CN109934418A - Short-term load forecasting method based on frequency domain decomposition and intelligent algorithm - Google Patents
Short-term load forecasting method based on frequency domain decomposition and intelligent algorithm Download PDFInfo
- Publication number
- CN109934418A CN109934418A CN201910243384.2A CN201910243384A CN109934418A CN 109934418 A CN109934418 A CN 109934418A CN 201910243384 A CN201910243384 A CN 201910243384A CN 109934418 A CN109934418 A CN 109934418A
- Authority
- CN
- China
- Prior art keywords
- component
- algorithm
- frequency domain
- decomposition
- short
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The present invention has that prediction technique is more single for existing short-term load forecasting method, and the not high problem of precision of prediction provides a kind of short-term load forecasting method based on frequency domain decomposition and intelligent algorithm.This method, comprising: decomposed with Load Time Series of the frequency domain decomposition algorithm to original loads data, obtain component diurnal periodicity, cycle component, low frequency component and high fdrequency component；Diurnal periodicity and cycle are predicted using neural network algorithm；Low frequency component is predicted using random forests algorithm；Second decomposition is carried out to high fdrequency component, the low frequency part after decomposition is predicted using neural network algorithm.The Short-term Load Forecasting Model based on frequency domain decomposition that the present invention is mentioned, prediction result have higher precision of prediction compared with Elman neural network, random forest prediction result.
Description
Technical field
The present invention relates to electric system to predict correlative technology field, specifically, being related to a kind of based on frequency domain decomposition and people
The short-term load forecasting method of work intelligent algorithm.
Background technique
Load prediction can be divided into long-term forecast (annual prediction), medium-term forecast (monthly prediction), short-term according to the prediction time limit
Predict (day degree prediction) and ultra-short term prediction (time-division prediction).Short-term load forecasting to how to arrange operation plan, interconnection hand over
Change the important in inhibiting such as power, Optimization of Unit Commitment.Electric load has height fluctuation and randomness in a short time, therefore
Short-term load forecasting is more difficult.With the gradually implementation of China's energy-saving and emission-reduction policy, the precision for improving load prediction is had become
One research topic to become more and more important.
In recent years, there is various load forecasting methods, time series method such as theoretical based on statistical analysis；Root
The neural network made inferences according to logic rules；Establish the branch in Statistical Learning Theory and Structural risk minization basis
Hold vector machine method；Using the combination forecasting method of transformation quantile estimate and gaussian kernel function；Random forest is as newer
Machine learning algorithm is widely used in load prediction research.
One of prior art applies Elman neural network, which is added by that will accept layer as delay operator
It is added to the hidden layer of feedforward network, to realize the purpose of memory and improve network stabilization.Another prior art proposes random gloomy
Lin Fa, adjustment parameter needed for this method is few, has stronger generalization ability, and fast convergence rate, and precision of prediction is high.
Non-linear extremely strong since the enchancement factor of load is excessive, the theoretical basis of conventional method is also limited.In recent years, right
Load sequence, which is first decomposed the method predicted again, becomes research hotspot.There is the prior art to propose Empirical mode decomposition, so
Different prediction techniques are combined to estimate short term afterwards.Also the prior art proposes a kind of based on small echo-atom sparse decomposition
Ultra-short term model improves atom sparse decomposition energy in conjunction with particle swarm optimization algorithm and orthogonal matching pursuit algorithm
Power is more advantageous to the precision of prediction of increasing productivity.Other prior arts, which are also disclosed, replaces sliding average using Akima interpolation method
Value method come handle local function with improve local mean value decompose (Local Mean Decomposition, LMD) algorithm, then make
With generalized regression nerve networks, the trend of each component is predicted, be superimposed each component and obtain the general trend of load sequence.
But existing short-term load forecasting method all has that prediction technique is more single, the not high problem of precision of prediction, anxious
Need a kind of short-term load forecasting method with higher precision.
Summary of the invention
The present invention has that prediction technique is more single, and precision of prediction is not high for existing short-term load forecasting method
Problem provides a kind of short-term load forecasting method based on frequency domain decomposition and intelligent algorithm.
