CN110009132A - A kind of short-term electric load fining prediction technique based on LSTM deep neural network - Google Patents
A kind of short-term electric load fining prediction technique based on LSTM deep neural network Download PDFInfo
- Publication number
- CN110009132A CN110009132A CN201910161661.5A CN201910161661A CN110009132A CN 110009132 A CN110009132 A CN 110009132A CN 201910161661 A CN201910161661 A CN 201910161661A CN 110009132 A CN110009132 A CN 110009132A
- Authority
- CN
- China
- Prior art keywords
- prediction
- day
- short
- electric load
- lstm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims description 22
- 238000010606 normalization Methods 0.000 claims description 20
- 238000003062 neural network model Methods 0.000 claims description 8
- 238000013135 deep learning Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 6
- 230000014509 gene expression Effects 0.000 claims description 6
- 241001269238 Data Species 0.000 claims description 4
- 238000001556 precipitation Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims description 2
- 238000011017 operating method Methods 0.000 claims 1
- 238000006243 chemical reaction Methods 0.000 abstract 4
- 238000012544 monitoring process Methods 0.000 abstract 2
- 239000000523 sample Substances 0.000 abstract 1
- 238000012360 testing method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000013277 forecasting method Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- 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)
- Feedback Control In General (AREA)
Abstract
A kind of short-term electric load fining prediction technique based on LSTM deep neural network, it includes electric secondary disk cabinet, the temperature sensor in electric secondary disk cabinet is arranged in sensor probe, signal focuses on center, long-range monitoring backstage, temperature sensor is connect by signal wire with the input terminal of the first signal conversion module, the output end of first signal conversion module is connect with the input terminal of second signal conversion module, the output end of second signal conversion module and the signal input part of signal processing module connect, the signal output end of signal processing module is connect with server.The short-term electric load based on LSTM deep neural network proposed the purpose of the invention is to accurate, efficient, good realization electric secondary disk cabinet remote temperature monitoring refines prediction technique.
Description
Technical field
The invention belongs to field of power systems, and in particular to a kind of short-term electric load based on LSTM deep neural network
Refine prediction technique.
Background technique
With the promotion of development and the user demand of electricity market, the safety of power grid and economical operation become to Guan Chong
It wants.Accurate short-term forecast is carried out to electric load, cost of electricity-generating can be reduced with effective guarantee electric power netting safe running, meet and use
Family demand and raising economic results in society.Since power system load has apparent cyclophysis, while influence factor is complicated,
Such as weather, rainfall, therefore select advanced and accurate short-term load forecasting method very necessary.
The modern intelligent method continued to bring out at this stage: wavelet analysis method, artificial neural network method, supporting vector method etc.,
Compared to some traditional short-term load forecasting methods: such as grey method, Kalman filtering method, expert system approach etc., gradually
The speed and precision of load prediction is improved, but there is also certain drawbacks.Such as artificial neural network method, though have stronger
Memory capability and non-linear mapping capability, but it is difficult to its network structure of determination and suitable optimizing parameter of science, there are parts
Minimal point problem.And supporting vector method on the basis of statistical theory is established, auto-selecting parameter and determining for kernel function need
By artificial experience, prediction effect also will affect.This method establishes LSTM deep learning neural network model, has autonomous learning
The strong feature with generalization ability, while the fining meteorological data including the temperature of subregion containing timesharing and rainfall is introduced, to 15min
The 96 point area power load curves a few days ago of time granularity carry out fining prediction.
Summary of the invention
The purpose of the invention is to provide a kind of fining gas introduced including the temperature of subregion containing timesharing and rainfall
Image data, and the short-term electric load fining based on LSTM deep neural network of independent learning ability and generalization ability by force
Prediction technique.
In order to realize above-mentioned technical characteristic, the technical solution adopted by the present invention is that:
A kind of short-term electric load fining prediction technique based on LSTM deep neural network, including following operation step
It is rapid:
1) acquisition prediction input data, when with 0~24 when every 15 minutes 96 electric load power datas form table
Show the vector power of daily electric load curve, choose the 96 point load power daily for the previous period of prediction day and predicts day
24 moment temperature and subregion rainfall form multidimensional characteristic input variable vector, to predict the 96 point load quantity of power of day as defeated
Variable vector out；Wherein, the range that the prediction day before yesterday is taken is bigger, and the model accordingly established is more accurate.
