CN109063911A - A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks - Google Patents
A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks Download PDFInfo
- Publication number
- CN109063911A CN109063911A CN201810876297.6A CN201810876297A CN109063911A CN 109063911 A CN109063911 A CN 109063911A CN 201810876297 A CN201810876297 A CN 201810876297A CN 109063911 A CN109063911 A CN 109063911A
- Authority
- CN
- China
- Prior art keywords
- load
- gru
- group
- prediction
- kinds
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 230000002776 aggregation Effects 0.000 title claims abstract description 44
- 238000004220 aggregation Methods 0.000 title claims abstract description 44
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 35
- 239000011159 matrix material Substances 0.000 claims abstract description 29
- 230000005611 electricity Effects 0.000 claims abstract description 16
- 238000007637 random forest analysis Methods 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 14
- 230000003595 spectral effect Effects 0.000 claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 13
- 230000002123 temporal effect Effects 0.000 claims abstract description 3
- 230000008569 process Effects 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000007500 overflow downdraw method Methods 0.000 abstract description 3
- 239000010410 layer Substances 0.000 description 10
- 238000012549 training Methods 0.000 description 10
- 230000015654 memory Effects 0.000 description 9
- 230000000694 effects Effects 0.000 description 8
- 238000001228 spectrum Methods 0.000 description 6
- 240000002853 Nelumbo nucifera Species 0.000 description 4
- 235000006508 Nelumbo nucifera Nutrition 0.000 description 4
- 235000006510 Nelumbo pentapetala Nutrition 0.000 description 4
- 230000004913 activation Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 238000007796 conventional method Methods 0.000 description 4
- 238000013277 forecasting method Methods 0.000 description 4
- 238000005096 rolling process Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 239000000047 product Substances 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 229910052739 hydrogen Inorganic materials 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 244000131316 Panax pseudoginseng Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000000155 melt Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002365 multiple layer Substances 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000005624 perturbation theories Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000004579 scanning voltage microscopy Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000005211 surface analysis Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000005612 types of electricity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The Load aggregation body regrouping prediction method based on gating cycle unit networks that the present invention relates to a kind of, include the following steps: using adapter distribution spectral clustering, customer charge data are clustered, to obtain the similar electricity consumption group of multiple part throttle characteristics, and acquire each group's load characteristic matrix；Three kinds of GRU networks are built, and the temporal aspect by extracting group is trained three kinds of GRU networks, obtains the prediction model of three kinds of GRU networks, Model Fusion is carried out to three kinds of GRU networks by random forests algorithm, obtains the load forecasting model of each group；Moment feature to be predicted is input in load forecasting model, the predicted load of each group is respectively obtained, different groups predicted value is summed to obtain the predicted value of final Load aggregation body；The present invention makes it possible to hold user load characteristics and changing rule, precision of prediction is high, strong applicability by introducing regrouping prediction, deep neural network, Model Fusion method.
Description
Technical field
The invention belongs to Load Prediction In Power Systems technical fields more particularly to a kind of based on gating cycle unit networks
Load aggregation body regrouping prediction method.
Background technique
Quickly and accurately load prediction is great to the safe and economic operation effect of electric system.Traditional load prediction mostly according to
Level division, such as system-level, bus grade, power transformation station level, micro-capacitance sensor grade etc. are carried out according to the physical structure that electric system measures, is led to
Often other levels are not applied for for the load forecasting method of a certain specific level exploitation.In recent years, with intelligent electric meter
Universal, Utilities Electric Co. can obtain the fine-grained customer charge data of magnanimity.Based on intelligent electric meter data, electricity can be got rid of
The limitation of Force system measuring structure can flexibly divide the Load aggregation body of different scales on demand and carry out load prediction, namely
It, can also be according to regional (such as building, cell, block, plot), row other than the Load Prediction In Power Systems of traditional level can be achieved
Industry (such as resident, industry and commerce), electricity price type (timesharing, peak valley etc.) etc. form Load aggregation body and carry out load prediction, to meet
The load prediction demand more refined.
The prediction of Load aggregation body is the bottom-up load forecasting method based on intelligent electric meter.Load aggregation body
It can divide on demand, more flexibly, but different demarcation method will lead to huge, the traditional load prediction side of prediction object scale difference
Method is only applicable to specific load scale, does not have generalization ability.Especially when load scale reduces, due to small-scale load
Population effect weaken, the mean absolute error percentage of load prediction (mean absolute percentage error,
MAPE) index is significantly improved with the reduction of prediction scale, therefore traditional prediction technique is not suitable for the prediction of Load aggregation body.
Flexible, scalable is divided for Load aggregation body, contacts the features such as close with user load characteristics, proposes that there is applicability
, high-precision prediction technique, be the difficult point of Load aggregation body prediction.