Technical problems to be solved needed for the present invention can be achieved through the following technical solutions:
A kind of short-term load forecasting method based on frequency domain decomposition and intelligent algorithm characterized by comprising
It is decomposed with Load Time Series of the frequency domain decomposition algorithm to original loads data, obtains component diurnal periodicity, week
Periodic component, low frequency component and high fdrequency component；
Diurnal periodicity and cycle are predicted using neural network algorithm；
Low frequency component is predicted using random forests algorithm；
Second decomposition is carried out to high fdrequency component, the low frequency part after decomposition is predicted using neural network algorithm.
It in the present invention, is decomposed with Load Time Series of the frequency domain decomposition algorithm to original loads data, obtains the diurnal
Phase component, cycle component, low frequency component and high fdrequency component, include the following steps:
1) it is decomposed, has been obtained orthogonal using the Load Time Series of Fourier transform pairs original loads data
Harmonic signal；
2) Load Time Series after being decomposed using the periodic characteristic of load variations to step 1) are reconstructed；
3) component diurnal periodicity, cycle component, low frequency component and high frequency are obtained by Fourier transform combination Euler's formula
Component.
In the present invention, in the step 1), following decompose is made to the Load Time Series P (t) of original loads data:
Wherein, N is historical load data number, aiAnd biFor coefficient.
In the present invention, in the step 2), it is w that P (t), which is decomposed into angular frequency,i=2 π × i ÷ N, (i=1,2 ..., N-
1) P (t) is reconstructed such as following formula by component:
P (t)=a0+ D (t)+W (t)+L (t)+H (t),
Wherein, a0+ D (t) is component diurnal periodicity, and W (t) is cycle component, and L (t) is low frequency component, and H (t) is high frequency division
Amount.
In the present invention, the step 3), comprising:
Discrete Fourier Transform is carried out to load sequence, coefficient a is obtained by spectrum valuei, bi；
Utilize inverse Fourier transform combination Euler's formula ejθ=cos θ+jsin θ acquires the Load Time Series after decomposing.
In the present invention, the neural network algorithm is Elman neural network algorithm.
In the present invention, the random forests algorithm, comprising:
Boot strap is used to extract i sample data set at random from original low frequency component data as the son of each decision tree
Sample set, each sample size is identical as original low frequency component data collection, and the data not being sampled every time constitute number outside bag
According to；
Post-class processing is established to each subsample collection respectively, i decision tree is constructed, in building process, for decision
Each node of tree, stochastical sampling original low frequency component data variables set obtain variable subset, according to gini index minimum criteria
Optimal characteristics are chosen from variable subset to be divided；
Every post-class processing recurrence branch growth from the top to the bottom reaches decision after the minimum dimension of setting leaf node
Tree stops growing, and all decision trees are combined into random forest；
The input test data in Random Forest model are respectively predicted sub- test sample collection using i decision tree, are taken every
The average value of a decision tree prediction result is predicted value.
In the present invention, second decomposition is carried out using Mallat algorithm to high fdrequency component.
Short-term load forecasting method based on frequency domain decomposition and intelligent algorithm of the invention, using frequency domain decomposition algorithm
Load data is decomposed, four diurnal periodicity, cycle, low frequency and high frequency components have been obtained.The characteristics of analyzing different components has needle
Prediction techniques different to the selection of property.To regular very strong diurnal periodicity, cycle component, predicted using Elman；Low frequency component
Sample is less, then is predicted using random forest；To fluctuating bigger high fdrequency component first with Mallat algorithm second decomposition, then
More stable component is taken to do sample combination Elman neural network prediction.The short term based on frequency domain decomposition that the present invention is mentioned
Prediction model, prediction result have higher precision of prediction compared with Elman neural network, random forest prediction result.
Detailed description of the invention
The present invention is further illustrated below in conjunction with the drawings and specific embodiments.
Fig. 1 is load forecasting model of the invention.
Fig. 2 is random forest calculation flow chart.
Fig. 3 is the decomposition of Mallat algorithm.
