CN109685290A - Deep learning-based power consumption prediction method, device and equipment - Google Patents
Deep learning-based power consumption prediction method, device and equipment Download PDFInfo
- Publication number
- CN109685290A CN109685290A CN201910109904.0A CN201910109904A CN109685290A CN 109685290 A CN109685290 A CN 109685290A CN 201910109904 A CN201910109904 A CN 201910109904A CN 109685290 A CN109685290 A CN 109685290A
- Authority
- CN
- China
- Prior art keywords
- sequence
- model
- prediction model
- demand forecasting
- value
- 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.)
- Granted
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 37
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 41
- 230000015654 memory Effects 0.000 claims abstract description 23
- 230000005611 electricity Effects 0.000 claims description 94
- 238000013277 forecasting method Methods 0.000 claims description 25
- 230000007787 long-term memory Effects 0.000 claims description 20
- 230000007246 mechanism Effects 0.000 claims description 9
- 238000011478 gradient descent method Methods 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000003860 storage Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 9
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 239000000243 solution Substances 0.000 description 5
- 230000007774 longterm Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000002123 temporal effect Effects 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000000714 time series forecasting Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000003466 anti-cipated effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000009466 transformation Effects 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a power consumption prediction method based on deep learning, which adopts a sequence-to-sequence seq2seq model comprising a long-short term memory network LSTM embedded block as a prediction model, enhances the learnability and the expressiveness of the sequence-to-sequence seq2seq model, obtains the prediction model by training historical power consumption data comprising a historical time sequence and exogenous characteristic data, enables the prediction model to learn and express the exogenous characteristic besides the time characteristic, increases the constraint on a plurality of factors influencing the trend of the power consumption, obtains the prediction model closer to the actual condition of the power consumption, predicts the power consumption through the prediction model, and obtains a more accurate power consumption prediction result. The invention also provides a device and equipment for predicting the power consumption based on deep learning, and the device and equipment have the beneficial effects.
Description
Technical field
The present invention relates to Electro-metering fields, more particularly to a kind of electricity demand forecasting method based on deep learning, dress
It sets and equipment.
Background technique
Electric system is a complicated real-time dynamic system, is related to power generation, transmission of electricity, with power transmission, electricity consumption, and scheduling pipe
All too many levels such as reason.Wherein, network load prediction is an important component of electric power system dispatching.
The electricity consumption of enterprise and load prediction belong to time series forecasting type, i.e., are predicted not according to historical time data
The value come.In the prior art for the processing of time series forecasting, autoregression integral sliding average (ARIMA) mould is generallyd use
Type carries out difference by the time series to non-stationary and obtains stable time series, then passes through auto-correlation coefficient and inclined phase
Relationship number is analyzed to obtain the parameter of model, and the data sequence that prediction object is formed over time is considered as a stochastic ordering
Column, with the model come this sequence of approximate description.But since ARIMA model is solely focused on the characteristic of time series itself, and
Often there is other features in addition to the time, this side that electricity demand forecasting is carried out using ARIMA model in electricity demand forecasting
The prediction result and practical electricity consumption deviation that formula obtains are larger.
The accuracy for how improving electricity demand forecasting is those skilled in the art's technical issues that need to address.
Summary of the invention
The electricity demand forecasting method, device and equipment based on deep learning that the object of the present invention is to provide a kind of, compared to
The prior art improves the accuracy of electricity demand forecasting.
In order to solve the above technical problems, the present invention provides a kind of electricity demand forecasting method based on deep learning, comprising:
Previously according to history electricity consumption data training prediction model；
Receive the power quantity predicting query time and exogenous characteristic parameter of input；
The power quantity predicting query time and the exogenous characteristic parameter are inputted into the prediction model, export electricity consumption
Predicted value；
Wherein, the history electricity consumption data includes the exogenous characteristic of historical time sequence and the historical time sequence
According to, the prediction model be specially include shot and long term memory network LSTM embedded block sequence to sequence seq2seq model.
Optionally, the encoder and decoder of the prediction model are shot and long term memory network LSTM.
Optionally, described previously according to history electricity consumption data training prediction model, it specifically includes:
According to the historical time sequence and the exogenous characteristic determine training input vector and with the instruction
Practice the corresponding label value of input vector；Wherein, the trained input vector includes parameter and characteristic parameter cycle of training；
The trained input vector is inputted into preset sequence to sequence seq2seq model structure, obtains trained values；
The error minimum value of the trained values Yu the label value is determined by preset rules；
The sequence is adjusted to the model parameter of sequence seq2seq model structure according to the error minimum value, obtains institute
State prediction model.
Optionally, the error minimum value that the trained values Yu the label value are determined by preset rules, specifically includes:
The negative gradient direction of the trained values Yu the label value is calculated using Adam gradient descent method, according to described negative
Gradient direction determines the error minimum value.
Optionally, the error minimum value that the trained values Yu the label value are determined by preset rules, specifically includes:
The negative gradient direction that the trained values Yu the label value are calculated using SGD stochastic gradient descent method, according to institute
It states negative gradient direction and determines the error minimum value.
Optionally, when the length of the historical time sequence is greater than threshold value, described that the trained input vector is defeated
When entering preset sequence to sequence seq2seq model structure, further includes:
It will be late by data point using the sliding window memory mechanism of fixed weight and introduce the preset sequence to sequence
Seq2seq model structure.