Since intelligent electric meter data and user load characteristics are in close relations, by clustering, find to bear between different user
Lotus changes general character rule, and Load aggregation body is divided into multiple electricity consumption groups accordingly, carries out modeling analysis, energy for load group
Improve the precision of prediction of Load aggregation body.In prediction technique, BP neural network, support vector machines (support vector
Machine, SVM) etc. be widely applied in load prediction.These algorithms are established between output and input by training
Non-linear relation, convert static modelling problem for dynamic time modeling problem.But it as typical time series data, bears
Lotus variation has dynamic characteristic, i.e. load variations rule is also changed by the past period in addition to being influenced by current time state
The influence of process.Conventional method mostly using similar day, the historical data of typical day as inputting, can not consider load variations when
Feature in sequence causes load prediction error larger.Potential electricity consumption Behavior law is found from user's history electricity consumption data, and
The development and change for speculating load are developed by data, are the key that accurate progress load predictions.With the development of deep learning, with
Shot and long term remembers recurrent neural network (the recurrent neural that (long short-term memory, LSTM) is representative
Network, RNN) it can be considered that correlation between time series, can be more fully described the variation of time series
Journey is widely applied in multiple fields such as speech recognition, natural language processings.LSTM is used for wind-powered electricity generation and resident load is pre-
It surveys, it was demonstrated that it can consider the inherent law of time series development evolvement, catch the substantive characteristics of time series, to improve prediction
Precision.But LSTM haves the shortcomings that the training time long, also, since Load aggregation body includes a variety of part throttle characteristics, different is negative
The neural network structure that lotus characteristic is applicable in is different.Therefore, these problems are based on, propose that the model based on deep neural network melts
The advantage that conjunction method, sufficiently integration utilize heterogeneous networks structure, helps to improve Load aggregation body precision of prediction.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, propose a kind of prediction side suitable for Load aggregation body
Method enables mentioned method to hold user load characteristics by introducing regrouping prediction, deep neural network, Model Fusion method
And changing rule, precision of prediction is high, strong applicability.
The present invention solves its technical problem and adopts the following technical solutions to achieve:
A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks, the regrouping prediction method include
Following steps:
(1) adapter distribution spectral clustering is used, customer charge data are clustered, to obtain multiple loads
The similar electricity consumption group of characteristic, and acquire each group's load characteristic matrix；
(2) the different GRU network of three kinds of structures is built, and the temporal aspect by extracting group is different to three kinds of structures
GRU network is trained, and obtains the prediction model of three kinds of GRU networks, carries out mould to three kinds of GRU networks by random forests algorithm
Type fusion, obtains the load forecasting model of each group；
(3) moment feature to be predicted is input in load forecasting model obtained in step (2), respectively obtains each group
The predicted load of body sums different groups predicted value to obtain the predicted value of final Load aggregation body.
It should be noted that in the step (1) electricity consumption group acquisition process are as follows: to each customer charge data by week
Mean value is taken, section [0,1] is zoomed to by maximum-Returning to one for minimum value, a load characteristic curve is obtained to each user, it will
All user personality curves are integrated into matrix, and carrying out cluster for matrix can be obtained electricity consumption group.
It may be noted that GRU network is made of input layer, output layer and hidden layer in the step (2), wherein hidden layer packet
Containing multiple cascade GRU units；The GRU unit includes a resetting door, and a update door is controlled defeated by door control mechanism
Out, recall info.
In addition, build the GRU network of three kinds of structures in the step (2) respectively to load group, it is deep by control network
Degree and the polymeric low frequency of GRU element number Learning work load, intermediate frequency, high-frequency characteristic, and learn different frequency domain load variations features,
The output of three deep neural networks is merged eventually by random forests algorithm.
The advantages and positive effects of the present invention are:
1, the present invention proposes to be grouped prediction to Load aggregation body according to part throttle characteristics, and by distributed spectrum clustering algorithm
On grouping and clustering applied to Load aggregation body, compares traditional K-means algorithm and improve clustering precision and stability, overcome
Single machine spectral clustering calculating speed is slow, the disadvantage more than committed memory；Model Fusion thought is applied to the pre- of Load aggregation body
It surveys, using the GRU deep neural network of different structure as meta-model, the dynamic modeling of time series is realized, by random
Forest algorithm merges multiple meta-models, can make full use of heterogeneous networks design feature, further increases load prediction essence
Degree；
2, compared to conventional methods such as BP, SVM, regrouping prediction+Model Fusion prediction technique of the invention is poly- to load
Fit forecasting problem has applicability, and all has higher precision of prediction under the conditions of different load scale；It is pre- by rolling
The time scale of prediction can be adjusted flexibly in the form of survey, on the prediction scale of 30min-24h, load forecasting method of the invention
All have higher precision of prediction.
Detailed description of the invention
Technical solution of the present invention is described in further detail below with reference to drawings and examples, but should
Know, these attached drawings are designed for task of explanation, therefore not as the restriction of the scope of the invention.In addition, except non-specifically
It points out, these attached drawings are meant only to conceptually illustrate structure construction described herein, without to be drawn to scale.
