CN109948845A - A kind of distribution network load shot and long term Memory Neural Networks prediction technique - Google Patents
A kind of distribution network load shot and long term Memory Neural Networks prediction technique Download PDFInfo
- Publication number
- CN109948845A CN109948845A CN201910197906.XA CN201910197906A CN109948845A CN 109948845 A CN109948845 A CN 109948845A CN 201910197906 A CN201910197906 A CN 201910197906A CN 109948845 A CN109948845 A CN 109948845A
- Authority
- CN
- China
- Prior art keywords
- distribution network
- network load
- lstm
- prediction
- photovoltaic power
- 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
Classifications
-
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of distribution network load shot and long term Memory Neural Networks prediction technique, it is to solve existing distribution network load prediction and the analytic process of photovoltaic power generation output forecasting is complicated, and be easy the problem of being influenced by correlation and redundancy.The specific steps of the present invention are as follows: step 1, constructs distribution network load and photovoltaic power generation output forecasting model respectively using LSTM；Step 2 automatically extracts historical context data using LSTM, obtains the predicted value of distribution network load and the predicted value of photovoltaic power output；The predicted value of distribution network load is subtracted the predicted value of photovoltaic power output, obtains the predicted value of power distribution network net load by step 3.For method of the invention under the premise of being not necessarily to carry out Feature Engineering, under Various Seasonal, meteorological condition, prediction effect is generally better than the net load prediction technique based on SVR, has good precision of prediction；Method of the invention can effectively meet power distribution network scheduling needs, have good practical value.
Description
Technical field
The present invention relates to power distribution network shot and long term load prediction field, specifically a kind of distribution network load shot and long term memory nerve
Neural network forecast method.
Background technique
Power distribution network short-term load forecasting belongs to the important evidence and premise of power distribution network scheduling, directly affects power distribution network operation
Reliability and security[1,2].Since power distribution network is widely distributed, network is complicated, uncertain strong.Simultaneously as photovoltaic power output by
Outside environmental elements are affected, and especially climatic factor influences to be especially apparent on it.As distributed photovoltaic power directly connects
Enter, after fluctuation power output directly cuts down sub-load, further improves the fluctuation and prediction difficulty of power distribution network net load.
Being usually used in distribution network load prediction and the method for photovoltaic power generation output forecasting includes statistic law, time series method and artificial intelligence
Energy method etc..It is proposed that a kind of statistical analysis technique based on load prediction error characteristics, in conjunction with prediction result and prediction
The error statistics regularity of distribution obtains probabilistic load prediction results.Someone uses in conjunction with chaology phase space reconstruction and builds
The method of vertical exponential model, predicts microgrid load using only demand history data.Factor of the someone for influence load
Multi-source heterogeneous differentiation processing is carried out, carries out predicting and using concurrent operation using SVR method, has taken into account the prediction essence of model
Degree and speed.Someone uses time series method, goes out force data to the photovoltaic comprising seasonal factor and establishes the progress of ARIMA model in advance
It surveys, precision of prediction is higher and error is relatively stable.Someone carries out Wavelet Denoising Method to photovoltaic power output historical data first, according to the time
The general characteristic of sequence data and discrete features carry out feature selecting, carry out photovoltaic power generation output forecasting using multicore SVR, achieve compared with
Good effect.Someone carries out photovoltaic power output using clustering method screening photovoltaic power output characteristic, using GA-BP neural network pre-
It surveys, prediction effect is ideal.Statistic law and time series method belong to conventional method, pass through historical load and photovoltaic power output data and curves
It is analyzed, finds data inherent law and predicted that model is simple, computational efficiency is high, it can be difficult to embodying extraneous factor pair
The influence of distribution network load and photovoltaic power output.Artificial intelligence approach can take out historical data and correlation by machine learning
Linear and nonlinear relationship in data simultaneously shows, and can divide the combined factors for influencing load and photovoltaic power output
Analysis, learning ability is strong, has well adapting to property and precision of prediction in terms of prediction.But traditional intelligence algorithm belongs to shallow-layer
Structure algorithm, it is difficult to represent the complex nonlinear relationship between mass data；Meanwhile when constructing prediction model, precision is by defeated
Enter feature influence, needs complicated feature selecting link to determine optimum prediction characteristic set.