Optionally, the sliding window memory mechanism using fixed weight introduces the lag data point described default
Sequence to sequence seq2seq model structure, specifically include:
The encoder output value of the lag data point is inputted into full articulamentum to reduce dimension, and after dimension being reduced
The input feature vector of encoder output value addition decoder；
It averages to preset data point and neighbour's data point, to reduce noise according to the mean value and compensate non-uniform
Interval.
Optionally, described previously according to history electricity consumption data training prediction model, it specifically includes:
Multiple initial predicted models are obtained using the training of multiple groups initialization data；Wherein, initialization data group and it is described just
Beginning prediction model corresponds；
Checkpoint is saved in each initial predicted model；
The forecast power of initial predicted model corresponding with the checkpoint is calculated according to the checkpoint；
Model Fusion is carried out to each initial predicted model according to each forecast power, obtains the prediction model.
In order to solve the above technical problems, the present invention also provides a kind of electricity demand forecasting device based on deep learning, comprising:
Training unit, for previously according to history electricity consumption data training prediction model；
Receiving unit, power quantity predicting query time for receiving input and exogenous characteristic parameter；
Computing unit, for the power quantity predicting query time and the exogenous characteristic parameter to be inputted the prediction mould
Type exports electricity demand forecasting value；
Wherein, the history electricity consumption data includes the exogenous characteristic of historical time sequence and the historical time sequence
According to, the prediction model be specially include shot and long term memory network LSTM embedded block sequence to sequence seq2seq model.
In order to solve the above technical problems, the present invention also provides a kind of electricity demand forecasting equipment based on deep learning, comprising:
Memory, for storing instruction, described instruction include the electricity consumption described in above-mentioned any one based on deep learning
The step of prediction technique；
Processor, for executing described instruction.
Electricity demand forecasting method provided by the present invention based on deep learning, using including shot and long term memory network LSTM
The sequence of embedded block to sequence seq2seq model be prediction model, using include historical time sequence and exogenous characteristic
The training of history electricity consumption data obtain the prediction model, and electricity consumption is predicted by the prediction model.In the prior art
Electricity demand forecasting is carried out using the ARIMA model for being only concerned about temporal characteristics, prediction result deviation is larger, and since shot and long term is remembered
Network LSTM allows exogenous feature injection model, and this programme increase shot and long term memory network LSTM embedded block enhances sequence and arrives
Sequence seq2seq model can learning-oriented and expressivity, make prediction model other than temporal characteristics, go back exogenous feature progress
Study and expression, increase the constraint to the Multiple factors for influencing electricity consumption trend, to obtain closer to electricity consumption reality
The prediction model of situation, and then obtained more accurate electricity demand forecasting result.The present invention also provides one kind to be based on depth
The electricity demand forecasting device and equipment of habit have above-mentioned beneficial effect, and details are not described herein.
Detailed description of the invention
It, below will be to embodiment or existing for the clearer technical solution for illustrating the embodiment of the present invention or the prior art
Attached drawing needed in technical description is briefly described, it should be apparent that, the accompanying drawings in the following description is only this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 (a) is the first first kind business electrical amount curve graph；
Fig. 1 (b) is second of first kind business electrical amount curve graph；
Fig. 2 (a) is the first second class business electrical amount curve graph；
Fig. 2 (b) is second of second class business electrical amount curve graphs；
Fig. 3 (a) is the first third class business electrical amount curve graph；
Fig. 3 (b) is second of third class business electrical amount curve graph；
Fig. 4 is the flow chart of the first electricity demand forecasting method based on deep learning provided in an embodiment of the present invention；
Fig. 5 is the neural unit structural schematic diagram of shot and long term memory network LSTM；
Fig. 6 is a kind of flow chart of specific embodiment of step S40 in Fig. 4 provided in an embodiment of the present invention；
Fig. 7 (a) is that a kind of lag data point provided in an embodiment of the present invention introduces schematic illustration；
Fig. 7 (b) is a kind of prediction principle schematic diagram introduced after lag data point provided in an embodiment of the present invention；
Fig. 8 is a kind of structural schematic diagram of the electricity demand forecasting device based on deep learning provided in an embodiment of the present invention；
Fig. 9 is a kind of structural schematic diagram of the electricity demand forecasting equipment based on deep learning provided in an embodiment of the present invention.
Specific embodiment
Core of the invention is to provide a kind of electricity demand forecasting method, device and equipment based on deep learning, compared to
The prior art improves the accuracy of electricity demand forecasting.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 (a) is the first first kind business electrical amount curve graph；Fig. 1 (b) is that second of first kind business electrical amount is bent
Line chart；Fig. 2 (a) is the first second class business electrical amount curve graph；Fig. 2 (b) is second of second class business electrical amount curves
Figure；Fig. 3 (a) is the first third class business electrical amount curve graph；Fig. 3 (b) is second of third class business electrical amount curve graph；
Fig. 4 is the flow chart of the first electricity demand forecasting method based on deep learning provided in an embodiment of the present invention；Fig. 5 is shot and long term
The neural unit structural schematic diagram of memory network LSTM.