Fig. 1 is load dendrogram；
Fig. 2 is the DB index of distributed spectrum clustering algorithm and K-means algorithm with number of clusters figure of changing；
Fig. 3 is three kinds of GRU network structures；
Fig. 4 is the prediction architecture diagram based on GRU network and Model Fusion；
Fig. 5 is that distinct methods predict error comparative situation figure under different user quantity；
Fig. 6 is four kinds of method precision of prediction MAPE with predicted time dimensional variation figure；
Specific embodiment
Firstly, it is necessary to which explanation, illustrates specific structure of the invention, feature and excellent for by way of example below
Point etc., however what all descriptions were intended merely to be illustrated, and should not be construed as to present invention formation any restrictions.This
Outside, any single technical characteristic for being described by or implying in each embodiment mentioned by this paper, still can be in these technologies spy
Continue any combination between sign (or its equivalent) or delete, to obtain the sheet that may do not referred to directly herein
More other embodiments of invention.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, in addition, term " includes " and " having " and their any deformation, it is intended that covering is not arranged
His includes, for example, the process, method, system, product or equipment for containing a series of steps or units are not necessarily limited to clearly
Those of list step or unit, but may include be not clearly listed or for these process, methods, product or equipment
Intrinsic other step or units.
In the description of the present invention, it should be noted that term " center ", "upper", "lower", "left", "right", "vertical",
The orientation or positional relationship of the instructions such as "horizontal", "inner", "outside" be the invention product using when the orientation or position usually put
Relationship is merely for convenience of description of the present invention and simplification of the description, rather than the device or element of indication or suggestion meaning must have
There is specific orientation, be constructed and operated in a specific orientation, therefore is not considered as limiting the invention.In addition, term " the
One ", " second ", " third " etc. are only used for distinguishing description, are not understood to indicate or imply relative importance.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.
The present invention just is illustrated in conjunction with Fig. 1 to Fig. 6 below.
Embodiment 1
Fig. 1 is load dendrogram；Fig. 2 is the DB index of distributed spectrum clustering algorithm and K-means algorithm with number of clusters
Figure of changing；Fig. 3 is three kinds of GRU network structures；Fig. 4 is the prediction architecture diagram based on GRU network and Model Fusion；Fig. 5
Error comparative situation figure is predicted for distinct methods under different user quantity；Fig. 6 be four kinds of method precision of prediction MAPE with prediction when
Between dimensional variation figure；As shown in figs. 1 to 6, the Load aggregation body grouping provided in this embodiment based on gating cycle unit networks
Prediction technique carries out the prediction of Load aggregation body by taking the intelligent electric meter data set of London as an example, and detailed process is as follows:
One, it is clustered using the load characteristics clustering method that distributed spectrum clusters
Firstly, taking mean value by week to each customer charge data Lm, section is zoomed to by maximum-Returning to one for minimum value
[0,1] obtains a load characteristic curve to each user in this way, and all user personality curves are integrated into Matrix CM×T, for
Matrix CM×TCarrying out cluster can be obtained electricity consumption group, calculate by intelligent electric meter typical 30 minute sampling interval, then each user
Indicatrix contains 48 × 7=336 point, and data dimension is very high, and number of users to be analyzed is more, causes traditional clustering method smart
Degree, speed, stability are unable to satisfy analysis and require.Spectral clustering then overcome the traditional clustering algorithms such as K-means can only identify it is convex
The data of spherical shape distribution, and the shortcomings that local optimum may be fallen into, it can be clustered on arbitrary shape sample, and converge on complete
Office is optimal, and essence is that clustering problem is converted to the optimal dividing problem of figure.Spectral clustering process is as follows:
By Matrix CM×TAcquire the Euclidean distance matrix H of n user between any twom,n, calculation formula is as follows:
In formula, Hm,nIndicate the Euclidean distance between m-th of user and nth user, Hm,nFor symmetrical matrix, and diagonal line
Element is 0.
Hm, the similarity matrix A of n are constructed using Gaussian function:
In formula, σmAnd σnAdaptively refer in practical operation for adaptive scale parameter and preassign several scale ginsengs
The value of number σ, executes spectral clustering respectively, finally chooses the σ for keeping cluster result best as parameter.
And then Laplacian Matrix L can be constructed:
L=D-1/2AD-1/2 (4)
According to perturbation theory, optimal classification number is determined by calculating the characteristic value of similarity matrix, if what is determined is optimal
Classification number is k, then its corresponding k feature vector is X1,X2,…,Xk, then gained eigenvectors matrix X=(X1,X2,…,
Xk).K-means clustering is used to gained eigenmatrix, obtains final electricity consumption group division result.
Spectral clustering calculation amount when calculating similar matrix and finding k feature vector is maximum, and occupies more
Memory space.To overcome the shortcomings of in traditional spectral clustering efficiency, distributed spectrum clustering algorithm is sparse similar using arest neighbors
Matrix replaces former similar matrix, while calculating feature vector using the distributed computing framework based on MapReduce.Firstly,
All n/p data point keys having the same are arranged in each map stage in the matrix that n/p row is stored on p node,
Reduce stage each node calculates local data and input xiDistance:
Wherein xjIt indicates the local data of node, to guarantee that gained distance matrix has symmetry, is arranged in the map stage
Two keys return to position No. and respective distance, respectively to determine each element position.Parallelization calculating between node is so that problem
Complexity is reduced to (7) by (6):
O(n2d+n2logt) (6)
O(n2d/p+(n2logt)/p) (7)
As shown in formula (2), same parallel step is also used to the calculating of similar matrix, obtains the phase of a rarefaction
Like matrix.The characteristic value of similar matrix is sought computationally intensive, and committed memory is more, since the rarefaction of similar matrix in spectral clustering is special
Point is solved the PARPACK algorithm of characteristic value using parallelization, is deployed in all calculate nodes respectively.Parallel computation characteristic value
Complexity formula (9) are reduced to by formula (8):
O(m3)+(O(nm)+O(nt))×O(m-k) (8)
O(m3)+(O(mn/p)+O(nt/p))×O(m-k) (9)
Spectral clustering realizes grouping by carrying out cluster to feature vector, and the present embodiment uses the K-means algorithm of parallelization
Realize sorting procedure.The built-in K-means algorithm of distributed version, can be called by MLlib packet in Spark.If p is
Calculate node number, distributed K-means theory of algorithm computation complexity are only the 1/p of single machine version.