Distribution network load predicts that someone is by using important in random forest mainly by power distribution network user power utilization behavioral implications
The correlation of Functional Analysis distribution network load and other influences factor is spent, the features such as temperature, dew point constitute distribution network load
Predicted characteristics collection.Photovoltaic is contributed, and fluctuation is big, and obvious by meteorological factor influence, someone is meteorological special using the analysis of Pearson's Y-factor method Y
Correlation between sign and photovoltaic power output, and similar day is chosen by calculating Euclidean distance and cosine similarity, determine cloud amount, it is wet
Degree, wind speed, the features such as air pressure are used for photovoltaic power generation output forecasting.The building of traditional characteristic set is general rule of thumb or based on feature phase
The analysis of closing property determines optimal characteristics set for distribution network load prediction and photovoltaic power generation output forecasting, and analytic process is complicated, and is easy
It is influenced by correlation and the problems such as redundancy.
Compared to traditional artificial intelligence method, shot and long term remembers (Long Short Term Memory, LSTM) as new
Generation deep learning method, carrying out analysis to time series data has advantage, excavates the inner link between time series data, energy emphatically
Enough adaptive building high-dimensional feature prediction models, and validity feature is automatically extracted according to training data characteristic, it is potential to excavate its
Correlation, is suitable for the complicated high-dimensional prediction model of building as a result, and people are also carrying out relevant research.
Summary of the invention
The purpose of the present invention is to provide a kind of distribution network load shot and long term Memory Neural Networks prediction techniques, on solving
State the problem of proposing in background technique.
To achieve the above object, the invention provides the following technical scheme:
A kind of distribution network load shot and long term Memory Neural Networks prediction technique, the specific steps are as follows:
Step 1 constructs distribution network load and photovoltaic power generation output forecasting model using LSTM respectively；
Step 2 automatically extracts historical context data using LSTM, realizes and predicts in conjunction with Short-term characteristic for a long time, effectively
The precision of prediction for improving distribution network load and photovoltaic power output obtains the predicted value of distribution network load and the predicted value of photovoltaic power output；
The predicted value of distribution network load is subtracted the predicted value of photovoltaic power output, obtains the pre- of power distribution network net load by step 3
Measured value.
As a further solution of the present invention: by historical context before building distribution network load and photovoltaic power generation output forecasting model
Property data are standardized.
As a further solution of the present invention: specific step is as follows for standardization: by each column historical context number
According to value range specification between [0,1], to historical context data sequence y1,y2K ynStandardization formula it is as follows:WhereinIndicate the mean value of the column historical context data,S is
The standard deviation of historical context data.
As a further solution of the present invention: in step 2 historical context data include continuous historical load data, when
Between data and temperature data.
As a further solution of the present invention: the historical context data after standardization being divided into training set, are tested
Card collection and 3 part of test set, the effect for verifying collection is to assess in regulating networks hyper parameter network.
As a further solution of the present invention: verifying collection being verified using K folding interior extrapolation method: data are divided into K
Subregion；Then the historical context data after standardization are turned into K identical models, for each model, by K-1
Subregion is as its training set, and remaining 1 subregion is as verifying collection；Finally, using the mean value of K result as to Algorithm Error
Verification result.
As a further solution of the present invention: the construction step of distribution network load prediction model is as follows: being built using Keras
LSTM model, time_step 24, corresponding 24 hours daily loads；Batch_size is 100, i.e., using 100 groups of samples as
One overall calculation loss function；Input_dim is identical as input data characteristic, and LSTM layers are 1 layer in hidden layer, hides section
Points are 300, in order to reduce influence of the over-fitting to prediction model, add dropout regularization, and dropout value is 0.2,
Activation primitive use Sigmoid function, loss function select mean absolute error (Mean Absolute Deviation,
MAE):Wherein xi indicates true value,Indicate predicted value.
As a further solution of the present invention: the construction step of photovoltaic power generation output forecasting model is as follows: being built using Keras
LSTM model, time_step 10, corresponding daily 10 hours photovoltaics power output；Batch_size be 40, input_dim with it is defeated
It is identical to enter data characteristics number, LSTM layers are 1 layer in hidden layer, and concealed nodes number is 300, in order to reduce over-fitting,
Dropout value is 0.2, and activation primitive is Sigmoid function, and loss function selects MAE.