Fig. 1 (a) to Fig. 3 (b) is vertical with electricity consumption (Electricity Consumption) for horizontal axis with date (Day)
The electricity consumption that axis describes different enterprises changes with time trend.In order to which business electrical situation is better anticipated, by more
The daily power consumption history curve of important electricity consumption enterprise, family is analyzed, according to the auto-correlation coefficient of curve and partial correlation coefficient to song
Line is classified, and discovery enterprise's daily power consumption curve can be divided mainly into three categories:
(1) leveling style curve, as shown in Fig. 1 (a) and Fig. 1 (b), in the industrial electrical user's body that large size continually and steadily produces
It is now obvious.
(2) preiodic type curve, as shown in Fig. 2 (a) and Fig. 2 (b), obvious with the performance of seasonal production.
(3) instability mode curve, as shown in Fig. 3 (a) and Fig. 3 (b), period of change is big, and regularity is poor.
It can be seen that being different from region and trade power consumption prediction, business electrical amount is simultaneously by the common shadow of both macro and micro many factors
It rings, if the analysis enterprise electricity impact factor of energy system, so that it may improve the prediction accuracy of business electrical amount.However pass through
In the prior art outside relevant to sequence can not be added in prediction calculates in such a way that training ARIMA model is predicted
Characteristic parameter, such as festivals or holidays, weather conditions, commodity future price.
Therefore, the embodiment of the present invention uses the sequence including shot and long term memory network LSTM embedded block to sequence seq2seq
Model is as prediction model, so that exogenous characteristic parameter is added.
As shown in figure 4, the electricity demand forecasting method provided in an embodiment of the present invention based on deep learning includes:
S40: previously according to history electricity consumption data training prediction model.
Wherein, history electricity consumption data includes the exogenous characteristic of historical time sequence and historical time sequence, prediction
Model be specially include shot and long term memory network LSTM embedded block sequence to sequence seq2seq model.
Shot and long term memory network LSTM is one kind of Recognition with Recurrent Neural Network RNN.Recognition with Recurrent Neural Network RNN to sequence before
Imformation memory is simultaneously applied in the calculating currently exported, i.e. the output of sequence and the output of previous state is related.And shot and long term is remembered
Recall network LSTM to have the ability that information is removed or added into location mode, be managed by door, including forget door, input
Door, out gate, structure are as shown in Figure 5.Wherein, X is input, and t is current time, and h is implicit layer state.Shot and long term remembers net
The algorithmic formula of network LSTM is as follows:
ft=σ (Wfxxt+Wfhht-1+bf) (1)
it=σ (Wixxt+Wihht-1+bi) (2)
gt=φ (Wgxxt+Wghht-1+bg) (3)
ot=σ (Woxxt+Wohht-1+bo) (4)
st=gt⊙it+st-1⊙ft (5)
ht=φ (st)⊙ot (6)
Wherein, ftIt indicates to forget door；itIndicate input gate；otIndicate out gate；gtIndicate memory unit；stIt is next to be transmitted to
The output result of a LSTM block；htFor the output result for being transmitted to next layer of neuron；σ indicates activation primitive sigmoid；φ
Indicate activation primitive tanh；Wx(including Wfx、Wix、Wgx、Wox) be input layer parameter；Wh(including Wfh、Wih、Wgh、Woh) it is hidden
The parameter of the input containing layer；xtFor input layer input value；ht-1For last moment hidden layer input value；B (including bf、bi、bg、bo) be
Constant.
The embodiment of the present invention predicts time series as prediction model to sequence seq2seq model using sequence.
Sequence is divided into two parts of encoder and decoder to sequence seq2seq model, inputs a Duan Xulie and feature into encoder,
Result after coding, which is input in decoder, to be decoded, and trained values are obtained.
Wherein, shot and long term memory network LSTM embedded block can be used as encoder into sequence seq2seq model in sequence
Using can also be used as decoder application, a kind of preferable application mode is that the encoder and decoder of prediction model are length
Short-term memory network LSTM.
S41: the power quantity predicting query time and exogenous characteristic parameter of input are received.
S42: by power quantity predicting query time and exogenous characteristic parameter input prediction model, electricity demand forecasting value is exported.
After training obtains prediction model, when can be inquired according to the power quantity predicting query time of input and the power quantity predicting
Between corresponding exogenous characteristic parameter carry out electricity demand forecasting, and export electricity demand forecasting value.
Electricity demand forecasting method provided in an embodiment of the present invention based on deep learning, using including shot and long term memory network
The sequence of LSTM embedded block to sequence seq2seq model be prediction model, using include historical time sequence and exogenous feature
The history electricity consumption data training of data obtains the prediction model, and is predicted by the prediction model electricity consumption.Existing skill
Using the ARIMA model progress electricity demand forecasting for being only concerned about temporal characteristics in art, prediction result deviation is larger, and due to shot and long term
Memory network LSTM allows exogenous feature injection model, and this programme increases shot and long term memory network LSTM embedded block and enhances sequence
Arrange sequence seq2seq model can learning-oriented and expressivity, make prediction model other than temporal characteristics, go back exogenous feature
Learnt and expressed, the constraint to the Multiple factors for influencing electricity consumption trend is increased, to obtain closer to electricity consumption
The prediction model of actual state, and then obtained more accurate electricity demand forecasting result.