2676 users in the intelligent electric meter data set of London are clustered, as seen from Figure 1: between different user group
There are apparent otherness, but it is similar in there is general character；(a) types of populations belongs to the comparatively stable type of electricity consumption, and load is whole
Body is in reduced levels；And (b) types of populations shows that such uses the head of a household then on the contrary, minimum load state remains at higher level
The phase equipment of work is more；(c), there is apparent two electricity consumption spikes in (e) types of populations sooner or later, but peak period and big
It is small different；(d) fluctuation of types of populations is stronger.
The calculated performance of lower the mentioned algorithm of surface analysis, contrast method are conventional individual version K-means algorithm, experimental ring
Border is 8 Node distribution formula computing clusters (Intel Xeon E7-8850v2*8,16G Registered DDR3 memory * 16).With
Davies-Bouldin index (DB) is used as cluster result evaluation index, DB Index Definition are as follows:
In formula,For average distance in the class of i-th, j class, wi,wjThe respectively cluster centre of two classes, the smaller meaning of DB
Taste inter- object distance is smaller, between class distance is bigger, Clustering Effect is better.
To guarantee to be comparable between cluster result, number of clusters is specified manually, since the selection of initial point will affect
Cluster result is repeated 10 times cluster to two methods of each specified number of clusters, records corresponding DB index.Two methods DB
Index is with number of clusters situation of change as shown in Fig. 2, from figure 2 it can be seen that corresponding to different number of clusters, spectral clustering is calculated
The DB value of method is respectively less than the DB value of K-means algorithm, illustrates that spectral clustering Clustering Effect is better than the poly- of K-means algorithm
Class effect.
Two, it is designed based on the prediction model of GRU network and Model Fusion
GRU network is made of input layer, output layer and hidden layer.Wherein hidden layer includes multiple cascade GRU units.
GRU unit includes a resetting door rt, a update door zt, the information such as output, memory are controlled by door control mechanism,
It makes prediction in current time step.GRU workflow is as follows: each moment, and GRU unit receives current state by updating door
xtWith the hidden state h at a upper momentt-1, after receiving input information, by matrix operation, neuron is determined by activation primitive
Whether activate.Similarly, resetting door equally receives xtWith ht-1, operation result determines that how many past information needs passes into silence.
Current time input is superimposed by operation with resetting door output, forms current memory content h ' by activation primitivet.Current memory
h′tH is inputted with backt-1By updating the dynamic control of door, the output content h of final door control unit is determinedt, while htAlso will
It is transmitted in next GRU unit.Calculation formula between each variable is as follows:
zt=σ (W(z)xt+U(z)ht-1) (11)
rt=σ (W(r)xt+U(r)ht-1) (12)
In formula, W(z)With U(z)Indicate the weight of update door, W(r)With U(r)Indicate the weight of forgetting door, W indicates to be formed with U
Network weight when current memory, σ (x) are activation primitive sigmoid, and tanh (x) is activation primitive tanh.
After obtaining GRU network, it is trained by the back-propagation algorithm (BPTT) being temporally unfolded.The present embodiment
Select the mean absolute error (MAE) of predicted load as loss function:
There are larger differences for load fluctuation between different load group.To relatively stable load group, shallow-layer is mostly single
The GRU network effect of member is preferable, and load group very strong for fluctuation, then should sufficiently be mentioned using the network of multiple-layer stacked
Take high-frequency characteristic.In order to adapt to the load character of different groups, the GRU net of three kinds of structures is proposed respectively to every type load group
Network merges three deep neural networks eventually by random forests algorithm sufficiently to learn different frequency domain load variations features
Output.The method of multi-model fusion ensure that prediction model has stronger applicability.
Three kinds of GRU network structures are as shown in Figure 3.By control network depth and GRU element number, targetedly learn
Low frequency, intermediate frequency, the high-frequency characteristic of Load aggregation body are merged eventually by random forest implementation model.The depth mind selected herein
It is as follows through network inputs feature:
(1) load data vector E, the E={ e at k moment of pastt-k,…,et-2,et-1, the k of the present embodiment setting is 6,
Load variations i.e. according to 3 hours in the past are predicted；
(2) belonging to point to be predicted coefficient I at the time of was divided into 48 future positions for 24 hours, then { 1,2 ..., 48 } I ∈；
(3) number of weeks D, D ∈ { 1,2 ..., 7 } belonging to point to be predicted；
(4) working day/festivals or holidays H, working day set 1, and festivals or holidays set 0；
(5) previous moment temperature T；
(6) current weather type W；13 kinds of weather categories such as the specified fine, light rain of London weather bureau, W ∈ 1,2 ...,
13}；
Since GRU network requirement input is between 0-1, vector E, T are handled by the normalized mode of minimax, by I,
D, H, W are converted into the form of heat coding.For categorical variable J, if J generic quantity is M, the variable J after heat coding contains
M bit, and only one bit for corresponding to generic is 1.