Fig. 4 is original loads sequence.
Fig. 5 is four components after frequency domain decomposition.
Fig. 6 is the prediction result of component diurnal periodicity.
Fig. 7 is the prediction result of cycle component.
Fig. 8 is the prediction result of low frequency component.
Fig. 9 is comparison of the high fdrequency component after mallat is decomposed.
Figure 10 is the prediction result of high fdrequency component.
Figure 11 is frequency domain decomposition model prediction result.
Specific embodiment
In order to make the technical means, creative features, achievement of purpose and effectiveness of the invention easy to understand, below with reference to tool
Body diagram, the present invention is further explained.
Idea of the invention is that being found existing by the analysis of method and actual demand to existing short-term load forecasting
Short-term load forecasting method all has that prediction technique is more single, and the not high problem of precision of prediction provides one kind through the invention
Short-term load forecasting method based on frequency domain decomposition and intelligent algorithm is to solve the above problems.
Referring to Fig. 1, load sequence inherently instability, after frequency domain decomposition, sub-load sequence shows one
Fixed regularity.To promote precision of prediction, according to different features, using different prediction techniques, prediction steps of the present invention
Core is as follows:
It is decomposed with Load Time Series of the frequency domain decomposition algorithm to original loads data, obtains component diurnal periodicity, week
Periodic component, low frequency component and high fdrequency component；
Diurnal periodicity and cycle are predicted using neural network algorithm；
Low frequency component is predicted using random forests algorithm；
Second decomposition is carried out to high fdrequency component, the low frequency part after decomposition is predicted using neural network algorithm.
I.e. by combination appropriate and the periodic feature of load variations, the load component that will change by the fixed cycle,
The load component that can directly extrapolate when prediction is divided into diurnal periodicity, cycle component；By each component period be greater than for 24 hours, it is meteorological because
The load that element etc. becomes related factor slowly is divided into low frequency component；The load component that will change at random can not establish model progress
The part of prediction is divided into high fdrequency component.Since diurnal periodicity, cycle component are the load components changed by the fixed cycle,
Good prediction effect can be reached using many methods, so, master is only to the prediction of residual components after frequency domain decomposition
Problem is wanted, the solution of the present invention will be illustrated in detail below.
Here, using frequency-domain analysis method analysis load data, to frequency domain decomposition algorithm to the load of original loads data
Time series is decomposed, and makees following decompose first with Fourier transform pairs Load Time Series P (t):
In formula, N is historical load data number, aiAnd biFor coefficient.After making Fourier decomposition to P (t), obtain each other
Orthogonal harmonic signal.Ibid, P (t) is decomposed into angular frequency is wi=2 π × i ÷ N, the component of (i=1,2 ..., N-1).Benefit
With the periodic characteristic of load variations, P (t) can be reconstructed into such as following formula:
P (t)=a0+D(t)+W(t)+L(t)+H(t) (2)
In formula, a0+ D (t) constitutes component diurnal periodicity, and W (t) is then cycle component, and diurnal periodicity, cycle component are all
By fixed mechanical periodicity.L (t), H (t) are low frequency component and high fdrequency component respectively.
Following formula is discrete Fourier transform (Discrete Fourier Tran-sform, DFT) and inverse discrete Fourier transform
Change (Inverse Discrete Fourier Transform, IDFT):
X(wi)=N (ai-jbi) (5)
Therefore, the load sequence after Discrete Fourier Transform can obtain coefficient a from spectrum valuei, bi.To reach
Original series are decomposed, a is obtained0The purpose of the sequence of four components of+D (t), W (t), L (t), H (t)；Obtaining coefficient ai, biIt
It, continues decomposition computation afterwards.
Inverse Fourier transform combination Euler's formula ejθ=cos θ+jsin θ acquires the sequence after decomposing:
It is derived according to above-mentioned formula, inverse Fourier transform operator can be used to obtain Decomposition Sequence.
Complementation operation is introduced, indicates x divided by the remainder of y with mod (x, y).Herein by taking electric load day samples at 96 points as an example.