Fig. 6 is a kind of flow chart of specific embodiment of step S40 in Fig. 4 provided in an embodiment of the present invention.Such as Fig. 6 institute
Show, on the basis of the above embodiments, in another embodiment, step S40 is specifically included:
S60: according to historical time sequence and exogenous characteristic determine training input vector and with training input to
Measure corresponding label value.
Wherein, training input vector includes parameter and characteristic parameter cycle of training.
In training prediction model, training input vector can be the form of three-dimensional vector.Such as input a three-dimensional vector
(1,30,12), wherein parameter 30 indicate training number of days, parameter 12 contain the same day be which working day, month, whether
Which working day whether this day for public holiday, futures and upper one year be, month, be the public holiday, futures, daily
The features such as electricity.Corresponding label value is the daily power consumption (1,30,1) on the same day, and setting batch_size is 16.Sequence is to sequence
The depth of the encoder and decoder of Seq2seq model can be set to 30, and the number of plies of shot and long term memory network LSTM is set as 2 layers.
According to pre-set search and transformation rule, historical time sequence and exogenous characteristic are converted into training
Input vector and label value corresponding with training input vector.Wherein, true according to maximum factor is influenced on user power consumption
Fixed exogenous feature, exogenous feature can be obtained by staff by the history electricity consumption data of analysis user, can also be with
It analyzes to obtain using specific parser.
S61: training input vector is inputted into preset sequence to sequence seq2seq model structure, obtains trained values.
S62: the error minimum value of trained values and label value is determined by preset rules.
In specific implementation, the error minimum value for determining trained values and label value, needs first to determine trained values and label value
Error negative gradient direction, to determine error minimum value according to negative gradient direction.It can be calculated using Adam gradient descent method
The negative gradient direction of trained values and label value can also calculate the negative of trained values and label value using SGD stochastic gradient descent method
Gradient direction.
Wherein, the algorithm principle of Adam gradient descent method is as follows:
Require: step-length ε (it is recommended that default are as follows: 0.001)
Require: the exponential decay rate of moments estimation, ρ1And ρ2In section [0,1) in.(it is recommended that default are as follows: be respectively 0.9
With 0.999)
Require: the small constant δ for numerical stability is (it is recommended that be defaulted as 10@8)
Require: initial parameter θ
Initialize single order and second moment variable s=0, r=0
Initialization time walks t=0
While does not reach stopping criterion do
It is adopted from training set comprising m sample { x(1)..., x(m)Small lot, corresponding target is y(i)。
Calculate gradient:
t←t-1
Update has inclined single order moments estimation: s ← ρ1s+(1-ρ1)g
Update has inclined second order moments estimation: r ← ρ2r+(1-ρ2)g⊙g
Correct the deviation of first moment:
Correct the deviation of second moment:
It calculates and updates:(by element application operating)
Using update: θ ← θ+Δ θ
end while
S63: the sequence is adjusted to the model parameter of sequence seq2seq model structure according to error minimum value, is obtained pre-
Survey model.
The embodiment of the invention provides a kind of specific embodiments of trained prediction model, have refined based on deep learning
The realization step of electricity demand forecasting method, improves practicability in practical applications and referential.
Fig. 7 (a) is that a kind of lag data point provided in an embodiment of the present invention introduces schematic illustration；Fig. 7 (b) is the present invention
Prediction principle schematic diagram after a kind of introducing lag data point that embodiment provides.
On the basis of the above embodiments, in another embodiment, when the length of historical time sequence is greater than threshold value,
When executing step S61, the electricity demand forecasting method based on deep learning further include:
It will be late by data point using the sliding window memory mechanism of fixed weight and introduce preset sequence to sequence seq2seq
Model structure.
Shot and long term memory network LSTM imitates relatively short sequence with extraordinary memory (within 100-300)
Fruit, but when wanting the longer time series of training, can not just be introduced into the same prediction model.Therefore work as historical time sequence
When the length of column is greater than preset threshold value, data point can be will be late by using the sliding window memory mechanism of fixed weight
(lagged data) introduces the memory that preset sequence reinforces neuron to sequence seq2seq model structure, such as current
The input of time series increases the electricity consumption data of corresponding current time previous year, is added and trains as sample characteristics.
In specific implementation, it will be late by data point using the sliding window memory mechanism of fixed weight and introduce preset sequence
To sequence seq2seq model structure, specifically include:
The encoder output value that will be late by data point inputs full articulamentum to reduce dimension, and will reduce the coding after dimension
The input feature vector of device output valve addition decoder；
It averages to preset data point and neighbour's data point, to reduce noise according to mean value and compensate non-uniform
Every.
As shown in Fig. 7 (a) and Fig. 7 (b), the data of last quarter and upper one year are such as wished to introduce into as input data, then
By the data the year before with the two time points before a season after encoder output, full articulamentum is fed to reduce dimension
Degree, and result is added in the input feature vector of decoder, and " feature " here is time point feature, including " upper one year ",
" last quarter ".The sliding window memory mechanism of fixed weight is for two purposes: first is that reducing processing higher-dimension input data
Computation burden, pass through structuring selection input subset, reduce data dimension；Second is that " eliminating the false and retaining the true ", allows task to handle
System, which focuses more on, finds in input data significantly useful information relevant to currently exporting, to improve the quality of output.