It will treated feature composition characteristic matrix X:
X={ E, I, D, H, T, W } (16)
All GRU networks are based on Keras framework establishment, are trained and survey on 1060 6G of GPU NVidia GTX
Examination, uses TensorFlow as computational back-end.In the training process of GRU network, realize that gradient decline is calculated by optimizer
Method, common optimizer have Adagrad, Adadelta, RMSprop, Adam.Herein using Adam as optimizer, advantage exists
In being able to achieve adaptive learning rate adjustment, training is efficient.For the random forests algorithm merged for multi-model, CART is set
Regression tree number is 50, and with no restrictions, the maximum characteristic that setting random forest uses is 3 to maximum decision tree depth.It is random gloomy
Woods model is trained after obtaining three GRU networks, and training data is verifying collection Lval。
After training obtains prediction model, to Load aggregation body to be predicted is given, i load is divided by cluster result
Group predicts each load group i using structure shown in Fig. 4, finally by i predicted value ei-tAdduction, can be obtained
The predicted value E of final t moment Load aggregation bodyt。
Three, Load aggregation body predicts error analysis
It is 2676 that Load aggregation body number of users M to be predicted, which is arranged, and Load aggregation body can be subdivided into 5 type load groups,
Every class includes 7392 groups of data sample, and wherein output power matrix Es dimension is 7392 × 1, and input feature vector matrix X-dimension is
7392 × 75, by data according to 8:1:1 ratio cut partition be training set, verifying collection and test set.Training set is for training GRU net
Network, verifying collection export different degree coefficient wk for training Random Forest model, and test set is for testing final mask performance.
In addition to MAPE, the mean absolute error (mean absolute error, MAE) of computation model predicts mould as supplement index
Type prediction effect in different load group is as shown in table 1.
Table 1
Seen from table 1, for different load groups, three kinds of network-critical degree difference are larger, while after Model Fusion
Precision of prediction is better than any single model, it was demonstrated that this paper institute climbing form type fusion method can make full use of heterogeneous networks structure special
Point realizes the automatic distribution of each network weight, to further increase precision of prediction.
For prove context of methods superiority, be respectively adopted BP neural network (three layers, neuronal quantity 128,256,
128), using the support vector machines of gaussian kernel function (support vector machine, SVM), random forest (random
Forest, RF) algorithm directly predicts Load aggregation body.Four kinds of method accuracy comparisons are as shown in table 2.
Table 2
As can be seen from Table 2, the algorithm in the embodiment of the present invention obtains mean absolute error percentage (MAPE) calculates
The mean absolute error (MAE) of model is low compared to BP neural network, SVM, random forests algorithm, illustrates in these four methods
In, the precision highest of the algorithm in the embodiment of the present invention.
Four, estimated performance compares under different scales Load aggregation body
Load aggregation body divides flexibly, and essentially different demarcation method finally only influences the quantity of syndication users.For
Algorithm of the present invention is verified to the applicability of different scales Load aggregation body, number of users is set by way of random sampling
M=500,1000,1500,2000,2676, compare the precision of prediction of the method and conventional method.Four kinds of method precision of predictions
It is as shown in Figure 5 with number of users situation of change.As seen from Figure 5, the mentioned method of this paper is due to using regrouping prediction, dynamic time
Modeling and Model Fusion technology, achieve highest precision of prediction under the conditions of different load scale.From different userbases
From the point of view of the MAE index of lower each method, the absolute error of prediction technique described herein is minimum, and increases error with number of users and become
Change less, performance is stablized, so that demonstrating proposed method herein all has preferable be applicable in the Load aggregation body of different scales
Property.And other three kinds of methods, especially SVM are done well when number of users is less, as number of users increases absolute error
It increases rapidly, estimated performance is unstable, and applicability is poor.
Five, rolling forecast effect
The predicted time scale of methods described herein can be adjusted flexibly by way of rolling forecast, Load aggregation is still set
The number of users of body is 2676, is respectively expanded to prediction scale for 24 hours by the following 30min, the precision of prediction MAPE of four kinds of methods
Variation is as shown in Figure 6.Curve is the result that multilinear fitting is carried out to scatterplot in figure.As seen from Figure 6, with predicted time scale
Increase, the precision of prediction of four kinds of methods decreases, and wherein random forests algorithm and the deterioration of SVM performance are significant.It is predicting
Scale is more than after 10 hours, and the error condition of two kinds of Artificial Neural Networks tends towards stability, and the accuracy benefits of this method obtain
It keeps, and predicts compared to ultra-short term more significant.