Compared with prior art, the beneficial effects of the present invention are:
The present invention carries out short-term prediction using load and photovoltaic power output of the LSTM to power distribution network respectively, obtains height by making the difference
The power distribution network net load prediction result of precision, measured data experiment show method of the invention without before carrying out Feature Engineering
It puts, under Various Seasonal, meteorological condition, prediction effect is generally better than the net load prediction technique based on SVR, has good
Precision of prediction；
Method of the invention can effectively meet power distribution network scheduling needs, have good practical value.
Detailed description of the invention
Fig. 1 is the work flow diagram of distribution network load shot and long term Memory Neural Networks prediction technique.
Fig. 2 is the memory unit structure schematic diagram of LSTMD in distribution network load shot and long term Memory Neural Networks prediction technique.
Fig. 3 is 23-29 days in January, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out distribution network load prediction respectively.
Fig. 4 is in March, 2017 6-12 bu in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Figure is not compared using the result that LSTM and SVR model carries out distribution network load prediction.
Fig. 5 is 17-23 days in July, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out distribution network load prediction respectively.
Fig. 6 is 23-29 days in October, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out distribution network load prediction respectively.
Fig. 7 is 23-29 days in January, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out photovoltaic power generation output forecasting respectively.
Fig. 8 is in March, 2017 6-12 bu in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Figure is not compared using the result that LSTM and SVR model carries out photovoltaic power generation output forecasting.
Fig. 9 is 17-23 days in July, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out photovoltaic power generation output forecasting respectively.
Figure 10 is in October, 2017 23-29 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Day compares figure using the result that LSTM and SVR model carries out photovoltaic power generation output forecasting respectively.
Figure 11 is 23-29 days in January, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out the prediction of power distribution network net load respectively.
Figure 12 is 6-12 days in March, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out the prediction of power distribution network net load respectively.
Figure 13 is 17-23 days in July, 2017 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Compare figure using the result that LSTM and SVR model carries out the prediction of power distribution network net load respectively.
Figure 14 is in October, 2017 23-29 in the embodiment 1 of distribution network load shot and long term Memory Neural Networks prediction technique
Day compares figure using the result that LSTM and SVR model carries out the prediction of power distribution network net load respectively.
Specific embodiment
The technical solution of the patent is explained in further detail With reference to embodiment.
Embodiment 1
1、LSTM
Deep learning belongs to a branch of machine learning, belongs to newer method, the company of relying primarily in data study
Continuous layer is learnt, these layers indicate and learnt usually using neural network model.Wherein, shot and long term is remembered
(LSTM) Processing with Neural Network time series problem effect is preferable, therefore, selects it to carry out distribution network load prediction and goes out with photovoltaic
Power prediction.LSTM belongs to one kind of Recognition with Recurrent Neural Network (Recurrent Neural Network, RNN), and special character exists
In: RNN only has the function of that memory is temporary, and LSTM has both shot and long term memory function.
Network training is participated in jointly with current input after the influence quantization for the content generation that RNN can input the past, i.e.,
The output of neuron is taken back into neuron as input again.But it needs the content of the state of current hidden layer and front is multiple
It is associated, causes operand rapidly with the increase of index rank, the training time of model greatly prolongs.I.e. with time interval
Increase, loses ability of the study apart from current point in time information farther out.LSTM solves the long-term Dependence Problem of RNN, special
Point is to be added to input gate in each layer structural unit of RNN, forget the valves such as door and out gate.Pass through " door ", control memory shape
State, the information for storing any time and distance, solve the problems, such as that RNN neural network only has short-term memory not have long-term memory,
The memory unit structure of LSTM is as shown in Figure 2.
It is C from unit input in Fig. 2t-1, right side output is Ct；Ct-1By multiplier multiplied by coefficient ft, then carry out common
SYSTEM OF LINEAR VECTOR superposition after export Ct。ftCalculation formula is；ft=Sigmoid (Wfg[ht-1,xt]+bf).The process that function calculates
It is exactly to realize the process for forgeing door, the value of Sigmoid function output is 1, then remembers completely, and 0 is forgets completely, and median is note
Recall and forgetting ratio, WfAnd bfAs undetermined coefficient, obtained by training study.
Other two sub-neural network layer is expressed as follows: it=Sigmoid (Wig[ht-1,xi]+bi), C 't=tanh (WCg
[ht-1,xt]+bc).Wherein, tanh function can be mapped to result [- 1,1], wherein parameter WcAnd bcAgain by training
Learn obtained coefficient.
Later, by Ct-1With C 'tLinear superposition calculating is carried out, thereby determines that the C of outputtIn how much information from this is defeated
Enter, how much information is from last legacy information: Ct=ft*Ct-1+it*C′t。
The h finally exportedtIt is divided into two parts, a part is output to next layer unit, and a part is output to one under same layer
A unit, may be expressed as: Ot=Sigmoid (W0g[ht-1,xt]+b0), ht=Ot*tanh(Ct)。
Wherein OtTo forget door, determine which information will be deleted from this unit, the C of outputtVector passes through and forgets door
Operation output is ht.Gate mechanism more than use, LSTM are just provided with the function of long-term memory.
2, data normalization and data set building
To improve model generalization ability, data need to be standardized, i.e., advised the value range of each column data
Model is between [0,1].To sequences y1,y2K ynStandardization formula it is as follows:WhereinIt indicates
The mean value of the column data,S is characterized the standard deviation of data.
Before training LSTM model, the historical data after standardization is divided into training set, verifies collection and test set 3
Point.Wherein, the effect for verifying collection is to assess in regulating networks hyper parameter network.In order to avoid verifying collection is too small or even
The so adjusting of property affecting parameters and the reliable assessment to model take K folding cross-validation method to be verified[12].Specific practice are as follows:
Data are divided into K subregion；Then, K identical models of instantiation, for each model, using K-1 subregion as its instruction
Practice collection, remaining 1 subregion is as verifying collection；Finally, using the mean value of K result as the verification result to Algorithm Error.This
It is 4 that K is taken in embodiment.
3, sample calculation analysis
To verify the validity of LSTM algorithm and advanced, respectively with LSTM and support vector regression (Support
Vector Regression, SVR) based on, power distribution network net load prediction model is constructed, and carry out the prediction of distribution net load.Choosing
2013-2017 In A Certain Locality, Jiangsu Province area power distribution network hour load, photovoltaic power output and other relevant weather characteristic history data are taken to carry out
Experiment.
Distribution network load prediction
LSTM is respectively adopted and SVR method establishes distribution network load prediction model.Input the characteristic such as table 1 of LSTM model
It is shown.Since LSTM can automatically analyze correlation between historical load feature and prediction target, characteristic set is herein with continuous
Historical load, temperature etc. combine date feature to construct, and do not carry out preferred.
1 distribution network load LSTM prediction model input feature vector of table
Using Keras deep learning program library, distribution hourly load forecasting is carried out, chooses Jiangsu distribution 2013-2017
Year hour load and temperature profile historical data, totally 43824 groups of data.Using 2017 annual datas as test set；In order in training
The hyper parameter of regulating networks in the process, reliably assessment models, carry out assessment verifying to model using 4 folding cross-validation methods, will
The historical data of 2013-2016 divides training set and verifying collection, division proportion 3:1 as unit of year.
In order to verify the validity of model, LSTM model prediction result and SVR model prediction result are compared into test,
The input feature vector selection of model as shown in table 1, builds LSTM model using Keras, major parameter setting are as follows: time_step is
24, corresponding 24 hours daily loads；Batch_size is 100, i.e., 100 groups of samples are calculated to loss letter as a whole
Number；Input_dim is identical as input data characteristic.LSTM layers are 1 layer in hidden layer, and concealed nodes number is 300.In order to reduce
Dropout regularization is added in influence of the over-fitting to prediction model, and dropout value is 0.2, and activation primitive uses Sigmoid
Function, loss function select mean absolute error (Mean Absolute Deviation, MAE):
Wherein xiIndicate true value,Indicate predicted value.
The input feature vector of SVR model is as shown in table 2, and the feature of output is Lh, indicate the prediction of current time distribution network load
Value, establishes SVR model and is predicted.
2 distribution network load SVR prediction model input feature vector of table
To verify new method validity, a weekly data is respectively extracted from four seasons in 2017, uses LSTM and SVR respectively
Model is predicted that experimental result is shown in Fig. 3-6.
In order to preferably embody comparison as a result, increase mean absolute percentage error (Mean Absolute Percent
Error, MAPE) it compares:Wherein xiIndicate true value,Indicate predicted value.At this time
The error comparison of the result and true value predicted using LSTM and SVR method is shown in Table 3.
3 Distribution Network Load Data precision of prediction of table compares
From table 3 and Fig. 3-6 as can be seen that the distribution network load prediction result of 168 hours shows LSTM in each season
The precision of prediction of method prediction result is above SVR method, is more suitable for distribution network load prediction.
Photovoltaic power generation output forecasting
Thinking is identical as distribution network load prediction, carries out photovoltaic power generation output forecasting, LSTM mould respectively using LSTM and SVR method
For the input feature vector set of type and SVR model respectively such as table 4 and table 5, the feature of output is Ph, indicate that current time photovoltaic power output is pre-
Measured value.
4 photovoltaic of table power output LSTM prediction model input feature vector
5 photovoltaic of table power output SVR prediction model input feature vector
Since solar irradiance is low before and after sunrise, and it is related to photovoltaic power generation starting Threshold, when photovoltaic power generation output forecasting
Section actual selection is 08:00-17:00, daily 10 hours.LSTM model, major parameter setting and 3.1 are built using Keras
Middle load forecasting model choosing method is similar: time_step 10, corresponding daily 10 hours photovoltaics power output；batch_size
It is identical as input data characteristic for 40, input_dim.LSTM layers are 1 layer in hidden layer, and concealed nodes number is 300.In order to subtract
Few over-fitting, dropout value are 0.2, and activation primitive is Sigmoid function, and loss function selects MAE.Meanwhile it using
SVR method predicts photovoltaic power output.Predicted value that two methods obtain and true value compare as is seen in figs 7-10, MAE and
MAPE error comparison such as table 6.
6 photovoltaic power generation output forecasting application condition of table
From table 6 and Fig. 7-10 as can be seen that each season photovoltaic power generation output forecasting the result shows that, the prediction error of SVR method is high
In LSTM method.Thus, in terms of predicting photovoltaic power output, LSTM method equally has advantage.
The prediction of power distribution network net load
When not considering other distributed generation resources in addition to distributed photovoltaic in power distribution network, distribution network load goes out with photovoltaic
The difference of power is power distribution network net load, it may be assumed that Lh'=Lh-Ph, wherein Lh' indicate power distribution network moment net load, LhIndicate power distribution network
Certain moment load, PhIndicate power distribution network moment photovoltaic power output.By can be calculated the power distribution network net load based on LSTM and SVR
As illustrated in figs. 11-14, each season power distribution network net load prediction error comparison is as shown in table 7 for predicted value and true value curve.
7 two methods of table predict net load value and true value application condition
From table 7 and Figure 11-14 as can be seen that MAE and MAPE is comprehensively compared as a result, in the prediction short-term net load side of power distribution network
Face, LSTM Various Seasonal prediction result have good precision of prediction and are superior to SVR.The method of the present invention is demonstrated as a result,
Validity and advance.
The present invention carries out short-term prediction using load and photovoltaic power output of the LSTM to power distribution network respectively, obtains height by making the difference
The power distribution network net load prediction result of precision.Measured data experiment shows method of the invention without before carrying out Feature Engineering
It puts, under Various Seasonal, meteorological condition, prediction effect is generally better than the net load prediction technique based on SVR, has good
Precision of prediction.Method of the invention can effectively meet power distribution network scheduling needs, have good practical value.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.No
It should treat any reference in the claims as limiting the claims involved.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped
Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should
It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
The other embodiments being understood that.
Claims (8)
1. a kind of distribution network load shot and long term Memory Neural Networks prediction technique, which is characterized in that specific step is as follows:
Step 1 constructs distribution network load prediction model and photovoltaic power generation output forecasting model using LSTM respectively；
Step 2 automatically extracts historical context data using LSTM and substitutes into distribution network load prediction model and photovoltaic power generation output forecasting
Model obtains the predicted value of distribution network load and the predicted value of photovoltaic power output；
The predicted value of distribution network load is subtracted the predicted value of photovoltaic power output, obtains the predicted value of power distribution network net load by step 3.
2. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 1, which is characterized in that described
Historical context data are standardized by building distribution network load with before photovoltaic power generation output forecasting model.
3. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 2, which is characterized in that described
Specific step is as follows for standardization: by the value range specification of each column historical context data between [0,1], to going through
History relevance data sequence y1,y2K ynStandardization formula it is as follows:WhereinIndicate the column
The mean value of historical context data,S is the standard deviation of historical context data.
4. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 1, which is characterized in that described
Historical context data include continuous historical load data, time data and temperature data in step 2.
5. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 2, which is characterized in that will mark
Standardization treated historical context data are divided into training set, verifying collection and 3 part of test set.
6. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 5, which is characterized in that testing
Card collection is verified using K folding interior extrapolation method: data are divided into K subregion；Then by the historical context after standardization
Data turn to K identical models, for each model, using K-1 subregion as its training set, remaining 1 subregion conduct
Verifying collection；Finally, using the mean value of K result as the verification result to Algorithm Error.
7. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 1, which is characterized in that described
The construction step of distribution network load prediction model is as follows: building LSTM model, time_step 24, batch_ using Keras
Size is 100, input_dim identical as input data characteristic, and LSTM layers are 1 layer in hidden layer, and concealed nodes number is 300,
Dropout regularization is added, activation primitive uses Sigmoid function, and loss function selects mean absolute error MAE:Wherein xiIndicate true value,Indicate predicted value.
8. distribution network load shot and long term Memory Neural Networks prediction technique according to claim 7, which is characterized in that described
The construction step of photovoltaic power generation output forecasting model is as follows: building LSTM model, time_step 10 using Keras；batch_size
Identical as input data characteristic for 40, input_dim, LSTM layers are 1 layer in hidden layer, and concealed nodes number is 300, addition
Dropout regularization, activation primitive are Sigmoid function, and loss function selects MAE.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910197906.XA CN109948845A (en) | 2019-03-15 | 2019-03-15 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910197906.XA CN109948845A (en) | 2019-03-15 | 2019-03-15 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109948845A true CN109948845A (en) | 2019-06-28 |
Family
ID=67009985
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910197906.XA Pending CN109948845A (en) | 2019-03-15 | 2019-03-15 | A kind of distribution network load shot and long term Memory Neural Networks prediction technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948845A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110474339A (en) * | 2019-08-07 | 2019-11-19 | 国网福建省电力有限公司 | A kind of electric network reactive-load control method based on the prediction of depth generation load |
| CN111582542A (en) * | 2020-03-31 | 2020-08-25 | 国网上海市电力公司 | Power load prediction method and system based on abnormal restoration |
| CN111598289A (en) * | 2020-03-30 | 2020-08-28 | 国网河北省电力有限公司 | Distributed optimization method of integrated energy system considering LSTM photovoltaic output prediction |
| CN111612253A (en) * | 2020-05-22 | 2020-09-01 | 广西电网有限责任公司 | Power distribution network net load prediction method and system based on Bayesian network |
| CN111680826A (en) * | 2020-05-14 | 2020-09-18 | 沂南力诺太阳能电力工程有限公司 | Photovoltaic power generation capacity prediction analysis method |
| CN111709554A (en) * | 2020-05-22 | 2020-09-25 | 广西电网有限责任公司 | Method and system for joint prediction of net loads of power distribution network |
| CN111738773A (en) * | 2020-07-01 | 2020-10-02 | 国网宁夏电力有限公司 | New energy and load-based net load peak-valley time interval dividing method and system |
| CN111985719A (en) * | 2020-08-27 | 2020-11-24 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN112134304A (en) * | 2020-09-22 | 2020-12-25 | 南方电网数字电网研究院有限公司 | Micro-grid full-automatic navigation method, system and device based on deep learning |
| CN112785088A (en) * | 2021-02-25 | 2021-05-11 | 西安理工大学 | Short-term daily load curve prediction method based on DCAE-LSTM |
| CN113011086A (en) * | 2021-03-02 | 2021-06-22 | 西南林业大学 | Estimation method of forest biomass based on GA-SVR algorithm |
| CN113988415A (en) * | 2021-10-28 | 2022-01-28 | 河北工业大学 | Medium-and-long-term power load prediction method |
| CN114066013A (en) * | 2021-10-21 | 2022-02-18 | 国网浙江省电力有限公司台州供电公司 | Net load prediction method and device for new energy power market |
| CN114841457A (en) * | 2022-05-18 | 2022-08-02 | 上海玫克生储能科技有限公司 | Power load estimation method and system, electronic device, and storage medium |
| CN116339899A (en) * | 2023-05-29 | 2023-06-27 | 内江师范学院 | Desktop icon management method and device based on artificial intelligence |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678415A (en) * | 2016-01-05 | 2016-06-15 | 湖南大学 | Method for predicting net load of distributed power supply power distribution network |
| CN106960252A (en) * | 2017-03-08 | 2017-07-18 | 深圳市景程信息科技有限公司 | Methods of electric load forecasting based on long Memory Neural Networks in short-term |
| CN108280551A (en) * | 2018-02-02 | 2018-07-13 | 华北电力大学 | A kind of photovoltaic power generation power prediction method using shot and long term memory network |
| CN108416690A (en) * | 2018-01-19 | 2018-08-17 | 中国矿业大学 | Load Forecasting based on depth LSTM neural networks |
| CN109284870A (en) * | 2018-10-08 | 2019-01-29 | 南昌大学 | Short-term method for forecasting photovoltaic power generation quantity based on shot and long term Memory Neural Networks |
-
2019
- 2019-03-15 CN CN201910197906.XA patent/CN109948845A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678415A (en) * | 2016-01-05 | 2016-06-15 | 湖南大学 | Method for predicting net load of distributed power supply power distribution network |
| CN106960252A (en) * | 2017-03-08 | 2017-07-18 | 深圳市景程信息科技有限公司 | Methods of electric load forecasting based on long Memory Neural Networks in short-term |
| CN108416690A (en) * | 2018-01-19 | 2018-08-17 | 中国矿业大学 | Load Forecasting based on depth LSTM neural networks |
| CN108280551A (en) * | 2018-02-02 | 2018-07-13 | 华北电力大学 | A kind of photovoltaic power generation power prediction method using shot and long term memory network |
| CN109284870A (en) * | 2018-10-08 | 2019-01-29 | 南昌大学 | Short-term method for forecasting photovoltaic power generation quantity based on shot and long term Memory Neural Networks |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110474339A (en) * | 2019-08-07 | 2019-11-19 | 国网福建省电力有限公司 | A kind of electric network reactive-load control method based on the prediction of depth generation load |
| CN110474339B (en) * | 2019-08-07 | 2022-06-03 | 国网福建省电力有限公司 | Power grid reactive power control method based on deep power generation load prediction |
| CN111598289A (en) * | 2020-03-30 | 2020-08-28 | 国网河北省电力有限公司 | Distributed optimization method of integrated energy system considering LSTM photovoltaic output prediction |
| CN111582542A (en) * | 2020-03-31 | 2020-08-25 | 国网上海市电力公司 | Power load prediction method and system based on abnormal restoration |
| CN111582542B (en) * | 2020-03-31 | 2023-10-03 | 国网上海市电力公司 | Power load prediction method and system based on anomaly repair |
| CN111680826A (en) * | 2020-05-14 | 2020-09-18 | 沂南力诺太阳能电力工程有限公司 | Photovoltaic power generation capacity prediction analysis method |
| CN111612253A (en) * | 2020-05-22 | 2020-09-01 | 广西电网有限责任公司 | Power distribution network net load prediction method and system based on Bayesian network |
| CN111709554A (en) * | 2020-05-22 | 2020-09-25 | 广西电网有限责任公司 | Method and system for joint prediction of net loads of power distribution network |
| CN111738773A (en) * | 2020-07-01 | 2020-10-02 | 国网宁夏电力有限公司 | New energy and load-based net load peak-valley time interval dividing method and system |
| CN111985719A (en) * | 2020-08-27 | 2020-11-24 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN111985719B (en) * | 2020-08-27 | 2023-07-25 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN112134304A (en) * | 2020-09-22 | 2020-12-25 | 南方电网数字电网研究院有限公司 | Micro-grid full-automatic navigation method, system and device based on deep learning |
| CN112785088A (en) * | 2021-02-25 | 2021-05-11 | 西安理工大学 | Short-term daily load curve prediction method based on DCAE-LSTM |
| CN112785088B (en) * | 2021-02-25 | 2023-06-30 | 西安理工大学 | DCAE-LSTM-based short-term daily load curve prediction method |
| CN113011086A (en) * | 2021-03-02 | 2021-06-22 | 西南林业大学 | Estimation method of forest biomass based on GA-SVR algorithm |
| WO2023065553A1 (en) * | 2021-10-21 | 2023-04-27 | 国网浙江省电力有限公司台州供电公司 | Net load prediction method and apparatus for new energy electric power market |
| CN114066013A (en) * | 2021-10-21 | 2022-02-18 | 国网浙江省电力有限公司台州供电公司 | Net load prediction method and device for new energy power market |
| CN113988415A (en) * | 2021-10-28 | 2022-01-28 | 河北工业大学 | Medium-and-long-term power load prediction method |
| CN113988415B (en) * | 2021-10-28 | 2024-06-04 | 河北工业大学 | Medium-and-long-term power load prediction method |
| CN114841457A (en) * | 2022-05-18 | 2022-08-02 | 上海玫克生储能科技有限公司 | Power load estimation method and system, electronic device, and storage medium |
| CN116339899A (en) * | 2023-05-29 | 2023-06-27 | 内江师范学院 | Desktop icon management method and device based on artificial intelligence |
| CN116339899B (en) * | 2023-05-29 | 2023-08-01 | 内江师范学院 | Desktop icon management method and device based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948845A (en) | A kind of distribution network load shot and long term Memory Neural Networks prediction technique | |
| CN109711620B (en) | Short-term power load prediction method based on GRU neural network and transfer learning | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Wu et al. | Interpretable wind speed prediction with multivariate time series and temporal fusion transformers | |
| Ke et al. | Short-term electrical load forecasting method based on stacked auto-encoding and GRU neural network | |
| CN112116144B (en) | Regional power distribution network short-term load prediction method | |
| CN110751318B (en) | Ultra-short-term power load prediction method based on IPSO-LSTM | |
| Dumitru et al. | Solar photovoltaic energy production forecast using neural networks | |
| CN112163689A (en) | Short-term load quantile probability prediction method based on depth Attention-LSTM | |
| CN113822481A (en) | Comprehensive energy load prediction method based on multi-task learning strategy and deep learning | |
| CN113537582B (en) | Photovoltaic power ultra-short-term prediction method based on short-wave radiation correction | |
| Tian et al. | An adaptive ensemble predictive strategy for multiple scale electrical energy usages forecasting | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| Massaoudi et al. | A novel approach based deep RNN using hybrid NARX-LSTM model for solar power forecasting | |
| CN111222689A (en) | LSTM load prediction method, medium, and electronic device based on multi-scale temporal features | |
| CN116526473A (en) | Particle swarm optimization LSTM-based electrothermal load prediction method | |
| CN115358437A (en) | Power supply load prediction method based on convolutional neural network | |
| CN111815039A (en) | Weekly scale wind power probability prediction method and system based on weather classification | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| Du et al. | An innovative interpretable combined learning model for wind speed forecasting | |
| CN113609762A (en) | Electric cooling and heating load combined prediction method and system based on GRU-MTL | |
| CN115481788B (en) | Phase change energy storage system load prediction method and system | |
| CN116090635A (en) | Meteorological-driven new energy generation power prediction method | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| Xu et al. | A framework for electricity load forecasting based on attention mechanism time series depthwise separable convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190628 |
|
| RJ01 | Rejection of invention patent application after publication |