Determine significant data point (preset data point) by fixed weight, significant data point and neighbour are averaged with reduce noise and
Compensate non-uniform interval.
Electricity demand forecasting method provided in an embodiment of the present invention based on deep learning, passes through the sliding window of fixed weight
Memory mechanism will be late by data point and introduce prediction model, so as to exceed the memory energy of shot and long term memory network LSTM to length
The time series of power is learnt, and more practical prediction model is obtained.
On the basis of the various embodiments described above, in another embodiment, to improve model prediction accuracy, step is being executed
When S40 training prediction model, specifically include:
Multiple initial predicted models are obtained using the training of multiple groups initialization data；Wherein, initialization data group and initial pre-
Model is surveyed to correspond；
Checkpoint is saved in each initial predicted model；
The forecast power of initial predicted model corresponding with checkpoint is calculated according to checkpoint；
Model Fusion is carried out to each initial predicted model according to each forecast power, obtains prediction model.
Initialization data, that is, model parameter initial value (seed value).Since training obtains on the basis of different initial values
Prediction model there is different performances, multiple groups model parameter can be trained using the different initial value of multiple groups, and from each
Checkpoint is saved in initial predicted model, and the forecast power of initial predicted model corresponding with checkpoint is calculated according to checkpoint,
To carry out Model Fusion to each initial predicted model according to each forecast power, prediction model is obtained.
Electricity demand forecasting method provided in an embodiment of the present invention based on deep learning, first passes through different initialization datas
Training obtains different initialization prediction models, determines each initial predicted model further according to the checkpoint in each initial predicted model
Forecast power, be weighted and averaged according to the forecast power and acquire final model parameter, to improve prediction model
Precision of prediction.
The corresponding each embodiment of electricity demand forecasting method based on deep learning as detailed above, on this basis, this
Invention also discloses the electricity demand forecasting device based on deep learning corresponding with the above method.
Fig. 8 is a kind of structural schematic diagram of the electricity demand forecasting device based on deep learning provided in an embodiment of the present invention.
As shown in figure 8, the electricity demand forecasting device based on deep learning includes:
Training unit 801, for previously according to history electricity consumption data training prediction model；
Receiving unit 802, power quantity predicting query time for receiving input and exogenous characteristic parameter；
Computing unit 803, for by power quantity predicting query time and exogenous characteristic parameter input prediction model, output to be used
Power quantity predicting value；
Wherein, history electricity consumption data includes the exogenous characteristic of historical time sequence and historical time sequence, prediction
Model be specially include shot and long term memory network LSTM embedded block sequence to sequence seq2seq model.
Since the embodiment of device part is corresponded to each other with the embodiment of method part, the embodiment of device part is asked
Referring to the description of the embodiment of method part, wouldn't repeat here.
Fig. 9 is a kind of structural schematic diagram of the electricity demand forecasting equipment based on deep learning provided in an embodiment of the present invention.
As shown in figure 9, bigger difference can be generated because configuration or performance are different by being somebody's turn to do the electricity demand forecasting equipment based on deep learning,
It may include one or more processors (central processing units, CPU) 910 (for example, one or one
The above processor) and memory 920, the 930 (example of storage medium of one or more storage application programs 933 or data 932
Such as one or more mass memory units).Wherein, memory 920 and storage medium 930 can be of short duration storage or lasting
Storage.The program for being stored in storage medium 930 may include one or more modules (diagram does not mark), and each module can
To include to the series of instructions operation in computing device.Further, processor 910 can be set to and storage medium
930 communications execute the series of instructions operation in storage medium 930 in the electricity demand forecasting equipment 900 based on deep learning.
Electricity demand forecasting equipment 900 based on deep learning can also include one or more power supplys 940, one or
More than one wired or wireless network interface 950, one or more input/output interfaces 960, and/or, one or one
The above operating system 931, such as Windows ServerTM, Mac OS XTM, UnixTM,LinuxTM, FreeBSDTMEtc..
Step in electricity demand forecasting method based on deep learning described in above-mentioned Fig. 4 and Fig. 6 is by being based on depth
The electricity demand forecasting device of habit is based on the structure shown in Fig. 9 and realizes.
It is apparent to those skilled in the art that for convenience and simplicity of description, foregoing description based on
The electricity demand forecasting equipment of deep learning and the specific work process of computer readable storage medium, can be real with reference to preceding method
The corresponding process in example is applied, details are not described herein.
In several embodiments provided herein, it should be understood that disclosed method, apparatus, equipment and calculating
Machine readable storage medium storing program for executing, may be implemented in other ways.For example, Installation practice described above is only schematic
, for example, the division of module, only a kind of logical function partition, there may be another division manner in actual implementation, such as
Multiple module or components can be combined or can be integrated into another system, or some features can be ignored or not executed.Separately
A bit, shown or discussed mutual coupling, direct-coupling or communication connection can be through some interfaces, device
Or the indirect coupling or communication connection of module, it can be electrical property, mechanical or other forms.Module as illustrated by the separation member
It may or may not be physically separated, the component shown as module may or may not be physics mould
Block, it can it is in one place, or may be distributed on multiple network modules.It can be selected according to the actual needs
In some or all of the modules achieve the purpose of the solution of this embodiment.
It, can also be in addition, can integrate in a processing module in each functional module in each embodiment of the application
It is that modules physically exist alone, can also be integrated in two or more modules in a module.Above-mentioned integrated mould
Block both can take the form of hardware realization, can also be realized in the form of software function module.
If integrated module is realized and when sold or used as an independent product in the form of software function module, can
To be stored in a computer readable storage medium.Based on this understanding, the technical solution of the application substantially or
Say that all or part of the part that contributes to existing technology or the technical solution can embody in the form of software products
Out, which is stored in a storage medium, including some instructions are used so that a computer equipment
The whole of (can be personal computer, funcall device or the network equipment etc.) execution each embodiment method of the application
Or part steps.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (Read-Only Memory,
ROM), random access memory (Random Access Memory, RAM), magnetic or disk etc. are various can store program
The medium of code.
A kind of electricity demand forecasting method, device and equipment based on deep learning provided by the present invention is carried out above
It is discussed in detail.Each embodiment is described in a progressive manner in specification, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For device disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.It should be pointed out that for those skilled in the art, without departing from the principle of the present invention,
Can be with several improvements and modifications are made to the present invention, these improvement and modification also fall into the protection scope of the claims in the present invention
It is interior.
It should also be noted that, in the present specification, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or apparatus that includes the element.
Claims (10)
1. a kind of electricity demand forecasting method based on deep learning characterized by comprising
Previously according to history electricity consumption data training prediction model；
Receive the power quantity predicting query time and exogenous characteristic parameter of input；
The power quantity predicting query time and the exogenous characteristic parameter are inputted into the prediction model, export electricity demand forecasting
Value；
Wherein, the history electricity consumption data includes the exogenous characteristic of historical time sequence and the historical time sequence,
The prediction model be specially include shot and long term memory network LSTM embedded block sequence to sequence seq2seq model.
2. electricity demand forecasting method according to claim 1, which is characterized in that the encoder of the prediction model and decoding
Device is shot and long term memory network LSTM.
3. electricity demand forecasting method according to claim 1, which is characterized in that described to be instructed previously according to history electricity consumption data
Practice prediction model, specifically include:
According to the historical time sequence and the exogenous characteristic determine training input vector and with the training it is defeated
The corresponding label value of incoming vector；Wherein, the trained input vector includes parameter and characteristic parameter cycle of training；
The trained input vector is inputted into preset sequence to sequence seq2seq model structure, obtains trained values；
The error minimum value of the trained values Yu the label value is determined by preset rules；
The sequence is adjusted to the model parameter of sequence seq2seq model structure according to the error minimum value, is obtained described pre-
Survey model.
4. electricity demand forecasting method according to claim 3, which is characterized in that described to determine the training by preset rules
The error minimum value of value and the label value, specifically includes:
The negative gradient direction that the trained values Yu the label value are calculated using Adam gradient descent method, according to the negative gradient
Direction determines the error minimum value.
5. electricity demand forecasting method according to claim 3, which is characterized in that described to determine the training by preset rules
The error minimum value of value and the label value, specifically includes:
The negative gradient direction of the trained values Yu the label value is calculated using SGD stochastic gradient descent method, according to described negative
Gradient direction determines the error minimum value.
6. electricity demand forecasting method according to claim 3, which is characterized in that when the length of the historical time sequence is big
When threshold value, it is described the trained input vector is inputted into preset sequence to sequence seq2seq model structure when, also wrap
It includes:
It will be late by data point using the sliding window memory mechanism of fixed weight and introduce the preset sequence to sequence seq2seq
Model structure.
7. electricity demand forecasting method according to claim 6, which is characterized in that the sliding window using fixed weight
The lag data point is introduced the preset sequence to sequence seq2seq model structure by memory mechanism, is specifically included:
The encoder output value of the lag data point is inputted into full articulamentum to reduce dimension, and the coding after dimension will be reduced
The input feature vector of device output valve addition decoder；
It averages to preset data point and neighbour's data point, to reduce noise according to the mean value and compensate non-uniform
Every.
8. according to claim 1 to electricity demand forecasting method described in 7 any one, which is characterized in that described previously according to going through
History electricity consumption data trains prediction model, specifically includes:
Multiple initial predicted models are obtained using the training of multiple groups initialization data；Wherein, initialization data group and described initial pre-
Model is surveyed to correspond；
Checkpoint is saved in each initial predicted model；
The forecast power of initial predicted model corresponding with the checkpoint is calculated according to the checkpoint；
Model Fusion is carried out to each initial predicted model according to each forecast power, obtains the prediction model.
9. a kind of electricity demand forecasting device based on deep learning characterized by comprising
Training unit, for previously according to history electricity consumption data training prediction model；
Receiving unit, power quantity predicting query time for receiving input and exogenous characteristic parameter；
Computing unit, for the power quantity predicting query time and the exogenous characteristic parameter to be inputted the prediction model,
Export electricity demand forecasting value；
Wherein, the history electricity consumption data includes the exogenous characteristic of historical time sequence and the historical time sequence,
The prediction model be specially include shot and long term memory network LSTM embedded block sequence to sequence seq2seq model.