In conclusion the present invention proposes to be grouped prediction to Load aggregation body according to part throttle characteristics, and by distributed spectrum
Clustering algorithm is applied on the grouping and clustering of Load aggregation body, compares traditional K-means algorithm and improves clustering precision and stabilization
Property, overcomes that single machine spectral clustering calculating speed is slow, the disadvantage more than committed memory；Model Fusion thought is gathered applied to load
Fit prediction realizes the dynamic modeling of time series using the GRU deep neural network of different structure as meta-model,
Multiple meta-models are merged by random forests algorithm, heterogeneous networks design feature can be made full use of, further increased negative
Lotus precision of prediction；Compared to conventional methods such as BP, SVM, regrouping prediction+Model Fusion prediction technique of the invention is poly- to load
Fit forecasting problem has applicability, and all has higher precision of prediction under the conditions of different load scale；It is pre- by rolling
The time scale of prediction can be adjusted flexibly in the form of survey, on the prediction scale of 30min-24h, load forecasting method of the invention
All have higher precision of prediction.
Above embodiments describe the invention in detail, but content is only the preferred embodiment of the present invention, no
It can be believed to be used to limit the scope of the invention.Any changes and modifications in accordance with the scope of the present application,
It should still fall within the scope of the patent of the present invention.
Claims (5)
1. a kind of Load aggregation body regrouping prediction method based on gating cycle unit networks, it is characterised in that: the grouping is pre-
Survey method includes the following steps:
(1) adapter distribution spectral clustering is used, customer charge data are clustered, to obtain multiple part throttle characteristics
Similar electricity consumption group, and acquire each group's load characteristic matrix；
(2) the different GRU network of three kinds of structures, and the GRU that the temporal aspect by extracting group is different to three kinds of structures are built
Network is trained, and obtains the prediction model of three kinds of GRU networks, carries out model to three kinds of GRU networks by random forests algorithm
Fusion, obtains the load forecasting model of each group；
(3) moment feature to be predicted is input in load forecasting model obtained in step (2), respectively obtains each group
Predicted load sums different groups predicted value to obtain the predicted value of final Load aggregation body.
2. a kind of Load aggregation body regrouping prediction method based on gating cycle unit networks according to claim 1,
It is characterized in that: the acquisition process of electricity consumption group in the step (1) are as follows: mean value is taken by week to each customer charge data, is passed through
Maximum-Returning to one for minimum value zooms to section [0,1], obtains a load characteristic curve to each user, and all users are special
Linearity curve is integrated into matrix, and carrying out cluster for matrix can be obtained electricity consumption group.
3. a kind of Load aggregation body regrouping prediction method based on gating cycle unit networks according to claim 1,
Be characterized in that: GRU network is made of input layer, output layer and hidden layer in the step (2), and wherein hidden layer includes multiple grades
The GRU unit of connection.
4. a kind of Load aggregation body regrouping prediction method based on gating cycle unit networks according to claim 3,
Be characterized in that: the GRU unit includes a resetting door, and a update door passes through door control mechanism control output, recall info.
5. a kind of Load aggregation body regrouping prediction method based on gating cycle unit networks according to claim 1,
Be characterized in that: building the GRU network of three kinds of structures in the step (2) respectively to load group, by control network depth and
The polymeric low frequency of GRU element number Learning work load, intermediate frequency, high-frequency characteristic, and learn different frequency domain load variations features, finally
The output of three deep neural networks is merged by random forests algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810876297.6A CN109063911B (en) | 2018-08-03 | 2018-08-03 | Load aggregation grouping prediction method based on gated cycle unit network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810876297.6A CN109063911B (en) | 2018-08-03 | 2018-08-03 | Load aggregation grouping prediction method based on gated cycle unit network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063911A true CN109063911A (en) | 2018-12-21 |
| CN109063911B CN109063911B (en) | 2021-07-23 |
Family
ID=64832974
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810876297.6A Active CN109063911B (en) | 2018-08-03 | 2018-08-03 | Load aggregation grouping prediction method based on gated cycle unit network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063911B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886465A (en) * | 2019-01-20 | 2019-06-14 | 东北电力大学 | A kind of distribution network load prediction technique based on intelligent electric meter user's clustering |