(1) period of D (t) is 96 and is with 1 day load component for mechanical periodicity.a0The angular frequency set of+D (t):
(2) period of W (t) is 7 × 96 and is with 7 days load components for mechanical periodicity.The angular frequency set of W (t) are as follows:
(3) a is deducted0After+D (t), W (t), remaining part is divided into L (t) and H (t).Due to some slowly varying
Correlative factor causes influence, i.e. low frequency component L (t) to load；H (t) then largely embodies the randomness of load.The two
Angular frequency set such as following formula (9) shown in.
Calculative target is the Fourier coefficient a in formula (1)i, bi, and establish and the frequency after Fourier transformation
Compose X (wi) between relationship.
Diurnal periodicity component and cycle component prediction, by frequency domain decomposition obtain diurnal periodicity component and periodic component advise
Rule property is apparent.The present invention predicts it using Elman neural network algorithm, Tansig hidden layer contain 11 neurons and
Purelin output layer contains a neuron.The output of hidden layer is lagged and is remembered by undertaking layer, and is inputted and be connected to implicit
Layer, the method for this linking keeps it sensitive to historical data, internal feedback is also added into network, thus increase its analysis and
Handle a series of ability of multidate informations.Therefore, diurnal periodicity, cycle component can accurately be described.
It is understood that the present invention can also use BP and LSTM nerve net in addition to Elman neural network algorithm
Network, but precision of prediction is all not so good as Elman neural network algorithm.Main reason is that Elman neural network structure is better than BP nerve
Network structure is more suitable for doing load prediction；LSTM neural network is then suitble to processing sequence data and needs more data training essences
Degree could be higher；In addition, there is also the predictions of other neural networks to be suitable for the present invention, it is not illustrated one by one herein.
The prediction of low frequency component, the low frequency component sequence samples obtained after Fourier decomposition are seldom, are not suitable for using nerve
Network algorithm.Consider that relatively small training sample can be used in the bootstrapping method of sampling, the prediction of present invention application random forest is low
Frequency component.As shown in Fig. 2, algorithm flow is as follows:
Step 1 uses boot strap to extract i sample data set at random from initial data as the increment of each decision tree
This collection, each sample size is identical as raw data set, and the data not being sampled every time constitute the outer data of bag.
Step 2 establishes post-class processing to each subsample collection respectively, constructs i decision tree, in building process, for
Each node of decision tree, stochastical sampling initial data variables set obtain variable subset, according to gini index minimum criteria from son
Selection optimal characteristics are concentrated to be divided.
Every post-class processing of step 3 recurrence branch growth from the top to the bottom reaches the minimum dimension of setting leaf node
Decision tree stops growing afterwards, and all decision trees are combined into random forest.
Step 4 input test data in Random Forest model, it is pre- to sub- test sample collection respectively using i decision tree
It surveys, taking the average value of each decision tree prediction result is predicted value.
The prediction of high fdrequency component, the high fdrequency component regularity obtained after Fourier decomposition is not strong, and that the present invention selects is original
Load sequence decomposition result is largely high frequency, so high fdrequency component prediction result directly determines the good of frequency domain decomposition prediction model
It is bad.In view of high-frequency load sequence have higher randomness and fluctuation it is bigger, Direct Modeling predict difficulty it is larger；It is analyzed, is selected
Second decomposition is carried out to high fdrequency component with Mallat algorithm.Low frequency part more gentle after decomposing is chosen, in conjunction with Elman nerve net
Network largely improves precision of prediction to former high-frequency load sequence prediction.
For Mallat algorithm, initiation sequenceIt is broken down into the layer corresponding to different frequency bands, and
And it may be constructed function f:
In formula, φ0nBe it is discrete after Orthogonal Wavelets.
It is extended to
In formula, φ1n、ψ1nIt is that each branch respectively corresponds orthogonal basis.Sequence C1It is former data column C0Decline form, C0With
C1Between information gap D1, φ1kIt is the orthogonal basis of the low frequency subspace after level of decomposition, has
It can also write
It is reduced to
C1=HC0 (15)
It is similar to have
This process is iterated, is had
Therefore, available
Obviously, it can be seen that each step of iteration has
C in formulaj=HCj-1, Dj=GCj-1。
It is the decomposable process of Mallat algorithm above.CjIt is C0Iteration decline decompose form, sampled point is every time than before it
One step halves, and algorithm terminates after L step is decomposed, i.e., walks C in L0It is decomposed into D1..., DLAnd CL, decomposable process is as shown in Figure 3.
Hereinafter, further illustrating the present invention by a specific example.
This example adjusts bore load data, on May totally 22 days 5, -2017 years on the 1st April in 2017 using city, Anhui Province system
Data, sampling period 15min amount to 3360 points.28 day datas predict following 7 days negative as training set before using
Lotus.Fig. 4 shows original loads sequence.
Preceding 28 days 2688 points are used to carry out frequency domain decomposition as sample.In Fig. 5, diurnal periodicity component and cycle component
For rule it is obvious that low frequency component is smooth curve, high fdrequency component fluctuation is very big, but gentler than original loads sequence.
It is based on frequency domain decomposition load forecasting model according to proposed in this paper, trains and predict to decompose using different methods
Sequence afterwards, then after 7 days 672 points as test sample.Prediction result such as Fig. 6~Figure 10 and table 1 are obtained respectively
It is shown.
The prediction result of 1 four components of table
In conjunction with table 1 above, it can be seen that four fractional prediction effects are all fine, and each section prediction result is combined just
It is final prediction result, as shown in figure 11, the average relative error comparison of each prediction day is shown in Table 2.
The average relative error of the prediction day of table 2
Frequency domain decomposition is first passed through it can be seen from Figure 11 table 2, then predicts short-term electric load using different prediction models,
Finally result, which is combined, has higher precision than Elman neural network, random forest method, the results showed that, the present invention
Method be suitable for short-term electric load prediction.
Only the preferred embodiment of the present invention has been described above, but is not to be construed as limiting the scope of the invention.This
Invention is not only limited to above embodiments, and specific structure is allowed to vary.In short, all guarantors in independent claims of the present invention
Made various change is within the scope of the invention in shield range.
Claims (8)
1. the short-term load forecasting method based on frequency domain decomposition and intelligent algorithm characterized by comprising
It is decomposed with Load Time Series of the frequency domain decomposition algorithm to original loads data, obtains component diurnal periodicity, cycle
Component, low frequency component and high fdrequency component；
Diurnal periodicity and cycle are predicted using neural network algorithm；
Low frequency component is predicted using random forests algorithm；
Second decomposition is carried out to high fdrequency component, the low frequency part after decomposition is predicted using neural network algorithm.
2. the short-term load forecasting method according to claim 1 based on frequency domain decomposition and intelligent algorithm, feature
It is: is decomposed with Load Time Series of the frequency domain decomposition algorithm to original loads data, obtains component diurnal periodicity, cycle
Component, low frequency component and high fdrequency component, include the following steps:
1) it is decomposed using the Load Time Series of Fourier transform pairs original loads data, has obtained orthogonal harmonic wave
Signal；
2) Load Time Series after being decomposed using the periodic characteristic of load variations to step 1) are reconstructed；
3) component diurnal periodicity, cycle component, low frequency component and high fdrequency component are obtained by Fourier transform combination Euler's formula.
3. the short-term load forecasting method according to claim 2 based on frequency domain decomposition and intelligent algorithm, feature
It is: in the step 1), following decompose is made to the Load Time Series P (t) of original loads data:
Wherein, N is historical load data number, aiAnd biFor coefficient.
4. the short-term load forecasting method according to claim 3 based on frequency domain decomposition and intelligent algorithm, feature
Be: in the step 2), it is w that P (t), which is decomposed into angular frequency,i=2 π × i ÷ N, the component of (i=1,2 ..., N-1), by P
(t) reconstruct such as following formula:
P (t)=a0+ D (t)+W (t)+L (t)+H (t),
Wherein, a0+ D (t) is component diurnal periodicity, and W (t) is cycle component, and L (t) is low frequency component, and H (t) is high fdrequency component.
5. the short-term load forecasting method according to claim 4 based on frequency domain decomposition and intelligent algorithm, feature
It is: the step 3), comprising:
Discrete Fourier Transform is carried out to load sequence, coefficient a is obtained by spectrum valuei, bi；
Utilize inverse Fourier transform combination Euler's formula ejθ=cos θ+jsin θ acquires the Load Time Series after decomposing.
6. the short-term load forecasting method according to claim 1 based on frequency domain decomposition and intelligent algorithm, feature
Be: the neural network algorithm is Elman neural network algorithm.
7. the short-term load forecasting method according to claim 1 based on frequency domain decomposition and intelligent algorithm, feature
It is: the random forests algorithm, comprising:
Boot strap is used to extract i sample data set at random from original low frequency component data as the subsample of each decision tree
Collection, each sample size is identical as original low frequency component data collection, and the data not being sampled every time constitute the outer data of bag；
Post-class processing is established to each subsample collection respectively, i decision tree is constructed, in building process, for decision tree
Each node, stochastical sampling original low frequency component data variables set obtain variable subset, according to gini index minimum criteria from change
Quantum concentrates selection optimal characteristics to be divided；
Every post-class processing recurrence branch growth from the top to the bottom, decision tree stops after reaching the minimum dimension of setting leaf node
It only grows, all decision trees are combined into random forest；
The input test data in Random Forest model respectively predict sub- test sample collection using i decision tree, take each determine
The average value of plan tree prediction result is predicted value.
8. the short-term load forecasting method according to claim 1 based on frequency domain decomposition and intelligent algorithm, feature
It is: second decomposition is carried out using Mallat algorithm to high fdrequency component.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811208917 | 2018-10-17 | ||
| CN2018112089175 | 2018-10-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109934418A true CN109934418A (en) | 2019-06-25 |
Family
ID=66988464
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910243384.2A Pending CN109934418A (en) | 2018-10-17 | 2019-03-28 | Short-term load forecasting method based on frequency domain decomposition and intelligent algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109934418A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160712A (en) * | 2019-12-06 | 2020-05-15 | 北京国电通网络技术有限公司 | User electricity utilization parameter adjusting method and device |
| CN111523087A (en) * | 2020-04-10 | 2020-08-11 | 北京航空航天大学 | Typhoon intensity long-term change trend analysis method |
| CN111598303A (en) * | 2020-04-20 | 2020-08-28 | 广东电网有限责任公司电力调度控制中心 | Summer short-term load prediction method based on meteorological component decomposition |
| CN112163710A (en) * | 2020-10-09 | 2021-01-01 | 武汉理工大学 | Short-term reactive load prediction method and device based on ensemble empirical mode |
| CN112990826A (en) * | 2021-04-15 | 2021-06-18 | 西南交通大学 | Short-time logistics demand prediction method, device, equipment and readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2001056342A (en) * | 1999-08-19 | 2001-02-27 | Masahiro Takei | Method for processing tomography image through use of wavelet transformation and its device |
| CN105320995A (en) * | 2014-08-19 | 2016-02-10 | 上海交通大学 | Frequency domain decomposition based wind power generation short-term load prediction method and apparatus |
| CN108022001A (en) * | 2017-09-20 | 2018-05-11 | 河海大学 | Short term probability density Forecasting Methodology based on PCA and quantile estimate forest |
-
2019
- 2019-03-28 CN CN201910243384.2A patent/CN109934418A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2001056342A (en) * | 1999-08-19 | 2001-02-27 | Masahiro Takei | Method for processing tomography image through use of wavelet transformation and its device |
| CN105320995A (en) * | 2014-08-19 | 2016-02-10 | 上海交通大学 | Frequency domain decomposition based wind power generation short-term load prediction method and apparatus |
| CN108022001A (en) * | 2017-09-20 | 2018-05-11 | 河海大学 | Short term probability density Forecasting Methodology based on PCA and quantile estimate forest |
Non-Patent Citations (3)
| Title |
|---|
| 刘达等: "基于EEMD和随机森林的月度负荷预测", 《智慧电力》 * |
| 向峥嵘等: "基于小波-神经网络的电力系统短期负荷预测", 《系统仿真学报》 * |
| 温锦斌等: "基于频域分解的短期风电负荷预测", 《电工技术学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160712A (en) * | 2019-12-06 | 2020-05-15 | 北京国电通网络技术有限公司 | User electricity utilization parameter adjusting method and device |
| CN111160712B (en) * | 2019-12-06 | 2023-09-22 | 北京国电通网络技术有限公司 | User electricity consumption parameter adjusting method and device |
| CN111523087A (en) * | 2020-04-10 | 2020-08-11 | 北京航空航天大学 | Typhoon intensity long-term change trend analysis method |
| CN111523087B (en) * | 2020-04-10 | 2021-04-16 | 北京航空航天大学 | Typhoon intensity long-term change trend analysis method |
| CN111598303A (en) * | 2020-04-20 | 2020-08-28 | 广东电网有限责任公司电力调度控制中心 | Summer short-term load prediction method based on meteorological component decomposition |
| CN112163710A (en) * | 2020-10-09 | 2021-01-01 | 武汉理工大学 | Short-term reactive load prediction method and device based on ensemble empirical mode |
| CN112990826A (en) * | 2021-04-15 | 2021-06-18 | 西南交通大学 | Short-time logistics demand prediction method, device, equipment and readable storage medium |
| CN112990826B (en) * | 2021-04-15 | 2021-11-02 | 西南交通大学 | Short-time logistics demand prediction method, device, equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109934418A (en) | Short-term load forecasting method based on frequency domain decomposition and intelligent algorithm | |
| Wang et al. | Deep learning method based on gated recurrent unit and variational mode decomposition for short-term wind power interval prediction | |
| CN109886498B (en) | EMD-GRU short-term power load prediction method based on feature selection | |
| Candogan et al. | Dynamics in near-potential games | |
| CN109871977A (en) | Load forecasting method based on wavelet transform and the minimum support vector machines of optimization | |
| Chen et al. | Kernel minimum error entropy algorithm | |
| Chen et al. | Mean square convergence analysis for kernel least mean square algorithm | |
| Hashemi et al. | A randomized greedy algorithm for near-optimal sensor scheduling in large-scale sensor networks | |
| CN111079899A (en) | Neural network model compression method, system, device and medium | |
| Ro et al. | FedJAX: Federated learning simulation with JAX | |
| CN112508299A (en) | Power load prediction method and device, terminal equipment and storage medium | |
| CN112581172A (en) | Multi-model fusion electricity sales quantity prediction method based on empirical mode decomposition | |
| Cloez et al. | Long-time behavior and Darwinian optimality for an asymmetric size-structured branching process | |
| Cai et al. | Wind power forecasting based on ensemble empirical mode decomposition with generalized regression neural network based on cross-validated method | |
| Liu et al. | Nonlinear regression A* OMP for compressive sensing signal reconstruction | |
| Bertram et al. | Combining mechanism and drift in community ecology: a novel statistical mechanics approach | |
| CN109978217A (en) | Methods of electric load forecasting based on singular spectrum analysis and memory network | |
| Bayestehtashk et al. | Parsimonious multivariate copula model for density estimation | |
| Zhang et al. | An effective wind speed prediction model combining secondary decomposition and regularised extreme learning machine optimised by cuckoo search algorithm | |
| CN116090615A (en) | Bus load noise reduction prediction method based on sequence decomposition | |
| CN115860270A (en) | Network supply load prediction system and method based on LSTM neural network | |
| Kaur et al. | Review of Decision Tree Data mining Algorithms: CART and C4. 5. | |
| CN115619449A (en) | Deep learning model-based fertilizer price prediction method, device and storage medium | |
| CN106899295A (en) | A kind of Randomized Quantizing signal decoding method and system | |
| Shi et al. | Recurrent neural network wind power prediction based on variational modal decomposition improvement |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190625 |