2): the Various types of data as input variable and output variable is uniformly normalized to one by data normalization pretreatment
A section；
3): will predict the 96 point load power of the day before yesterday, predict the 24 moment temperature and subregion prediction of precipitation value normalizing of day
Change data and form multidimensional characteristic input variable, it is short-term to establish LSTM as output variable for the 96 point load performance numbers to predict day
Power load forecasting module specifies the number of shot and long term Memory Neural Networks LSTM input node according to the number of input variable,
The suitable hidden layer node number of setting, and represent the output node number of prediction daily load vector power；
4): consistent with the transfer of the day type on the same day according to the prediction day before yesterday is principle, is selected from the near to the distant by the range prediction time
Select the training sample of specified number；
5): the training sample selected based on step 4 carries out the LSTM Short-term Load Forecasting that step 3 is established
Training, acquisition make the smallest optimal model parameters of training sample global error；
6): after the LSTM Short-term Load Forecasting for obtaining optimized parameter, by the electric load function of the day before yesterday to be predicted
The normalization datas such as rate and the meteorology of day to be predicted obtain day electric load power curve prediction to be predicted as mode input
Value.
In step 1), daily 96 electric load P during 1 year before acquisition prediction dayd=[P1,P2,…,P96]d、24
Moment temperature Td=[T1,T2,…,T24]d, subregion rainfall Hd=[H1,H2,…,HM]dWith date type Sd(1 is working day, 0
For day off) historical data, and prediction 24 moment of same day day temperature Tf, subregion rainfall HfPrediction data and date class
Type Sf, wherein d ∈ { 1,2 ..., D }, D are the total number of days of historical sample, the subregion number that M includes by prediction area.
In step 2), at the load power data that acquire respectively to step 1, temperature record and rainfall product data normalization
Different data, is normalized to that same scale [- 1,1] is inner by reason, if the data before normalization are xi, maximum, smallest sample
Value is respectively x, the data after normalized are xi', it is as follows specifically to normalize formula by sample number N:
In step 3), the LSTM deep learning neural network model of short-term electric load prediction is established, wherein input becomes
Amount chooses prediction 96 electric load normalized value P ' of the day before yesterdayd-1=[P '1,P′2,…,P′96]d-1, prediction 24 moment of day temperature return
One changes data T 'd=[T '1,T′2,…,T′24]dWith the normalization rainfall H ' of all subregiond=[H '1,H′2,…,H′M]d, mould
Shape parameter θ vector unified representation, output variable are to predict the load prediction vector power of 96 points of dayPrediction model is as follows:
The LSTM neural network includes 120+M input layer, 20 hidden layer nodes and 96 output node layers: defeated
Entering 1~96 node in layer is [P '1,P′2,…,P′96]d-1, 97~120 nodes expression [T '1,T′2,…,T′24]d, 121~120+
M node indicates [H '1,H′2,…,H′M]d；Hidden layer node number is determined as 20 according to training effect comprehensive comparison；Output layer
96 points expression
In step 4), shift according to the day type of the prediction day before yesterday and the same day to [Sf-1, Sf] it is unanimously principle, from going through
30 training sample composition training sample set TS are selected from the near to the distant by the range prediction time in history sample dayf:
In step 5), the LSTM short-term electric load prediction mould of training sample and step 3 foundation selected based on step 4
Type is trained the parameter vector θ of LSTM, obtains so that prediction power vectorWith actual power vector PdBetween minimum two
Multiply optimal L STM prediction model parameters when error E minimum
In step 6), the LSTM Short-term Load Forecasting of optimized parameter is obtained, by the electric power of the day before yesterday to be predicted
Load power normalized vector P 'f-1=[P '1,P′2,…,P′96]f-1, day to be predicted meteorological normalization data T 'f=[T '1,
T′2,…,T′24]fWith H 'f=[H '1,H′2,…,H′M]fAs mode input, it is pre- to obtain day electric load power curve to be predicted
Measured value
The present invention has the following technical effect that
The present invention passes through collection electric load, the historical data of weather meteorology and the weather weather prognosis number of corresponding prediction day
According to；And to the advanced row data normalized of the data such as the electric load of collection, meteorology；Specifically, some day since zero point,
It was used as a moment every 15 minutes, then daily power is indicated with 96 points of this day of load power value, will predict the day before yesterday
96 point load performance numbers, the 24 moment temperature for predicting day and the composition multidimensional characteristic input of subregion prediction of precipitation value normalization data
Variable establishes LSTM deep learning network to predict the 96 point load power of day as output variable, and according to the prediction day before yesterday and
The day type transfer on the same day is unanimously principle, selects training sample from the near to the distant by the range prediction time, carries out to LSTM parameter
Training, obtains optimal LSTM prediction model.By the continuous training and optimization to parameter, so that model more refines, in advance
The accuracy of survey is correspondingly improved.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 is that the short-term electric load based on LSTM deep neural network refines prediction technique flow chart；
Fig. 2 is the LSTM neural network structure figure of short-term electric load prediction；
Fig. 3 is according to one embodiment of the invention in 1 daily load true value April in 2018 and LSTM model predication value, RBF
The comparison of model predication value, time series neural network model predicted value；
Fig. 4 is that three kinds of prediction models predict the comparison diagram of error between in April, 2018 according to one embodiment of the invention.
Specific embodiment
As shown in Figure 1, step 1: daily 96 electric load P during 1 year before acquisition prediction dayd=[P1,P2,…,
P96]d, 24 moment temperature Td=[T1,T2,…,T24]d, subregion rainfall Hd=[H1,H2,…,HM]dWith date type Sd(1 is work
Make day, 0 is day off) historical data, and prediction 24 moment of same day day temperature Tf, subregion rainfall HfPrediction data and
Date type Sf, wherein d ∈ { 1,2 ..., D }, D are the total number of days of historical sample, the subregion number that M includes by prediction area.
Step 2: data normalization, due to step 1 acquire load power data, temperature record and rainfall product data it
Between numerical value difference it is bigger, it is inner to need to normalize to different data same scale [- 1,1]；By pretreated input
Sample is xi'；Sample data before normalized is xi, maximum, smallest sample value are respectively x, sample number N, tool
It is as follows that body handles formula:
Step 3: establishing the LSTM deep learning neural network model of short-term electric load prediction, and wherein input variable is selected
Take prediction 96 electric load normalized value P ' of the day before yesterdayd-1=[P '1,P′2,…,P′96]d-1, prediction day 24 moment temperature normalization
Data T 'd=[T '1,T′2,…,T′24]dWith the normalization rainfall H ' of all subregiond=[H '1,H′2,…,H′M]d, model ginseng
Number θ vector unified representation, output variable are to predict the load prediction vector power of 96 points of dayIn advance
It is as follows to survey model:
The LSTM neural network includes 120+M input layer, 20 hidden layer nodes and 96 output node layers: defeated
Entering 1~96 node in layer is [P '1,P′2,…,P′96]d-1, 97~120 nodes expression [T '1,T′2,…,T′24]d, 121~120+
M node indicates [H '1,H′2,…,H′M]d；Hidden layer node number is determined as 20 according to training effect comprehensive comparison；Output layer
96 points expression
According to fig. 2, the LSTM neural network structure of short-term electric load prediction is explained:
(1) it determines which status information deleted by forgeing the Sigmoid layer of door, forgets door to ht-1And xtIt is seen
It examines, to state Ct-1In each element, the value between output one 0~1 determines which information be filtered with this.It is defeated
Enter given data and substitute into following formula:
ft=σ (wf·[ht-1,xt]+bf)
Which (2) determine to update new information into state.This process is made of two parts.First in input gate
The Sigmoid layers of decision information to be updated, then calculate new candidate value by tanh layersAndIt may be added to
In state；Secondly, LSTM, which will be combined the two parts, is used for more new state information；By input data and forget door
In obtainable status information formula it is as follows:
(3) by Ct-1Update CtIn.Specific steps are as follows: Ct-1With ftIt is multiplied, removes the information for determining to delete, add
itWithProduct, obtain new candidate value Ct, it can be changed according to the degree for updating each state.Formula is as follows:
(4) last output content is determined.Output can be based on current state, and carry out the screening of a part.Structure first
One Sigmoid layers of Output Gate (out gate) is made which ingredient of last output state information determined, and then will
State (allows output result projection between -1~+1) and O after passing through tanhtBe multiplied, obtain desired output as a result, to
Obtain the value of Model Parameter:
ot=σ (Wo·[ht-1,xt]+bo)
ht=ot·tanh(Ct)
Step 4: it shifts according to the day type of the prediction day before yesterday and the same day to [Sf-1, Sf] it is unanimously principle, from history sample
30 training sample composition training sample set TS are selected from the near to the distant by the range prediction time in this dayf:
if[Sf-1-i,Sf-i]=[Sf-1,Sf] i=1,2 ..., D
f-i∈TSf
Step 5: the LSTM Short-term Load Forecasting that the training sample and step 3 selected based on step 4 is established,
The parameter vector θ of LSTM is trained, is obtained so that prediction power vectorWith actual power vector PdBetween least square
Optimal L STM prediction model parameters when error E minimum
Step 6: obtaining the LSTM Short-term Load Forecasting of optimized parameter, by the electric load of the day before yesterday to be predicted
Power normalization vector P 'f-1=[P '1,P′2,…,P′96]f-1, day to be predicted meteorological normalization data T 'f=[T '1,T
′2,…,T′24]fWith H 'f=[H '1,H′2,…,H′M]fAs mode input, it is pre- to obtain day electric load power curve to be predicted
Measured value
In the present embodiment, using the actual load data in somewhere on April 1st, 2018 to April 30 as test object, lead to
Cross the method for the invention, load data predicted, and with the methods of RBF, Time Serial Neural Network predicted value into
Row comparison, can be seen that for the prediction result comparison diagram on April 1 from Fig. 3 and predicts load curve and reality using LSTM method
Load curve is more nearly, and no matter from overall trend or local detail all achieves preferable effect, and average opposite prediction misses
Difference is 2.9143%, hence it is evident that lower than 4.9596% and time series neural network model prediction result of RBF model prediction result
4.0488%, forecasting accuracy is higher.From Day Load Curve Forecasting error of the Fig. 4 during test entire in April, 2018
From the point of view of comparison diagram, the prediction error of most test day LSTM models be it is the smallest in three kinds of error prediction models, be still
LSTM is substantially better than other control methods, and the ensemble average prediction error of LSTM model is only 3.8464% during test, is lower than
RBF model average forecasting error 6.052% predicts error 5.294% also below time series neural network model.It is possible thereby to
Verifying, the short-term electric load according to an embodiment of the present invention with deep learning ability refine prediction model, are able to ascend
The precision and fining degree of prediction.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure made by bright specification and accompanying drawing content or equivalent function transformation, are applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (8)
1. a kind of short-term electric load based on LSTM deep neural network refines prediction technique, which is characterized in that including with
Lower operating procedure:
1) acquisition prediction input data, when with 0~24 when formed every 15 minutes 96 electric load power datas and indicate every
The vector power of day electric load curve, choose prediction day 96 point load power daily for the previous period and prediction day 24 when
It carves temperature and subregion rainfall forms multidimensional characteristic input variable vector, to predict that the 96 point load quantity of power of day become as output
Measure vector；
2): data normalization pretreatment will uniformly normalize to an area as the Various types of data of input variable and output variable
Between；
3): the 96 point load power, the 24 moment temperature for predicting day and the subregion prediction of precipitation value that predict the day before yesterday are normalized into number
According to composition multidimensional characteristic input variable, the 96 point load performance numbers to predict day establish LSTM short term power as output variable
Load forecasting model specifies the number of shot and long term Memory Neural Networks LSTM input node, setting according to the number of input variable
Suitable hidden layer node number, and represent the output node number of prediction daily load vector power；
4): consistent with the transfer of the day type on the same day according to the prediction day before yesterday is principle, selects to refer to from the near to the distant by the range prediction time
Determine the training sample of number；
5): the training sample selected based on step 4, the LSTM Short-term Load Forecasting established to step 3 are trained,
Acquisition makes the smallest optimal model parameters of training sample global error；
6): after the LSTM Short-term Load Forecasting for obtaining optimized parameter, by the electric load power of the day before yesterday to be predicted and
The normalization datas such as the meteorology of day to be predicted obtain day electric load power curve predicted value to be predicted as mode input.
2. a kind of short-term electric load based on LSTM deep neural network according to claim 1 refines prediction side
Method, which is characterized in that in step 1), daily 96 electric load P during 1 year before acquisition prediction dayd=[P1,P2,…,
P96]d, 24 moment temperature Td=[T1,T2,…,T24]d, subregion rainfall Hd=[H1,H2,…,HM]dWith date type Sd(1 is work
Make day, 0 is day off) historical data, and prediction 24 moment of same day day temperature Tf, subregion rainfall HfPrediction data and
Date type Sf, wherein d ∈ { 1,2 ..., D }, D are the total number of days of historical sample, the subregion number that M includes by prediction area.
3. a kind of short-term electric load based on LSTM deep neural network according to claim 1 or 2 refines prediction
Method, which is characterized in that in step 2), step 1 is acquired respectively load power data, temperature record and rainfall product data
Different data, is normalized to that same scale [- 1,1] is inner by normalized, if the data before normalization are xi, maximum,
Smallest sample value is respectively x, the data after normalized are x 'i, sample number N, specifically normalization formula is as follows:
4. a kind of short-term electric load based on LSTM deep neural network according to claim 3 refines prediction side
Method, it is characterised in that: in step 3), the LSTM deep learning neural network model of short-term electric load prediction is established, wherein
Input variable chooses prediction 96 electric load normalized value P ' of the day before yesterdayd-1=[P '1,P′2,…,P′96]d-1, prediction 24 moment of day
Temperature normalization data T 'd=[T '1,T′2,…,T′24]dWith the normalization rainfall H ' of all subregiond=[H '1,H′2,…,
H′M]d, model parameter θ vector unified representation, output variable is to predict the load prediction vector power of 96 points of dayPrediction model is as follows:
5. a kind of short-term electric load based on LSTM deep neural network according to claim 4 refines prediction side
Method, which is characterized in that in step 4), shift according to the day type of the prediction day before yesterday and the same day to [Sf-1, Sf] consistent for original
Then, several training samples are selected to form training sample set TS from the near to the distant by the range prediction time from historical sample dayf:
6. a kind of short-term electric load based on LSTM deep neural network described according to claim 1 or 2 or 4 or 5 is fine
Change prediction technique, which is characterized in that in step 5), the LSTM that the training sample and step 3 selected based on step 4 is established is short-term
Power load forecasting module is trained the parameter vector θ of LSTM, obtains so that prediction power vectorWith actual power to
Measure PdBetween minimum mean-square error E minimum when optimal L STM prediction model parameters
7. a kind of short-term electric load based on LSTM deep neural network according to claim 6 refines prediction side
Method, which is characterized in that in step 6), the LSTM Short-term Load Forecasting of optimized parameter is obtained, by the day before yesterday to be predicted
Electric load power normalization vector P 'f-1=[P '1,P′2,…,P′96]f-1, day to be predicted meteorological normalization data T 'f=
[T′1,T′2,…,T′24]fWith H 'f=[H '1,H′2,…,H′M]fAs mode input, day electric load power to be predicted is obtained
Curve prediction value
8. a kind of short-term electric load based on LSTM deep neural network according to claim 4 refines prediction side
Method, which is characterized in that in LSTM deep learning neural network model, which includes 120+M input layer section
Point, 20 hidden layer nodes and 96 output node layers: 1~96 node is [P ' in input layer1,P′2,…,P′96]d-1, 97~
120 nodes indicate [T '1,T′2,…,T′24]d, 121~120+M node expression [H '1,H′2,…,H′M]d；Hidden layer node number
It is determined as 20 according to training effect comprehensive comparison；96 points of expressions of output layer
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910161661.5A CN110009132A (en) | 2019-03-04 | 2019-03-04 | A kind of short-term electric load fining prediction technique based on LSTM deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910161661.5A CN110009132A (en) | 2019-03-04 | 2019-03-04 | A kind of short-term electric load fining prediction technique based on LSTM deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110009132A true CN110009132A (en) | 2019-07-12 |
Family
ID=67166359