| CN109948722A (en) * | 2019-03-27 | 2019-06-28 | 中国人民解放军战略支援部队航天工程大学 | A method of identification extraterrestrial target |
| CN109948861A (en) * | 2019-03-26 | 2019-06-28 | 西南交通大学 | A kind of urban rail based on mode decomposition and deep learning passenger flow forecasting in short-term |
| CN110110915A (en) * | 2019-04-28 | 2019-08-09 | 上海电力学院 | A kind of integrated prediction technique of the load based on CNN-SVR model |
| CN110610767A (en) * | 2019-08-01 | 2019-12-24 | 平安科技（深圳）有限公司 | Morbidity monitoring method, device, equipment and storage medium |
| CN110703899A (en) * | 2019-09-09 | 2020-01-17 | 创新奇智(南京)科技有限公司 | Data center energy efficiency optimization method based on transfer learning |
| CN110728403A (en) * | 2019-10-11 | 2020-01-24 | 国网江西省电力有限公司经济技术研究院 | Rural medium-long term power grid load prediction method |
| CN111062476A (en) * | 2019-12-06 | 2020-04-24 | 重庆大学 | Water quality prediction method based on gated circulation unit network integration |
| CN111222467A (en) * | 2020-01-08 | 2020-06-02 | 燕山大学 | Wind driven generator component fault classification detection method |
| CN111325315A (en) * | 2019-11-22 | 2020-06-23 | 国网新疆电力有限公司电力科学研究院 | Distribution transformer power failure and power loss prediction method based on deep learning |
| CN111563610A (en) * | 2020-03-30 | 2020-08-21 | 山东电力工程咨询院有限公司 | LSTM neural network-based building electrical load comprehensive prediction method and system |
| CN111784019A (en) * | 2019-12-26 | 2020-10-16 | 国网北京市电力公司 | Power load processing method and device |
| CN111797883A (en) * | 2019-09-30 | 2020-10-20 | 浙江浙能中煤舟山煤电有限责任公司 | Coal type identification method based on random forest |
| CN112308345A (en) * | 2020-11-30 | 2021-02-02 | 中国联合网络通信集团有限公司 | Communication network load prediction method, device and server |
| CN112766537A (en) * | 2020-12-24 | 2021-05-07 | 沈阳工程学院 | Short-term electric load prediction method |
| CN113222645A (en) * | 2021-04-23 | 2021-08-06 | 深圳市城市交通规划设计研究中心股份有限公司 | Urban hot spot area peak trip demand prediction method based on multi-source data fusion |
| CN113361776A (en) * | 2021-06-08 | 2021-09-07 | 国网上海市电力公司 | Power load probability prediction method based on user power consumption behavior clustering |
| CN113450141A (en) * | 2021-06-09 | 2021-09-28 | 重庆锦禹云能源科技有限公司 | Intelligent prediction method and device based on electricity selling quantity characteristics of large-power customer groups |
| CN113988436A (en) * | 2021-11-01 | 2022-01-28 | 广西电网有限责任公司 | Power consumption prediction method based on LSTM neural network and hierarchical relation correction |
| CN114254828A (en) * | 2021-12-21 | 2022-03-29 | 云南电网有限责任公司电力科学研究院 | Power load prediction method based on hybrid convolution feature extractor and GRU |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
| CN107330362A (en) * | 2017-05-25 | 2017-11-07 | 北京大学 | A kind of video classification methods based on space-time notice |
| CN107730044A (en) * | 2017-10-20 | 2018-02-23 | 燕山大学 | A kind of hybrid forecasting method of renewable energy power generation and load |
-
2018
- 2018-08-03 CN CN201810876297.6A patent/CN109063911B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
| CN107330362A (en) * | 2017-05-25 | 2017-11-07 | 北京大学 | A kind of video classification methods based on space-time notice |
| CN107730044A (en) * | 2017-10-20 | 2018-02-23 | 燕山大学 | A kind of hybrid forecasting method of renewable energy power generation and load |
Non-Patent Citations (2)
| Title |
|---|
| CHEN W Y: "Parallel spectral clustering in distributed systems", 《IEEE TRANSACTIONS PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| QUILUMBA F L: "Using Smart Meter Data to Improve the Accuracy of Intraday Load Forecasting Considering Customer Behavior Similarities", 《IEEE TRANSACTION ON SMART GRID》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886465A (en) * | 2019-01-20 | 2019-06-14 | 东北电力大学 | A kind of distribution network load prediction technique based on intelligent electric meter user's clustering |
| CN109886465B (en) * | 2019-01-20 | 2022-03-18 | 东北电力大学 | Power distribution network load prediction method based on intelligent electric meter user cluster analysis |
| CN109948861A (en) * | 2019-03-26 | 2019-06-28 | 西南交通大学 | A kind of urban rail based on mode decomposition and deep learning passenger flow forecasting in short-term |
| CN109948722A (en) * | 2019-03-27 | 2019-06-28 | 中国人民解放军战略支援部队航天工程大学 | A method of identification extraterrestrial target |
| CN110110915A (en) * | 2019-04-28 | 2019-08-09 | 上海电力学院 | A kind of integrated prediction technique of the load based on CNN-SVR model |
| JP2022536785A (en) * | 2019-08-01 | 2022-08-18 | 平安科技（深▲せん▼）有限公司 | Method, apparatus, equipment and storage medium for monitoring incidence rate |
| CN110610767A (en) * | 2019-08-01 | 2019-12-24 | 平安科技（深圳）有限公司 | Morbidity monitoring method, device, equipment and storage medium |
| JP7295278B2 (en) | 2019-08-01 | 2023-06-20 | 平安科技（深▲せん▼）有限公司 | Method, apparatus, equipment and storage medium for monitoring incidence rate |
| CN110703899A (en) * | 2019-09-09 | 2020-01-17 | 创新奇智(南京)科技有限公司 | Data center energy efficiency optimization method based on transfer learning |
| CN111797883B (en) * | 2019-09-30 | 2021-04-16 | 浙江浙能中煤舟山煤电有限责任公司 | Coal type identification method based on random forest |
| CN111797883A (en) * | 2019-09-30 | 2020-10-20 | 浙江浙能中煤舟山煤电有限责任公司 | Coal type identification method based on random forest |
| CN110728403A (en) * | 2019-10-11 | 2020-01-24 | 国网江西省电力有限公司经济技术研究院 | Rural medium-long term power grid load prediction method |
| CN110728403B (en) * | 2019-10-11 | 2023-04-18 | 国网江西省电力有限公司经济技术研究院 | Rural medium-long term power grid load prediction method |
| CN111325315A (en) * | 2019-11-22 | 2020-06-23 | 国网新疆电力有限公司电力科学研究院 | Distribution transformer power failure and power loss prediction method based on deep learning |
| CN111062476A (en) * | 2019-12-06 | 2020-04-24 | 重庆大学 | Water quality prediction method based on gated circulation unit network integration |
| CN111784019A (en) * | 2019-12-26 | 2020-10-16 | 国网北京市电力公司 | Power load processing method and device |
| CN111222467B (en) * | 2020-01-08 | 2022-03-25 | 燕山大学 | Wind driven generator component fault classification detection method |
| CN111222467A (en) * | 2020-01-08 | 2020-06-02 | 燕山大学 | Wind driven generator component fault classification detection method |
| CN111563610A (en) * | 2020-03-30 | 2020-08-21 | 山东电力工程咨询院有限公司 | LSTM neural network-based building electrical load comprehensive prediction method and system |
| CN111563610B (en) * | 2020-03-30 | 2023-07-18 | 山东电力工程咨询院有限公司 | Building electric load comprehensive prediction method and system based on LSTM neural network |
| CN112308345A (en) * | 2020-11-30 | 2021-02-02 | 中国联合网络通信集团有限公司 | Communication network load prediction method, device and server |
| CN112766537A (en) * | 2020-12-24 | 2021-05-07 | 沈阳工程学院 | Short-term electric load prediction method |
| CN112766537B (en) * | 2020-12-24 | 2023-06-06 | 沈阳工程学院 | Short-term electric load prediction method |
| CN113222645A (en) * | 2021-04-23 | 2021-08-06 | 深圳市城市交通规划设计研究中心股份有限公司 | Urban hot spot area peak trip demand prediction method based on multi-source data fusion |
| CN113361776A (en) * | 2021-06-08 | 2021-09-07 | 国网上海市电力公司 | Power load probability prediction method based on user power consumption behavior clustering |
| CN113450141A (en) * | 2021-06-09 | 2021-09-28 | 重庆锦禹云能源科技有限公司 | Intelligent prediction method and device based on electricity selling quantity characteristics of large-power customer groups |
| CN113450141B (en) * | 2021-06-09 | 2023-09-01 | 重庆锦禹云能源科技有限公司 | Intelligent prediction method and device based on electricity sales quantity characteristics of large power customer group |
| CN113988436A (en) * | 2021-11-01 | 2022-01-28 | 广西电网有限责任公司 | Power consumption prediction method based on LSTM neural network and hierarchical relation correction |
| CN113988436B (en) * | 2021-11-01 | 2023-04-28 | 广西电网有限责任公司 | Power consumption prediction method based on LSTM neural network and hierarchical relation correction |
| CN114254828A (en) * | 2021-12-21 | 2022-03-29 | 云南电网有限责任公司电力科学研究院 | Power load prediction method based on hybrid convolution feature extractor and GRU |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063911B (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109063911A (en) | A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks | |
| CN109754113B (en) | Load prediction method based on dynamic time warping and long-and-short time memory | |
| Yunpeng et al. | Multi-step ahead time series forecasting for different data patterns based on LSTM recurrent neural network | |
| Xuan et al. | Multi-model fusion short-term load forecasting based on random forest feature selection and hybrid neural network | |
| Dong et al. | Short-term load forecasting in smart grid: A combined CNN and K-means clustering approach | |
| CN111860951A (en) | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network | |
| Xiao et al. | Ensemble ANNs-PSO-GA approach for day-ahead stock e-exchange prices forecasting | |
| CN101383023B (en) | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation | |
| CN109359786A (en) | A kind of power station area short-term load forecasting method | |
| CN106709820A (en) | Power system load prediction method and device based on deep belief network | |
| CN106600059A (en) | Intelligent power grid short-term load predication method based on improved RBF neural network | |
| CN106779087A (en) | A kind of general-purpose machinery learning data analysis platform | |
| CN108549960A (en) | A kind of 24 hours Methods of electric load forecasting | |
| CN110414788A (en) | A kind of power quality prediction technique based on similar day and improvement LSTM | |
| CN112396152A (en) | Flood forecasting method based on CS-LSTM | |
| CN109255726A (en) | A kind of ultra-short term wind power prediction method of Hybrid Intelligent Technology | |
| CN110322075A (en) | A kind of scenic spot passenger flow forecast method and system based on hybrid optimization RBF neural | |
| CN109214503A (en) | Project of transmitting and converting electricity cost forecasting method based on KPCA-LA-RBM | |
| CN116128150B (en) | Photovoltaic intelligent output prediction method based on two-stage optimization | |
| Akpinar et al. | Forecasting natural gas consumption with hybrid neural networks—Artificial bee colony | |
| CN113361785A (en) | Power distribution network short-term load prediction method and device, terminal and storage medium | |
| CN115186803A (en) | Data center computing power load demand combination prediction method and system considering PUE | |
| CN111882114B (en) | Short-time traffic flow prediction model construction method and prediction method | |
| CN114118596A (en) | Photovoltaic power generation capacity prediction method and device | |
| Seera et al. | Improving the Fuzzy Min–Max neural network performance with an ensemble of clustering trees |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |