CN110232483A - Deep learning load forecasting method, device and terminal device - Google Patents
Deep learning load forecasting method, device and terminal device Download PDFInfo
- Publication number
- CN110232483A CN110232483A CN201910527965.9A CN201910527965A CN110232483A CN 110232483 A CN110232483 A CN 110232483A CN 201910527965 A CN201910527965 A CN 201910527965A CN 110232483 A CN110232483 A CN 110232483A
- Authority
- CN
- China
- Prior art keywords
- sub
- time window
- traveling time
- load
- default
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 70
- 238000013277 forecasting method Methods 0.000 title claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 123
- 238000000034 method Methods 0.000 claims description 43
- 238000004422 calculation algorithm Methods 0.000 claims description 40
- 238000004590 computer program Methods 0.000 claims description 20
- 230000015654 memory Effects 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 238000005259 measurement Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 10
- 230000002159 abnormal effect Effects 0.000 claims description 9
- 240000002853 Nelumbo nucifera Species 0.000 claims description 6
- 235000006508 Nelumbo nucifera Nutrition 0.000 claims description 6
- 235000006510 Nelumbo pentapetala Nutrition 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 4
- 230000006870 function Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 8
- 238000013461 design Methods 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000005611 electricity Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 description 1
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001764 infiltration Methods 0.000 description 1
- 230000008595 infiltration Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000013517 stratification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention is suitable for technical field of data prediction, provides a kind of deep learning load forecasting method, device and terminal device.The deep learning load forecasting method includes: that forecast interval is divided into the multiple first sub- forecast intervals using the first default traveling time window, and historical load data is divided into the multiple first sub- weight training data, wherein the length of the historical load data is corresponding with the length of the forecast interval;Using the deep learning prediction model of the multiple first the multiple first sub- forecast interval of sub- weight training data training, and obtain the predicted load of every sub- forecast interval;Final predicted value is determined according to the predicted value of each sub- forecast interval.Above-mentioned deep learning load forecasting method is by adjusting the fine granularity and selection historical load data appropriate of cutting forecast interval, and suitable for the load prediction of different durations, and the predicted value obtained is more accurate reliable.
Description
Technical field
The invention belongs to data prediction fields more particularly to a kind of deep learning load forecasting method, device and terminal to set
It is standby.
Background technique
Load prediction is according to historical load changing rule, in conjunction with factors such as weather, temperature, economy, politics to following several
The load of hour, several days or several years several months carry out the prediction of science.
Currently, traditional load forecasting method requires greatly according to different predicted times using different load prediction sides
Method, the load forecasting model complexity constructed in this way is high, and versatility is low.And with new energy (such as wind-power electricity generation and photovoltaic power generation
Deng) infiltration and sustainable growth and flexible controllable burden use electrotransfer, the randomness of power load and uncertain constantly increase
Add, just to load prediction accurate, reliable more stringent requirements are proposed for this.
Summary of the invention
In view of this, the embodiment of the invention provides a kind of deep learning load forecasting method, device and terminal device, with
The load forecasting model complexity for solving to construct in the prior art is high, versatility is low and load prediction is inaccurate reliably asks
Topic.
The first aspect of the embodiment of the present invention provides a kind of deep learning load forecasting method, comprising:
Forecast interval is divided into the multiple first sub- forecast intervals using the first default traveling time window, and history is born
Lotus data are divided into the multiple first sub- weight training data, wherein the length of the historical load data and the forecast interval
Length is corresponding;
Deep learning using the multiple first the multiple first sub- forecast interval of sub- weight training data training is pre-
Model is surveyed, and obtains the predicted load of every sub- forecast interval;
Final predicted value is determined according to the predicted value of each sub- forecast interval.
Preferably, the described first default traveling time window is obtained by clustering algorithm, process are as follows:
It randomly obtains one group second and presets traveling time window, using one group of second preset time window by the history
Load data is divided into the sub- weight training data of multiple groups second, wherein the sub- weight training data of the multiple groups second are calculated as cluster
The input of method;
The smallest clusters number is found according to the measurement standard of clustering algorithm;
The minimum value of clusters number is selected, and using the traveling time window of maximum length corresponding to the minimum value as described in
First default traveling time window.
Preferably, the measurement standard using mean profile coefficient and interval stats amount as the clustering algorithm.
Preferably, described one group second setting method for presetting traveling time window are as follows:
The initial length of each second default traveling time window is identical, will be described using the described second default traveling time window
Historical load data is divided into one group of second sub- weight training data, and in the upper length to the described second default traveling time window
After degree increases preset step-length Δ T, according to the current length of the second default traveling time window after increase to the historical load number
According to being divided, until obtaining the sub- weight training data of multiple groups second;Or
The initial length of each second default traveling time window is different, according to the setting of the change rate of historical load data
The length of second default traveling time window, and increase preset step-length Δ in the upper length to the described second default traveling time window
After T, the historical load data is divided according to the current length of the second default traveling time window after increase, until
To the sub- weight training data of multiple groups second.
Preferably, the move mode of the described first default traveling time window are as follows:
Current first default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t=
Twin,i, then the end position of the initial position of the next first default traveling time window and the current first default traveling time window is held in the mouth
It connects, wherein Twin,iFor the length of the current first default traveling time window;Or
Current first default traveling time window when performing the next step mobile, if default moving step length Δ t meet Δ t <
Twin,i, then the initial position of the next first default traveling time window and the current first default traveling time window be there are Chong Die, wherein
Twin,iFor the length of the current first default traveling time window;
The move mode of the second default traveling time window are as follows:
Current second default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t=
Twin,i, then the end position of the initial position of the next second default traveling time window and the current second default traveling time window is held in the mouth
It connects, wherein Twin,iFor the length of the current second default traveling time window;Or
Current second default traveling time window when performing the next step mobile, if default moving step length Δ t meet Δ t <
Twin,i, then the initial position of the next second default traveling time window and the current second default traveling time window Twin,iThere are overlapping,
Wherein, Twin,iFor the length of the current second default traveling time window.
Preferably, in the depth using the multiple first the multiple first sub- forecast interval of sub- weight training data training
Degree study prediction model, and before obtaining the predicted load of every sub- forecast interval, the deep learning load forecasting method is also
Include:
The multiple first sub- weight training data are pre-processed, the multiple first sub- weight training data are corrected
In abnormal data or fill up missing data in the multiple first sub- weight training data;
To pretreated multiple first sub- weight training data normalization processing;
The depth using the multiple first the multiple first sub- forecast interval of sub- weight training data training
Prediction model is practised, and obtains the predicted load of every sub- forecast interval specifically:
The multiple first sub- weight training data are pre-processed, the multiple first sub- weight training data are corrected
In abnormal data or fill up missing data in the multiple first sub- weight training data;
To pretreated multiple first sub- weight training data normalization processing;
Utilize the multiple first the multiple first sub- forecast intervals of sub- weight training data training after normalized
Deep learning prediction model, and obtain the predicted load of every sub- forecast interval.
Preferably, the deep learning prediction model is to be predicted using the depth LSTM of Keras deep learning framework establishment
Model, the training process of the depth LSTM prediction model are as follows:
Obtain the first sub- weight training data of i-th of sub- forecast interval;
Utilize grid search LSTM hyper parameter;
Depth LSTM prediction model is established using the described first sub- weight training data and the LSTM hyper parameter.
Preferably, the algorithm of the deep learning prediction model is that depth encodes neural network certainly.
The second aspect of the embodiment of the present invention provides a kind of deep learning load prediction device, comprising:
Cutting module for forecast interval to be divided into the multiple first sub- forecast intervals, and historical load data is drawn
It is divided into the multiple first sub- weight training data;
Prediction module, for utilizing the multiple first the multiple first sub- forecast interval of sub- weight training data training
Deep learning prediction model, and obtain the predicted load of every sub- forecast interval;
Determining module, for determining final predicted value according to the predicted load of each sub- forecast interval.
The third aspect of the embodiment of the present invention provides a kind of terminal device, including memory, processor and is stored in
In the memory and the computer program that can run on the processor, when the processor executes the computer program
The step of realizing the as above any one deep learning load forecasting method.
The fourth aspect of the embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Media storage has computer program, realizes that as above any one deep learning is negative when the computer program is executed by processor
The step of lotus prediction technique.
The embodiment of the present invention proposes a kind of deep learning load forecasting method, by presetting traveling time window for forecast interval
Multiple sub- forecast intervals are divided into, historical load data is divided into multiple sub- weight training data;It is instructed by multiple sub- loads
Practice the corresponding multiple sub- forecast intervals of data training, by adjusting the first default traveling time window length to adjust the pre- of cutting
Survey the fine granularity and selection historical load data appropriate in section, the load prediction suitable for different durations.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description is only of the invention some
Embodiment for those of ordinary skill in the art without creative efforts, can also be attached according to these
Figure obtains other attached drawings.
Fig. 1 is a kind of flow diagram of deep learning load forecasting method provided by the invention;
Fig. 2 is a kind of specific example figure of deep learning load forecasting method provided by the invention;
Fig. 3 is time window cutting schematic diagram provided in an embodiment of the present invention;
Fig. 4 is load prediction comparison diagram provided in an embodiment of the present invention;
Fig. 5 is the schematic diagram of deep learning load prediction device provided in an embodiment of the present invention;
Fig. 6 is the schematic diagram of terminal device provided in an embodiment of the present invention.
Specific embodiment
In being described below, for illustration and not for limitation, the tool of such as particular system structure, technology etc is proposed
Body details, to understand thoroughly the embodiment of the present invention.However, it will be clear to one skilled in the art that there is no these specific
The present invention also may be implemented in the other embodiments of details.In other situations, it omits to well-known system, device, electricity
The detailed description of road and method, in case unnecessary details interferes description of the invention.
In order to illustrate technical solutions according to the invention, the following is a description of specific embodiments.
Fig. 1 shows a kind of flow diagram of depth load forecasting method provided by the invention, referring to Fig. 1, to this hair
Details are as follows for the deep learning load forecasting method of bright offer.
Forecast interval is divided into the multiple first sub- forecast intervals using the first default traveling time window by step S101, with
And historical load data is divided into the multiple first sub- weight training data, wherein the length of the historical load data with it is described
The length of forecast interval is corresponding.
Traditional load forecasting method is usually gathered historical load data using different for short-term load forecasting
Class algorithm is to find out Overload Class similar with day is predicted, according to Overload Class training front stage BP cascade similar with prediction day
Neural network;For Mid-long term load forecasting, generally by functional data analysis theory and nonparametric probability side
Then the Load Forecast Algorithm of method constructor type non parametric regression corrects the prediction of nonparametric Regression Model by quadratic programming
Curve finally obtains the prediction curve of specified forecast interval;Therefore, traditional load forecasting method, usually will be for difference
The different load forecasting method of the load prediction block design of duration, these methods have the negative of lower versatility or building
Lotus prediction model is sufficiently complex.
Specifically, first with the first default traveling time window by forecast interval and historical load number in the embodiment of the present invention
According to being divided, long load prediction section, is conducive to when so as to be shorter by the load prediction interval division of longer duration
Simplify the unification of load forecasting model and a variety of load forecasting methods.
Wherein, what historical load data represented is the load data in the past period, the length of historical load data
It is corresponding with the length of forecast interval, the historical load data of corresponding length is selected according to the duration of specific forecast interval;Prediction
The duration in section can be divided into ultra-short term, short-term, medium and long term according to the purpose of load prediction, ultra-short term it is pre-
Section general control is surveyed within one hour following, short-term load forecasting refers to that daily load prediction or all load predictions, mid-term are negative
Lotus predicts the load prediction for referring to the moon to year, and long term load forecasting refers to the load in The Next 3-5 Years even a longer period of time
Prediction, the duration of forecast interval are independently set according to actual needs, and the embodiment of the present invention is it is not limited here.
Preferably, the first default traveling time window can select to obtain by clustering algorithm, and it is pre- to obtain first by clustering algorithm
If the process of traveling time window are as follows: randomly obtain one group second and preset traveling time window, when being preset using described one group second
Between window the historical load data is divided into the sub- weight training data of multiple groups second, wherein the sub- weight training of the multiple groups second
Input of the data as clustering algorithm;The smallest clusters number is found according to the measurement standard of clustering algorithm;Select clusters number
Minimum value, and using the traveling time window of maximum length corresponding to the minimum value as the described first default traveling time window.
It specifically, clustering algorithm is a kind of statistical analysis technique for studying classification problem, while being also the one of data mining
A important algorithm can deeply excavate the inner link between historical load data using clustering algorithm, so that it is pre- to improve load
The accuracy of survey.
Wherein, clustering algorithm can be divided into partitioning (Partitioning Methods), stratification (Hierarchical
Methods), the method based on density (Density-Based Methods), the method (Grid-Based based on grid
Methods) and the method based on model (Model-Based Methods) etc., specific clustering algorithm can be selected independently, can
It is to find the smallest clusters number, the smallest clusters number is corresponding different measurement standards to be arranged according to actual needs
The traveling time window of maximum length can reduce the complexity of constructed load forecasting model as the first default traveling time window
Degree.
Preferably, the measurement standard using mean profile coefficient and interval stats amount as the clustering algorithm.
Wherein, silhouette coefficient is a kind of evaluation method of Clustering Effect quality, it combine two kinds of cohesion degree and separating degree because
Element evaluates different clustering algorithms or identical clustering algorithm different running method to influence caused by cluster result;Specific to this
In inventive embodiments, the silhouette coefficient of i-th of sub- weight training data sample are as follows:
In formula:
For i-th of sub- weight training data sample, the average distance of its every other sample into its affiliated class is calculated,
Remember ai;
For i-th of sub- weight training data sample, calculating it, all samples in nearest inhomogeneity are averaged with it
Distance remembers bi;Wherein " distance " refers to dissimilar degree, and distance is bigger, and it is higher to represent dissmilarity degree, and Euclidean distance
Just meet this condition, therefore, calculate the distance between above-mentioned sample using Euclidean distance, it may be assumed that
In formula: d (x, y) indicates the Euclidean distance between two data samples x and y;The length of two samples is n.
For the value of silhouette coefficient in the section of [- 1,1], the bigger expression cohesion degree of value and separating degree are all more excellent;
The silhouette coefficient of all sub- weight training data samples is averaging, is exactly the silhouette coefficient of the cluster result.
Wherein, interval stats amount (Gap Statistic) determines the number of determined class to solve clustering problem;It will
Variation is compared with its desired value under reference load data distribution in total cluster of different cluster numbers k values.This reference number
It is generated according to using monte carlo method, i.e., for each sample, calculates its maximum value and minimum value, it is then random uniformly to generate
Random number of the minimum value to maximum value.For real data and reference data, calculated in total cluster using different cluster k values
Variation.Interval stats amount calculates as follows in the case where given cluster numbers k value:
In formula:Indicate that reference length is the expectation of the sample of n;CrIndicate r-th of cluster classification, nr=| Cr|, DrIt indicates
Euclidean distance in class between sample point.
Steps are as follows for the calculating of this evaluation index:
1) actual multiple sub- weight training data sets are clustered, changes clusters number k=1,2 ..., kamx, and count
Calculate corresponding Wk;
2) it generates reference data set and it is clustered, change clusters number k=1,2 ..., kamx, calculate corresponding Gn(k);
3) it enablesB is the number for generating reference data set.Then standard deviation is calculatedAnd it sets
The smallest clusters number k is finally selected, G is metn(k)≥Gn(k+1)-sk+1.K is optimal cluster numbers at this time
Mesh selects the traveling time window of the corresponding maximum length of this clusters number as the described first default traveling time window.
When two kinds of measurement standards selection min cluster number difference when, preferentially select cluster data relatively small as
Optimal clusters number, to determine the length of the first default traveling time window.
Step S102 utilizes the depth of the multiple first the multiple first sub- forecast interval of sub- weight training data training
Degree study prediction model, and obtain the predicted load of every sub- forecast interval.
Specifically, utilizing the depth of the multiple first the multiple first sub- forecast interval of sub- weight training data training
Learn prediction model, the predicted load of the every sub- forecast interval obtained can be made more accurate.
Step S103 determines final predicted value according to the predicted value of each sub- forecast interval.
Specifically, the predicted value of each sub- forecast interval is integrated sequentially in time, and to the prediction after integration
Value carries out renormalization, obtains final predicted value.
The present invention provides a kind of deep learning Load Forecast Algorithms, will predict including the use of the first default traveling time window
Interval division is the multiple first sub- forecast intervals, and historical load data is divided into the multiple first sub- weight training data,
Wherein the length of the historical load data is corresponding with the length of the forecast interval;Utilize the multiple first sub- weight training
The deep learning prediction model of the multiple first sub- forecast interval of data training, and the load for obtaining every sub- forecast interval is pre-
Measured value;Final predicted value is determined according to the predicted value of each sub- forecast interval.
As it can be seen that in the present invention, can using the first default traveling time window by forecast interval and historical load data cutting,
It, can be by length with the deep learning prediction model of obtained after cutting first the first sub- forecast interval of sub- weight training data training
Spending longer forecast interval cutting is the shorter forecast interval of length, in order to real using a kind of deep learning load forecasting method
The load prediction of existing different time length forecast interval, and by adjusting the fine granularity of the sub- forecast interval after cutting, can also be with
Increase the accuracy of prediction result.
On the basis of the above embodiments:
Embodiment as one preferred, described one group second setting method for presetting traveling time window are as follows:
The initial length of each second default traveling time window is identical, will be described using the described second default traveling time window
Historical load data is divided into one group of second sub- weight training data and in the upper length to the described second default traveling time window
After degree increases preset step-length Δ T, according to the current length of the second default traveling time window after increase to the historical load number
According to being divided, until obtaining the sub- weight training data of multiple groups second;Or
The initial length of each second default traveling time window is different, according to the setting of the change rate of historical load data
The length of second default traveling time window, and increase preset step-length Δ in the upper length to the described second default traveling time window
After T, the historical load data is divided according to the current length of the second default traveling time window after increase, until
To the sub- weight training data of multiple groups second.
Specifically, the historical load data divided respectively to each second default traveling time window clusters, reference
Multiple cluster results, can choose a kind of length of more preferably the first preset time window, and then optimize trained deep learning
Prediction model obtains more accurate prediction result.
Wherein, it is pre- not to be further added by second then for each second default traveling time window length after adding preset step-length Δ T sufficiently large
If the length of traveling time window, determines according to actual conditions, the embodiment of the present invention is without limitation by preset step-length Δ T.
Embodiment as one preferred, the move mode of the first default traveling time window are as follows:
Current first default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t=
Twin,i, then the end position of the initial position of the next first default traveling time window and the current first default traveling time window is held in the mouth
It connects, wherein Twin,iFor the length of the current first default traveling time window;Or
Current first default traveling time window when performing the next step mobile, if default moving step length Δ t meet Δ t <
Twin,i, then the initial position of the next first default traveling time window and the current first default traveling time window be there are Chong Die, wherein
Twin,iFor the length of the current first default traveling time window;
The move mode of the second default traveling time window are as follows:
Current second default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t=
Twin,i, then the end position of the initial position of the next second default traveling time window and the current second default traveling time window is held in the mouth
It connects, wherein Twin,iFor the length of the current second default traveling time window;Or
Current second default traveling time window when performing the next step mobile, if default moving step length Δ t meet Δ t <
Twin,i, then the initial position of the next second default traveling time window and the current second default traveling time window Twin,iThere are overlapping,
Wherein, Twin,iFor the length of the current second default traveling time window.
Specifically, moving step length Δ t meets Δ t≤Twin,iWith guarantee historical load data can by complete cutting without
It omits, when moving step length Δ t meets Δ t < Twin,iWhen, forecast interval and historical load data can obtain more fine-grained stroke
Point, in order to which this deep learning load forecasting method is suitable for the load prediction in different duration prediction sections.
Embodiment as one preferred is utilizing the multiple first sub- weight training data training the multiple first
The deep learning prediction model of sub- forecast interval, and before obtaining the predicted load of every sub- forecast interval, the deep learning
Load forecasting method further include:
The multiple first sub- weight training data are pre-processed, the multiple first sub- weight training data are corrected
In abnormal data or fill up missing data in the multiple first sub- weight training data;
To pretreated multiple first sub- weight training data normalization processing;
The depth using the multiple first the multiple first sub- forecast interval of sub- weight training data training
Prediction model is practised, and obtains the predicted load of every sub- forecast interval specifically:
The multiple first sub- weight training data are pre-processed, the multiple first sub- weight training data are corrected
In abnormal data or fill up missing data in the multiple first sub- weight training data;
To pretreated multiple first sub- weight training data normalization processing;
Utilize the multiple first the multiple first sub- forecast intervals of sub- weight training data training after normalized
Deep learning prediction model, and obtain the predicted load of every sub- forecast interval.
Specifically, the historical load data of equal length corresponding with forecast interval is drawn using the first default traveling time window
After being divided into multiple sub- weight training data, the problems such as, ageing equipment improper since in sampling process, there are manual operations, cause
There may be missing values and abnormal data for historical load data, i.e. there may be missing values and exception for every sub- weight training data
Data, therefore, it is necessary to first be pre-processed to the corresponding sub- weight training data of every sub- forecast interval, correct abnormal data or
Fill up missing data;
Wherein, since deep learning prediction algorithm is more sensitive to data scale ratio, we are to pretreated data
It is normalized again:
In formula: the data matrix after X ' expression normalization;tkIndicate the length of k-th of Sub Data Set;xiIndicate the i-th of X '
Row vector;minxiAnd maxxiIndicate xiMinimum value and maximum value.
Specifically, making data processing get up more just within the scope of pretreated data are namely mapped to 0~1
It is prompt quick.
Certainly, other than the sub- weight training data after division are normalized, other forms can also be carried out
Processing, the embodiment of the present invention is it is not limited here.
Embodiment as one preferred, deep learning prediction model are the depth using Keras deep learning framework establishment
Spend LSTM prediction model, the training process of depth LSTM prediction model are as follows:
Obtain the first sub- weight training data of i-th of sub- forecast interval;
Utilize grid search LSTM hyper parameter;
Depth LSTM prediction model is established using the described first sub- weight training data and the LSTM hyper parameter.
Specifically, Keras is a high level neural network API (Application Program Interface), by pure
Python (a kind of computer programming language) writes, its purpose be in order to support quick experiment, can be yours
Idea is rapidly converted into as a result, having the advantages that user friendly, modularization and easily extension.
Wherein, LSTM is depth shot and long term memory network (Long Short Term Memory), it is a kind of improved
RNN (Recurrent Neural Network, Recognition with Recurrent Neural Network) model, relative to the RNN of standard, LSTM is more suitable for place
Relatively long critical event is spaced and postponed in reason and predicted time sequence.
Embodiment as one preferred, the algorithm of deep learning prediction model are that depth encodes neural network certainly.
Certainly, in addition to LSTM and depth are from neural network is encoded, the algorithm of deep learning prediction model can also use it
The neural network of his type replaces, and the embodiment of the present invention is it is not limited here.
Referring to FIG. 2, Fig. 2 is a kind of specific example figure of deep learning load forecasting method provided by the invention, specifically
Exemplary step is as follows:
Selection input load data:
Siding-to-siding block length T including determining load prediction, for different siding-to-siding block lengths, (short-term or Mid-long Term Load is pre-
Survey) and the desired load resolution ratio (interval between i.e. adjacent load data point) of load prediction, select different history negative
Lotus data.Such as following one day load data of requirement forecast, it can choose the historical load data of identical number of weeks.
Traveling time window cutting data set:
If the length of forecast interval is T, the length of i-th of second default traveling time windows is Twin,i.Pass through design second
It predicts traveling time window and moves the second default traveling time window since the starting endpoint of historical load data section T to history
The end caps of load data section T, by going through in each second default traveling time window institute scope in moving process
The cutting of history load data and individually storage record are as the second sub- weight training data.It is cut at this time with the second default traveling time window
Divided data collection is as a preliminary cutting, and the second default traveling time window will train number with the history of forecast interval equal length
According to the multiple second sub- weight training data are divided into, the multiple second sub- weight training data after cutting are as the defeated of clustering algorithm
Enter, to find the first default traveling time window using clustering algorithm.
Clustering algorithm determines optimal cutting:
It should not only be arranged too short of the length of time window but also should not be arranged too long, determine suitable by clustering algorithm
The length of one default traveling time window.
Firstly, being calculated using the multiple second sub- weight training data after the second default traveling time window cutting as cluster
The input of method finds optimal number of clusters as measurement standard using mean profile coefficient and interval stats amount;
Secondly, maximum traveling time window length corresponding to the minimum value and this minimum value of selection clusters number.
Finally the traveling time window of the maximum length according to corresponding to the minimum value of clusters number is default as described first
Traveling time window.
When two kinds of measurement standards selection min cluster number difference when, preferentially select clusters number relatively small as
Optimal clusters number, with the default traveling time window length of determination suitable first.
Optimal cutting data normalization:
The historical load data of equal length corresponding with forecast interval is divided into using suitable first movement time window
After multiple first sub- weight training data, need first to carry out the corresponding first sub- weight training data of every sub- forecast interval pre-
Processing corrects abnormal data or fills up missing data;
Since deep learning prediction algorithm is more sensitive to data scale ratio, we carry out pretreated data again
Normalized, the data after normalized easily facilitate data processing between 0~1.
After normalization data, other characteristics can be increased on the basis of each first sub- weight training data, with
The precision of prediction is improved, according to sample frequency, other characteristics, each first sub- weight training data can determine depth
The input quantity and output quantity of learning network;
The modeling and prediction of sub- forecast interval:
After every sub- weight training data prediction normalization, prediction model is established for corresponding sub- forecast interval, is had
Body:
The initial data of i-th of sub- forecast interval is determined first;
Secondly grid search LSTM hyper parameter is utilized;
Then prediction model is established according to LSTM hyper parameter;
The predicted load in i-th of subinterval is finally obtained according to prediction model.
Wherein, Keras deep learning framework establishment depth LSTM model is utilized.When due to LSTM model foundation it needs to be determined that
Multiple hyper parameters in model, such as number, loss function, the model the number of iterations of hidden state etc..It is more using grid search
The optimal value of a hyper parameter, and then establish depth LSTM prediction model.
Integrate the predicted value of each sub- forecast interval, renormalization, output prediction load data.
Specifically, assuming that sub- forecast interval there are k, need to judge whether the serial number i of i-th of sub- forecast interval is greater than k, when
When i > k, the predicted value of each sub- forecast interval is just integrated, to ensure that the predicted value of every sub- forecast interval is all counted on.
As it can be seen that in this deep learning load prediction specific example, according to the length of different forecast intervals, thus it is possible to vary cutting
The fine granularity in section, such as forecast interval are long, total forecast interval fine granularity highland can be divided into multiple careful sons
Forecast interval;Thus it is suitable for the load prediction of different durations, the prediction algorithm of different prediction lengths can be unified, reduces load
The complexity of prediction model building, enhances its versatility;When since the suitable first default movement has been determined using clustering algorithm
Between window, it is right and before the deep learning prediction model using first sub- weight training data the first sub- forecast interval of training
First sub- weight training data are pre-processed and have been normalized, and then improve prediction progress and data processing speed.
Embodiment as one preferred verifies above-mentioned deep learning load prediction for predicting following 1 day load
Method, wherein historical load data resolution ratio is 15 minutes.Assuming that forecast interval length is T, actual load data are xt, t=
1,2,…,T;Remember that the load data through deep learning model prediction isT=1,2 ..., T;Predict one day data, wherein T
Value is 96.The precision of prediction is measured using following 3 kinds of evaluation indexes:
(1) root-mean-square error
(2) mean absolute error
(3) precision of prediction
Successively according to step described previously, optimal time window length is found out using clustering algorithm first.It is optimal in this example
Clusters number be 4, in addition, the division of sub- forecast interval is as shown in Fig. 3.Then it is utilized respectively using grid data service pre-
Survey the hyper parameter of the corresponding sub- weight training data training deep neural network in section.Instruction is finally passed through based on historical load data
LSTM model prediction load data after the completion of white silk, furthermore this algorithm is compared with traditional BP neural network, the two it is pre-
It is as shown in Fig. 4 to survey result figure.
The error of the algorithm of the invention design is respectively as follows: error1=9271.51, error2=75.34;Traditional BP mind
Prediction error through network are as follows: error1=26192.09, error2=128.77;By comparing this algorithm compared with traditional neural net
The error of network reduces respectively are as follows: Δ error1=16920.58, Δ error2=53.43.Furthermore the precision of prediction of LSTM algorithm is
η=2.19%, and BP neural network precision of prediction is η=5.15%, precision of prediction improves 2.96%.
As can be seen that using the depth LSTM load forecasting method based on traveling time window, the predicted load of acquisition is more
To be accurate reliable, this deep learning load forecasting method is applied to electric system, can be mentioned for the economic load dispatching of electric system
For solid reference, reliable basis is provided to formulate load dispatching policy and setting area tou power price.
In addition to this, this deep learning load forecasting method can also be applied to the prediction of resident region consumption gas quantity
Deng the embodiment of the present invention is not construed as limiting this.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
Fig. 5 is a kind of structural schematic diagram for deep learning load prediction device that one embodiment of the invention provides, referring to figure
5, which may include cutting module 50, prediction module 51 and output module 52.
Cutting module 50 is used to for forecast interval being divided into the multiple first sub- forecast intervals, and historical load data is drawn
It is divided into the multiple first sub- weight training data;Prediction module 51 is used to utilize the multiple first sub- weight training data training institute
The deep learning prediction model of the multiple first sub- forecast intervals is stated, and obtains the predicted load of every sub- forecast interval;Output
Module 52 is used to determine final predicted value according to the predicted load of each sub- forecast interval.
Aforementioned depth Learning work load is please referred to for the introduction of the deep learning load prediction device in the embodiment of the present invention
Prediction technique embodiment, details are not described herein for the embodiment of the present invention.
Fig. 6 is the schematic diagram for the terminal device that one embodiment of the invention provides.As shown in fig. 6, the terminal of the embodiment is set
Standby 6 include: processor 60, memory 61 and are stored in the meter that can be run in the memory 61 and on the processor 60
Calculation machine program 62.The processor 60 realizes above-mentioned each deep learning load forecasting method when executing the computer program 62
Step in embodiment, such as step shown in FIG. 1.Alternatively, realization when the processor 60 executes the computer program 62
The function of each module/unit in above-mentioned each Installation practice, such as the function of module 50 to 52 shown in Fig. 5.
Illustratively, the computer program 62 can be divided into one or more module/units, it is one or
Multiple module/units are stored in the memory 61, and are executed by the processor 60, to complete the present invention.Described one
A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, which is used for
Implementation procedure of the computer program 62 in the load prediction equipment 6 is described.For example, the computer program 62 can be with
It is divided into cutting module, prediction module and output module, each module concrete function is as follows:
Cutting module is used to forecast interval being divided into the multiple first sub- forecast intervals, and historical load data is divided
For the multiple first sub- weight training data;Prediction module is used for described more using the multiple first sub- weight training data training
The deep learning prediction model of a first sub- forecast interval, and obtain the predicted load of every sub- forecast interval;Output module
For determining final predicted value according to the predicted load of each sub- forecast interval.
The load prediction equipment 6 can be the calculating such as desktop PC, notebook, palm PC and cloud server
Equipment.The load prediction equipment may include, but be not limited only to, processor 60, memory 61.Those skilled in the art can manage
Solution, Fig. 6 is only the example of terminal device 6, does not constitute the restriction to terminal device 6, may include more or more than illustrating
Few component perhaps combines certain components or different components, such as the terminal device can also be set including input and output
Standby, network access equipment, bus etc..
Alleged processor 60 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor is also possible to any conventional processor
Deng.
The memory 61 can be the internal storage unit of the terminal device 6, such as the hard disk of load prediction equipment 6
Or memory.The memory 61 is also possible to the External memory equipment of the load prediction equipment 6, such as the load prediction is set
The plug-in type hard disk being equipped on standby 6, intelligent memory card (Smart Media Card, SMC), secure digital (Secure
Digital, SD) card, flash card (Flash Card) etc..Further, the memory 61 can also both include the load
The internal storage unit of pre- measurement equipment 6 also includes External memory equipment.The memory 61 is for storing the computer program
And other programs and data needed for the load prediction equipment.The memory 61 can be also used for temporarily storing
Output or the data that will be exported.
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.Each functional unit in embodiment, module can integrate in one processing unit, can also
To be that each unit physically exists alone, can also be integrated in one unit with two or more units, it is above-mentioned integrated
Unit both can take the form of hardware realization, can also realize in the form of software functional units.In addition, each function list
Member, the specific name of module are also only for convenience of distinguishing each other, the protection scope being not intended to limit this application.Above system
The specific work process of middle unit, module, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
In embodiment provided by the present invention, it should be understood that disclosed device/terminal device and method, it can be with
It realizes by another way.For example, device described above/terminal device embodiment is only schematical, for example, institute
The division of module or unit is stated, only a kind of logical function partition, there may be another division manner in actual implementation, such as
Multiple units or components can be combined or can be integrated into another system, or some features can be ignored or not executed.Separately
A bit, shown or discussed mutual coupling or direct-coupling or communication connection can be through some interfaces, device
Or the INDIRECT COUPLING or communication connection of unit, it can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated module/unit be realized in the form of SFU software functional unit and as independent product sale or
In use, can store in a computer readable storage medium.Based on this understanding, the present invention realizes above-mentioned implementation
All or part of the process in example method, can also instruct relevant hardware to complete, the meter by computer program
Calculation machine program can be stored in a computer readable storage medium, the computer program when being executed by processor, it can be achieved that on
The step of stating each embodiment of the method.Wherein, the computer program includes computer program code, the computer program
Code can be source code form, object identification code form, executable file or certain intermediate forms etc..Computer-readable Jie
Matter may include: can carry the computer program code any entity or device, recording medium, USB flash disk, mobile hard disk,
Magnetic disk, CD, computer storage, read-only memory (ROM, Read-Only Memory), random access memory (RAM,
Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..It should be noted that described
The content that computer-readable medium includes can carry out increasing appropriate according to the requirement made laws in jurisdiction with patent practice
Subtract, such as does not include electric carrier signal and electricity according to legislation and patent practice, computer-readable medium in certain jurisdictions
Believe signal.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations;Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features;And these are modified
Or replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (11)
1. a kind of deep learning load forecasting method characterized by comprising
Forecast interval is divided into the multiple first sub- forecast intervals using the first default traveling time window, and by historical load number
According to the multiple first sub- weight training data are divided into, wherein the length of the length of the historical load data and the forecast interval
It is corresponding;
Mould is predicted using the deep learning of the multiple first the multiple first sub- forecast interval of sub- weight training data training
Type, and obtain the predicted load of every sub- forecast interval;
Final predicted value is determined according to the predicted value of each sub- forecast interval.
2. deep learning load forecasting method as described in claim 1, which is characterized in that the first default traveling time window
It is obtained by clustering algorithm, process are as follows:
It randomly obtains one group second and presets traveling time window, using one group of second preset time window by the historical load
Data are divided into the sub- weight training data of multiple groups second, wherein the sub- weight training data of the multiple groups second are as clustering algorithm
Input;
The smallest clusters number is found according to the measurement standard of clustering algorithm;
The minimum value of clusters number is selected, and using the traveling time window of maximum length corresponding to the minimum value as described first
Default traveling time window.
3. deep learning load forecasting method as claimed in claim 2, which is characterized in that use mean profile coefficient and interval
Measurement standard of the statistic as the clustering algorithm.
4. deep learning load forecasting method as claimed in claim 2, which is characterized in that when described one group second default movement
Between window setting method are as follows:
The initial length of each second default traveling time window is identical, using the described second default traveling time window by the history
Load data is divided into one group of second sub- weight training data, and increases in the upper length to the described second default traveling time window
After adding preset step-length Δ T, according to the current length of the second default traveling time window after increase to the historical load data into
Row divides, until obtaining the sub- weight training data of multiple groups second;Or
The initial length of each second default traveling time window is different, according to the change rate of historical load data setting described second
The length of default traveling time window, and after the upper length to the described second default traveling time window increases preset step-length Δ T,
The historical load data is divided according to the current length of the second default traveling time window after increase, until obtaining more
The second sub- weight training data of group.
5. such as the described in any item deep learning load forecasting methods of claim 2 to 4, which is characterized in that described first is default
The move mode of traveling time window are as follows:
Current first default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t=Twin,i,
Then the initial position of the next first default traveling time window is connected with the end position of the current first default traveling time window,
In, Twin,iFor the length of the current first default traveling time window;Or
Current first default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t < Twin,i, then
The initial position of next first default traveling time window and the current first default traveling time window Twin,iThere are overlappings, wherein
Twin,iFor the length of the current first default traveling time window;
The move mode of the second default traveling time window are as follows:
Current second default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t=Twin,i,
Then the initial position of the next second default traveling time window is connected with the end position of the current second default traveling time window,
In, Twin,iFor the length of the current second default traveling time window;Or
Current second default traveling time window is when performing the next step mobile, if default moving step length Δ t meets Δ t < Twin,i, then
The initial position of next second default traveling time window and the current second default traveling time window Twin,iThere are overlappings, wherein
Twin,iFor the length of the current second default traveling time window.
6. deep learning load forecasting method as claimed in claim 5, which is characterized in that negative using the multiple first son
The deep learning prediction model of the multiple first sub- forecast interval of lotus training data training, and obtain every sub- forecast interval
Before predicted load, the deep learning load forecasting method further include:
The multiple first sub- weight training data are pre-processed, are corrected in the multiple first sub- weight training data
Abnormal data fills up missing data in the multiple first sub- weight training data;
To pretreated multiple first sub- weight training data normalization processing;
The deep learning using the multiple first the multiple first sub- forecast interval of sub- weight training data training is pre-
Model is surveyed, and obtains the predicted load of every sub- forecast interval specifically:
The multiple first sub- weight training data are pre-processed, are corrected in the multiple first sub- weight training data
Abnormal data fills up missing data in the multiple first sub- weight training data;
To pretreated multiple first sub- weight training data normalization processing;
Utilize the depth of the multiple first the multiple first sub- forecast intervals of sub- weight training data training after normalized
Learn prediction model, and obtains the predicted load of every sub- forecast interval.
7. such as the described in any item deep learning load forecasting methods of Claims 1-4, which is characterized in that the deep learning
Prediction model is the depth LSTM prediction model using Keras deep learning framework establishment, the deep learning prediction model
Training process are as follows:
Obtain the first sub- weight training data of i-th of sub- forecast interval;
Utilize grid search LSTM hyper parameter;
Depth LSTM prediction model is established using the described first sub- weight training data and the LSTM hyper parameter.
8. such as the described in any item deep learning load forecasting methods of Claims 1-4, which is characterized in that the deep learning
The algorithm of prediction model is that depth encodes neural network certainly.
9. a kind of deep learning load prediction device characterized by comprising
Cutting module for forecast interval to be divided into the multiple first sub- forecast intervals, and historical load data is divided into
Multiple first sub- weight training data;
Prediction module, for the depth using the multiple first the multiple first sub- forecast interval of sub- weight training data training
Degree study prediction model, and obtain the predicted load of every sub- forecast interval;
Determining module, for determining final predicted value according to the predicted load of each sub- forecast interval.
10. a kind of terminal device, including memory, processor and storage are in the memory and can be on the processor
The computer program of operation, which is characterized in that the processor realizes such as claim 1 to 8 when executing the computer program
The step of any one deep learning load forecasting method.
11. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In realization deep learning load prediction side as described in any one of claim 1 to 8 when the computer program is executed by processor
The step of method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910527965.9A CN110232483B (en) | 2019-06-18 | 2019-06-18 | Deep learning load prediction method and device and terminal equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910527965.9A CN110232483B (en) | 2019-06-18 | 2019-06-18 | Deep learning load prediction method and device and terminal equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110232483A true CN110232483A (en) | 2019-09-13 |
CN110232483B CN110232483B (en) | 2021-05-04 |
Family
ID=67859762
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910527965.9A Active CN110232483B (en) | 2019-06-18 | 2019-06-18 | Deep learning load prediction method and device and terminal equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110232483B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110701732A (en) * | 2019-12-10 | 2020-01-17 | 南昌掘策数据服务有限公司 | Energy consumption data analysis method and system and energy saving method and system of central air conditioner |
CN111126565A (en) * | 2019-11-28 | 2020-05-08 | 广东电网有限责任公司 | Method and device for predicting block load density index based on deep learning |
CN111419249A (en) * | 2020-03-26 | 2020-07-17 | 心图熵动科技(苏州)有限责任公司 | Generation method and prediction system of depression prediction model |
CN111694830A (en) * | 2020-06-12 | 2020-09-22 | 复旦大学 | Missing data completion method based on deep ensemble learning |
CN112330010A (en) * | 2020-11-03 | 2021-02-05 | 长安大学 | Power consumer load interval prediction method based on deep learning |
CN112734106A (en) * | 2021-01-08 | 2021-04-30 | 深圳市国电科技通信有限公司 | Method and device for predicting energy load |
CN112862143A (en) * | 2019-11-28 | 2021-05-28 | 新奥数能科技有限公司 | Load and price prediction method |
CN113761317A (en) * | 2020-07-28 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Bullet screen based data processing method and device |
CN115174237A (en) * | 2022-07-08 | 2022-10-11 | 河北科技大学 | Method and device for detecting malicious traffic of Internet of things system and electronic equipment |
CN116995673A (en) * | 2023-09-26 | 2023-11-03 | 宁德时代新能源科技股份有限公司 | Power load prediction method, power load prediction model training method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103606013A (en) * | 2013-12-06 | 2014-02-26 | 国家电网公司 | User annual power consumption prediction method based on support vector machine |
CN103901305A (en) * | 2014-03-31 | 2014-07-02 | 广东电网公司电力科学研究院 | Online early warning method for power equipment |
US20170206615A1 (en) * | 2012-01-23 | 2017-07-20 | Earth Networks, Inc. | Optimizing and controlling the energy consumption of a building |
CN108009673A (en) * | 2017-11-24 | 2018-05-08 | 国网北京市电力公司 | Novel load Forecasting Methodology and device based on deep learning |
-
2019
- 2019-06-18 CN CN201910527965.9A patent/CN110232483B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170206615A1 (en) * | 2012-01-23 | 2017-07-20 | Earth Networks, Inc. | Optimizing and controlling the energy consumption of a building |
CN103606013A (en) * | 2013-12-06 | 2014-02-26 | 国家电网公司 | User annual power consumption prediction method based on support vector machine |
CN103901305A (en) * | 2014-03-31 | 2014-07-02 | 广东电网公司电力科学研究院 | Online early warning method for power equipment |
CN108009673A (en) * | 2017-11-24 | 2018-05-08 | 国网北京市电力公司 | Novel load Forecasting Methodology and device based on deep learning |
Non-Patent Citations (2)
Title |
---|
SUN DA-SHUAI,MA LI-XIN,WANG SHOU-ZHENG: "The Design of Short-term Load Forecast Systems Based on the Theory of Complex", 《2010 INTERNATIONAL CONFERENCE ON INTELLIGENT SYSTEM DESIGN AND ENGINEERING APPLICATION》 * |
王颖: ""典型发达和发展中国家电力负荷预测分析"", 《中国优秀硕士学位论文全文数据库经济与管理科学辑》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111126565A (en) * | 2019-11-28 | 2020-05-08 | 广东电网有限责任公司 | Method and device for predicting block load density index based on deep learning |
CN112862143A (en) * | 2019-11-28 | 2021-05-28 | 新奥数能科技有限公司 | Load and price prediction method |
CN110701732A (en) * | 2019-12-10 | 2020-01-17 | 南昌掘策数据服务有限公司 | Energy consumption data analysis method and system and energy saving method and system of central air conditioner |
CN111419249A (en) * | 2020-03-26 | 2020-07-17 | 心图熵动科技(苏州)有限责任公司 | Generation method and prediction system of depression prediction model |
CN111694830A (en) * | 2020-06-12 | 2020-09-22 | 复旦大学 | Missing data completion method based on deep ensemble learning |
CN113761317A (en) * | 2020-07-28 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Bullet screen based data processing method and device |
CN112330010A (en) * | 2020-11-03 | 2021-02-05 | 长安大学 | Power consumer load interval prediction method based on deep learning |
CN112734106A (en) * | 2021-01-08 | 2021-04-30 | 深圳市国电科技通信有限公司 | Method and device for predicting energy load |
CN115174237A (en) * | 2022-07-08 | 2022-10-11 | 河北科技大学 | Method and device for detecting malicious traffic of Internet of things system and electronic equipment |
CN116995673A (en) * | 2023-09-26 | 2023-11-03 | 宁德时代新能源科技股份有限公司 | Power load prediction method, power load prediction model training method and device |
CN116995673B (en) * | 2023-09-26 | 2024-02-20 | 宁德时代新能源科技股份有限公司 | Power load prediction method, power load prediction model training method and device |
Also Published As
Publication number | Publication date |
---|---|
CN110232483B (en) | 2021-05-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110232483A (en) | Deep learning load forecasting method, device and terminal device | |
CN108846517B (en) | Integration method for predicating quantile probabilistic short-term power load | |
WO2017076154A1 (en) | Method and apparatus for predicting network event and establishing network event prediction model | |
CN108364106A (en) | A kind of expense report Risk Forecast Method, device, terminal device and storage medium | |
CN111160617B (en) | Power daily load prediction method and device | |
CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
CN105760952A (en) | Load prediction method based on Kalman filtering and self-adaptive fuzzy neural network | |
CN109102155B (en) | Ultra-short-term node marginal electricity price probability prediction method and system | |
Tian et al. | A network traffic prediction method based on IFS algorithm optimised LSSVM | |
CN116307215A (en) | Load prediction method, device, equipment and storage medium of power system | |
CN102768701A (en) | High-voltage switch cabinet insulator electric field optimization method based on quantum genetic algorithm | |
CN109508826A (en) | The schedulable capacity prediction methods of electric car cluster of decision tree are promoted based on gradient | |
CN109633448A (en) | Identify the method, apparatus and terminal device of cell health state | |
Lee et al. | Probabilistic wind power forecasting based on the laplace distribution and golden search | |
CN111985719A (en) | Power load prediction method based on improved long-term and short-term memory network | |
CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
CN113516275A (en) | Power distribution network ultra-short term load prediction method and device and terminal equipment | |
AU2021106200A4 (en) | Wind power probability prediction method based on quantile regression | |
CN115310650A (en) | Low-complexity high-precision time sequence multi-step prediction method and system | |
CN117094535B (en) | Artificial intelligence-based energy supply management method and system | |
Langsari et al. | Optimizing COCOMO II parameters using particle swarm method | |
CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
CN111259340B (en) | Saturation load prediction method based on logistic regression | |
CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
CN116470491A (en) | Photovoltaic power probability prediction method and system based on copula function |
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 |