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910161661.5A Pending CN110009132A (en) | 2019-03-04 | 2019-03-04 | A kind of short-term electric load fining prediction technique based on LSTM deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110009132A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110543942A (en) * | 2019-08-28 | 2019-12-06 | 广西大学 | Multi-space-time long and short memory depth network accurate prediction method |
| CN113849052A (en) * | 2021-08-20 | 2021-12-28 | 广州云硕科技发展有限公司 | Machine room temperature prediction method and system based on artificial intelligence |
Citations (3)
| 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 |
| CN106971238A (en) * | 2017-03-10 | 2017-07-21 | 东华大学 | The Short-Term Load Forecasting Method of Elman neutral nets is obscured based on T S |
| CN109359786A (en) * | 2018-12-05 | 2019-02-19 | 国网江苏省电力有限公司扬州供电分公司 | A kind of power station area short-term load forecasting method |
-
2019
- 2019-03-04 CN CN201910161661.5A patent/CN110009132A/en active Pending
Patent Citations (3)
| 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 |
| CN106971238A (en) * | 2017-03-10 | 2017-07-21 | 东华大学 | The Short-Term Load Forecasting Method of Elman neutral nets is obscured based on T S |
| CN109359786A (en) * | 2018-12-05 | 2019-02-19 | 国网江苏省电力有限公司扬州供电分公司 | A kind of power station area short-term load forecasting method |
Non-Patent Citations (4)
| Title |
|---|
| 吴润泽;包正睿;宋雪莹;邓伟;: "基于深度学习的电网短期负荷预测方法研究", 现代电力, no. 02 * |
| 彭文 等: "电力市场中基于 Attention-LSTM 的短期负荷预测模型", 《电网技术》, vol. 43, no. 5, pages 1745 - 1751 * |
| 徐立本，姜云飞，张世伟等主编: "《机器学习引论》", 吉林大学出版社, pages: 347 - 348 * |
| 杨金喜;高洁;孔伯骏;吴佳佳;薛晨;: "基于多模型长短时记忆神经网络的电力负荷预测方法研究", 机电信息, no. 30 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110543942A (en) * | 2019-08-28 | 2019-12-06 | 广西大学 | Multi-space-time long and short memory depth network accurate prediction method |
| CN113849052A (en) * | 2021-08-20 | 2021-12-28 | 广州云硕科技发展有限公司 | Machine room temperature prediction method and system based on artificial intelligence |
| CN113849052B (en) * | 2021-08-20 | 2022-11-04 | 广州云硕科技发展有限公司 | Machine room temperature prediction method and system based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084367B (en) | Soil moisture content prediction method based on LSTM deep learning model | |
| CN109272146B (en) | Flood prediction method based on deep learning model and BP neural network correction | |
| CN109299812B (en) | Flood prediction method based on deep learning model and KNN real-time correction | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| CN109583565B (en) | Flood prediction method based on attention model long-time and short-time memory network | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN111665575B (en) | Medium-and-long-term rainfall grading coupling forecasting method and system based on statistical power | |
| CN106951611A (en) | A kind of severe cold area energy-saving design in construction optimization method based on user's behavior | |
| CN111767517B (en) | BiGRU multi-step prediction method, system and storage medium applied to flood prediction | |
| CN109886496B (en) | Agricultural yield prediction method based on meteorological information | |
| CN109143408B (en) | Dynamic region combined short-time rainfall forecasting method based on MLP | |
| CN113705877A (en) | Real-time monthly runoff forecasting method based on deep learning model | |
| CN117977568A (en) | Power load prediction method based on nested LSTM and quantile calculation | |
| CN115495991A (en) | Rainfall interval prediction method based on time convolution network | |
| CN113033081A (en) | Runoff simulation method and system based on SOM-BPNN model | |
| CN115115125A (en) | Photovoltaic power interval probability prediction method based on deep learning fusion model | |
| CN110009132A (en) | A kind of short-term electric load fining prediction technique based on LSTM deep neural network | |
| CN112132334A (en) | Method for predicting yield of urban domestic garbage | |
| CN111859787A (en) | Daily runoff forecasting method based on coupling of meteorological hydrological forecasting factor classification and deep learning model | |
| CN114897264A (en) | Photovoltaic output interval prediction method under small sample scene based on transfer learning | |
| CN115222138A (en) | Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid | |
| CN113268929B (en) | Short-term load interval prediction method and device | |
| CN115238948A (en) | Method and device for predicting power generation capacity of small hydropower station | |
| CN116341391B (en) | Precipitation prediction method based on STPM-XGBoost model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190712 |