical appliance load characteristics and deep learning |
| CN111489036B (en) * | 2020-04-14 | 2023-06-09 | 天津相和电气科技有限公司 | Resident load prediction method and device based on electrical load characteristics and deep learning |
| CN112036658A (en) * | 2020-09-09 | 2020-12-04 | 华东交通大学 | Wind power real-time prediction method based on random forgetting long-term and short-term memory |
| CN112036658B (en) * | 2020-09-09 | 2022-06-14 | 华东交通大学 | Wind power real-time prediction method based on forgetting long-term and short-term memory with difference |
| 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 |
| CN113408795A (en) * | 2021-06-03 | 2021-09-17 | 国网河北省电力有限公司高邑县供电分公司 | Power load prediction system and method based on grey theory |
| CN113537578A (en) * | 2021-06-25 | 2021-10-22 | 国网江苏省电力有限公司信息通信分公司 | Power consumer behavior prediction method |
| WO2023284731A1 (en) * | 2021-07-13 | 2023-01-19 | 中兴通讯股份有限公司 | Parameter setting method and apparatus, and electronic device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416690A (en) | Load Forecasting based on depth LSTM neural networks | |
| CN106781489B (en) | A kind of road network trend prediction method based on recurrent neural network | |
| WO2020248228A1 (en) | Computing node load prediction method in a hadoop platform | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN110648014A (en) | Regional wind power prediction method and system based on space-time quantile regression | |
| CN109255726A (en) | A kind of ultra-short term wind power prediction method of Hybrid Intelligent Technology | |
| CN109637196A (en) | En-route sector traffic probability density prediction technique | |
| CN112329990A (en) | User power load prediction method based on LSTM-BP neural network | |
| CN113537626B (en) | Method for predicting neural network combined time sequence by aggregating information difference | |
| CN110837915A (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN111768622A (en) | Short-time traffic prediction method based on improved wolf algorithm | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN109740818A (en) | A kind of probability density forecasting system applied to en-route sector traffic | |
| CN115310782A (en) | Power consumer demand response potential evaluation method and device based on neural turing machine | |
| Robati et al. | Inflation rate modeling: adaptive neuro-fuzzy inference system approach and particle swarm optimization algorithm (ANFIS-PSO) | |
| Lao et al. | The optimized multivariate grey prediction model based on dynamic background value and its application | |
| Han et al. | Short-term 4D trajectory prediction based on LSTM neural network | |
| Wu et al. | Automatic fuzzy model identification for short-term load forecast | |
| Zhang et al. | Direction-decision learning based pedestrian flow behavior investigation | |
| Wang et al. | Electric vehicle charging load clustering and load forecasting based on long short term memory neural network | |
| Setiawan et al. | Evaluation of container forecasting methods for analyzing port container terminal performance using agent-based simulation | |
| CN116431439A (en) | Energy consumption monitoring and management system for green data center | |
| Li et al. | Hydropower generation forecasting via deep neural network | |
| Lin et al. | The short-term exit traffic prediction of a toll station based on LSTM | |
| CN112232557B (en) | Short-term prediction method for health degree of switch machine based on long-short-term memory network |
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 |
Application publication date: 20180817 |
|
| RJ01 | Rejection of invention patent application after publication |