10. a kind of electricity demand forecasting equipment based on deep learning characterized by comprising
Memory, for storing instruction, described instruction include the use described in claim 1 to 8 any one based on deep learning
The step of power predicating method；
Processor, for executing described instruction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910109904.0A CN109685290B (en) | 2019-02-11 | 2019-02-11 | Power consumption prediction method, device and equipment based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910109904.0A CN109685290B (en) | 2019-02-11 | 2019-02-11 | Power consumption prediction method, device and equipment based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109685290A true CN109685290A (en) | 2019-04-26 |
| CN109685290B CN109685290B (en) | 2023-06-16 |
Family
ID=66195775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910109904.0A Active CN109685290B (en) | 2019-02-11 | 2019-02-11 | Power consumption prediction method, device and equipment based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685290B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110264251A (en) * | 2019-05-31 | 2019-09-20 | 杭州博钊科技有限公司 | The data organization form and the prediction technique based on multi-task learning for characterizing cash flow |
| CN111178611A (en) * | 2019-12-23 | 2020-05-19 | 广西电网有限责任公司 | Method for predicting daily electric quantity |
| CN111539841A (en) * | 2019-12-31 | 2020-08-14 | 远景智能国际私人投资有限公司 | Electric quantity prediction method, device, equipment and readable storage medium |
| CN111626536A (en) * | 2019-12-26 | 2020-09-04 | 上海理工大学 | Residential electricity consumption energy efficiency evaluation method based on data driving |
| CN111667377A (en) * | 2020-04-24 | 2020-09-15 | 南方电网能源发展研究院有限责任公司 | Full-automatic power consumption prediction method and system |
| CN111797386A (en) * | 2020-06-28 | 2020-10-20 | 四川长虹电器股份有限公司 | Credible user behavior detection method based on Internet of things |
| CN111861032A (en) * | 2020-07-29 | 2020-10-30 | 广州海关技术中心 | Bulk resource type mineral product release risk prediction method based on deep learning |
| CN112116265A (en) * | 2020-09-25 | 2020-12-22 | 国网上海市电力公司 | Industry landscape index construction method driven by electric power data |
| CN112529144A (en) * | 2019-09-17 | 2021-03-19 | 中国科学院分子细胞科学卓越创新中心 | Short-term time sequence prediction-oriented expected learning method and system |
| CN113268403A (en) * | 2021-05-25 | 2021-08-17 | 中国联合网络通信集团有限公司 | Time series analysis and prediction method, device, equipment and storage medium |
| CN114549095A (en) * | 2022-04-25 | 2022-05-27 | 江西师范大学 | Sliding window type electricity selling amount prediction method based on time sequence |
| CN114743072A (en) * | 2022-05-24 | 2022-07-12 | 中国科学院计算机网络信息中心 | Training method of short-term time sequence prediction model |
| CN115202202A (en) * | 2022-06-20 | 2022-10-18 | 山东大学 | Electric equipment control method and system based on artificial intelligence algorithm |
| CN116205382A (en) * | 2023-05-05 | 2023-06-02 | 南方电网科学研究院有限责任公司 | Prediction method and prediction device for electricity consumption, electronic device and electronic equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106886846A (en) * | 2017-04-26 | 2017-06-23 | 中南大学 | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term |
| CN107590567A (en) * | 2017-09-13 | 2018-01-16 | 南京航空航天大学 | Recurrent neural network short-term load prediction method based on information entropy clustering and attention mechanism |
| CN108197751A (en) * | 2018-01-23 | 2018-06-22 | 国网山东省电力公司电力科学研究院 | Seq2seq network Short-Term Load Forecasting Methods based on multilayer Bi-GRU |
| CN108764568A (en) * | 2018-05-28 | 2018-11-06 | 哈尔滨工业大学 | A kind of data prediction model tuning method and device based on LSTM networks |
| CN109214592A (en) * | 2018-10-17 | 2019-01-15 | 北京工商大学 | A kind of Air Quality Forecast method of the deep learning of multi-model fusion |
| US20190220748A1 (en) * | 2016-05-20 | 2019-07-18 | Google Llc | Training machine learning models |
-
2019
- 2019-02-11 CN CN201910109904.0A patent/CN109685290B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190220748A1 (en) * | 2016-05-20 | 2019-07-18 | Google Llc | Training machine learning models |
| CN106886846A (en) * | 2017-04-26 | 2017-06-23 | 中南大学 | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term |
| CN107590567A (en) * | 2017-09-13 | 2018-01-16 | 南京航空航天大学 | Recurrent neural network short-term load prediction method based on information entropy clustering and attention mechanism |
| CN108197751A (en) * | 2018-01-23 | 2018-06-22 | 国网山东省电力公司电力科学研究院 | Seq2seq network Short-Term Load Forecasting Methods based on multilayer Bi-GRU |
| CN108764568A (en) * | 2018-05-28 | 2018-11-06 | 哈尔滨工业大学 | A kind of data prediction model tuning method and device based on LSTM networks |
| CN109214592A (en) * | 2018-10-17 | 2019-01-15 | 北京工商大学 | A kind of Air Quality Forecast method of the deep learning of multi-model fusion |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110264251A (en) * | 2019-05-31 | 2019-09-20 | 杭州博钊科技有限公司 | The data organization form and the prediction technique based on multi-task learning for characterizing cash flow |
| CN112529144A (en) * | 2019-09-17 | 2021-03-19 | 中国科学院分子细胞科学卓越创新中心 | Short-term time sequence prediction-oriented expected learning method and system |
| CN112529144B (en) * | 2019-09-17 | 2023-10-13 | 中国科学院分子细胞科学卓越创新中心 | Predictive learning method and system for short-term time sequence prediction |
| CN111178611A (en) * | 2019-12-23 | 2020-05-19 | 广西电网有限责任公司 | Method for predicting daily electric quantity |
| CN111178611B (en) * | 2019-12-23 | 2022-09-23 | 广西电网有限责任公司 | Method for predicting daily electric quantity |
| CN111626536A (en) * | 2019-12-26 | 2020-09-04 | 上海理工大学 | Residential electricity consumption energy efficiency evaluation method based on data driving |
| CN111539841A (en) * | 2019-12-31 | 2020-08-14 | 远景智能国际私人投资有限公司 | Electric quantity prediction method, device, equipment and readable storage medium |
| CN111667377A (en) * | 2020-04-24 | 2020-09-15 | 南方电网能源发展研究院有限责任公司 | Full-automatic power consumption prediction method and system |
| CN111797386A (en) * | 2020-06-28 | 2020-10-20 | 四川长虹电器股份有限公司 | Credible user behavior detection method based on Internet of things |
| CN111861032A (en) * | 2020-07-29 | 2020-10-30 | 广州海关技术中心 | Bulk resource type mineral product release risk prediction method based on deep learning |
| CN112116265A (en) * | 2020-09-25 | 2020-12-22 | 国网上海市电力公司 | Industry landscape index construction method driven by electric power data |
| CN113268403A (en) * | 2021-05-25 | 2021-08-17 | 中国联合网络通信集团有限公司 | Time series analysis and prediction method, device, equipment and storage medium |
| CN113268403B (en) * | 2021-05-25 | 2023-10-31 | 中国联合网络通信集团有限公司 | Time series analysis and prediction method, device, equipment and storage medium |
| CN114549095A (en) * | 2022-04-25 | 2022-05-27 | 江西师范大学 | Sliding window type electricity selling amount prediction method based on time sequence |
| CN114549095B (en) * | 2022-04-25 | 2022-08-09 | 江西师范大学 | Sliding window type electricity selling amount prediction method based on time sequence |
| CN114743072A (en) * | 2022-05-24 | 2022-07-12 | 中国科学院计算机网络信息中心 | Training method of short-term time sequence prediction model |
| CN115202202A (en) * | 2022-06-20 | 2022-10-18 | 山东大学 | Electric equipment control method and system based on artificial intelligence algorithm |
| CN116205382A (en) * | 2023-05-05 | 2023-06-02 | 南方电网科学研究院有限责任公司 | Prediction method and prediction device for electricity consumption, electronic device and electronic equipment |
| CN116205382B (en) * | 2023-05-05 | 2023-08-22 | 南方电网科学研究院有限责任公司 | Prediction method and prediction device for electricity consumption, electronic device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109685290B (en) | 2023-06-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685290A (en) | Deep learning-based power consumption prediction method, device and equipment | |
| Wang et al. | Bi-directional long short-term memory method based on attention mechanism and rolling update for short-term load forecasting | |
| CN111401433B (en) | User information acquisition method and device, electronic equipment and storage medium | |
| CN107622329A (en) | The Methods of electric load forecasting of Memory Neural Networks in short-term is grown based on Multiple Time Scales | |
| CN108510113A (en) | A kind of application of XGBoost in short-term load forecasting | |
| CN108320016A (en) | Short-term prediction method for building energy consumption | |
| CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
| CN106779859A (en) | A kind of real-time Method for Sales Forecast method of mobile terminal product | |
| CN115099502B (en) | Short-term power load prediction method based on inter-user power consumption behavior similarity | |
| CN113393057A (en) | Wheat yield integrated prediction method based on deep fusion machine learning model | |
| CN112418495A (en) | Building energy consumption prediction method based on longicorn stigma optimization algorithm and neural network | |
| CN110163444A (en) | A kind of water demand prediction method based on GASA-SVR | |
| CN111507505A (en) | Method for constructing reservoir daily input prediction model | |
| CN113610328A (en) | Power generation load prediction method | |
| CN105844353A (en) | Aquatic product price prediction method and device | |
| CN112508286A (en) | Short-term load prediction method based on Kmeans-BilSTM-DMD model | |
| CN116432697A (en) | Time sequence prediction method integrating long-term memory network and attention mechanism | |
| Yu et al. | Ceam: A novel approach using cycle embeddings with attention mechanism for stock price prediction | |
| CN109146553A (en) | Spot Price forecasting system and its method based on multi-density cluster and multicore SVM | |
| CN114091776A (en) | K-means-based multi-branch AGCNN short-term power load prediction method | |
| CN116608551B (en) | Cold load prediction method and device, electronic equipment and storage medium | |
| Atabay et al. | Multivariate time series forecasting using arimax, sarimax, and rnn-based deep learning models on electricity consumption | |
| Zhang et al. | Building energy consumption prediction based on temporal-aware attention and energy consumption states | |
| CN116402194A (en) | Multi-time scale load prediction method based on hybrid neural network | |
| Yu et al. | Clothing sales forecast based on arima-bp neural network combination 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |