CN107145976A - A kind of method for predicting user power utilization load - Google Patents
A kind of method for predicting user power utilization load Download PDFInfo
- Publication number
- CN107145976A CN107145976A CN201710292676.6A CN201710292676A CN107145976A CN 107145976 A CN107145976 A CN 107145976A CN 201710292676 A CN201710292676 A CN 201710292676A CN 107145976 A CN107145976 A CN 107145976A
- Authority
- CN
- China
- Prior art keywords
- power load
- user
- influence index
- power
- class
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/80—Management or planning
- Y02P90/82—Energy audits or management systems therefor
Abstract
The present invention provides a kind of method for predicting user power utilization load, using the teaching of the invention it is possible to provide the accuracy rate predicted the outcome.Methods described includes:The power load historical data of multiple users is obtained, the power load historical data to acquisition is classified, regard the partial history data in every class result as training set;Build the power load influence index of user;Using the power load influence index of structure as feature, the training set as object set, grader is built using decision tree;User to be predicted is obtained for the desired value of the power load influence index built, desired value is inputted into the grader to predict the classification belonging to the user to be predicted, the power load that the power load historical data for user's generic to be predicted passes through user to be predicted described in Neural Network Prediction.The present invention relates to technical field of power systems.
Description
Technical field
The present invention relates to technical field of power systems, a kind of method for predicting user power utilization load is particularly related to.
Background technology
With intelligent grid, the construction of microgrid and the development of information technology so that with the function such as prediction and early warning
The construction of intelligence system receives much concern.The load prediction data more voluminous of key effect is played for electricity market decision-making and miscellaneous
Disorderly, accurately and timely predict that electric load plays an important role in Power System Planning and operation.In the prior art, it is general to utilize
The power load information of all users got predicts the customer charge of single user, and the accuracy rate that predicts the outcome is low.
The content of the invention
The technical problem to be solved in the present invention is to provide a kind of method for predicting user power utilization load, to solve prior art
The problem of existing accuracy rate that predicts the outcome is low.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of method for predicting user power utilization load, including:
The power load historical data of multiple users is obtained, the power load historical data to acquisition is classified, will be every
Partial history data in class result are used as training set;
Build the power load influence index of user;
Using the power load influence index of structure as feature, the training set as object set, using decision tree come structure
Build grader;
User to be predicted is obtained for the desired value of the power load influence index built, desired value is inputted into the classification
Device predicts the classification belonging to the user to be predicted, for the power load history number of user's generic to be predicted
According to the power load by user to be predicted described in Neural Network Prediction.
Further, the power load historical data of described pair of acquisition, which carries out classification, includes:
The power load historical data of acquisition is classified using hierarchical clustering algorithm.
Further, it is described that the power load historical data progress classification of acquisition is included using hierarchical clustering algorithm:
The power load historical data of acquisition is analyzed using Agglomerative Hierarchical Clustering algorithm, walked by daily power load
Gesture is clustered, divides classification.
Further, the power load influence index includes:Temperature, week, whether be festivals or holidays, at one day festivals or holidays,
Two days festivals or holidays, three days festivals or holidays, four days festivals or holidays and five days festivals or holidays.
Further, the power load influence index for building user includes:
Primarily determine that the power load influence index of user;
The degree of association that power load is determined between each power load influence index for primarily determining that;
Judge whether the degree of association is more than predetermined threshold value;
If more than default threshold value, the primarily determine that and degree of association to be more than to the power load influence index of predetermined threshold value
It is used as final power load influence index.
Further, the degree of association determined between power load and each power load influence index primarily determined that
Including:
Using gray relative analysis method, power load is determined between each power load influence index for primarily determining that
The degree of association.
Further, the power load influence index using structure is used as object set, profit as feature, the training set
Included with decision tree to build grader:
S1, using the power load influence index of structure as feature, all training samples regard one as in the training set
Node;
S2, travels through each partitioning scheme of each power load influence index, according to default determination the best cutting point
Condition, determine the best cutting point;
S3, according to the best cutting point of determination, by a node allocation into two nodes N1 and N2
S4, S2-S3 is continued executing with to node N1 and N2 respectively, until each node meets default class condition.
Further, using the power load influence index of structure, as feature, training set utilizes decision-making as object set
Tree is come after building grader, methods described also includes:
It regard the remaining historical data in every class result as test set;
Grader is verified using power load influence index and the test set, by the result record mixed
Confuse in matrix;
According to confusion matrix modified result grader.
Further, obtaining the desired value of power load influence index of the prediction user treated for building will predict
The desired value of certain day.
The above-mentioned technical proposal of the present invention has the beneficial effect that:
In such scheme, the power load historical data of multiple users is obtained, the power load historical data to acquisition is entered
Row classification, regard the partial history data in every class result as training set;Build the power load influence index of user;It will build
Power load influence index as feature, the training set as object set, build grader using decision tree;Acquisition is treated
User is predicted for the desired value of the power load influence index built, it is described to predict that desired value is inputted into the grader
Classification belonging to user to be predicted, the power load historical data for user's generic to be predicted passes through neutral net
Algorithm predicts the power load of the user to be predicted.So, the power load of the classification according to belonging to user to be predicted is passed through
Historical data, by the power load of user to be predicted described in Neural Network Prediction, with utilizing all users got
Power load information predict that the customer charge of single user is compared, it is more accurate to predict the outcome.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of the method for prediction user power utilization load provided in an embodiment of the present invention;
Fig. 2 is the power load schematic diagram data in the whole year in 1997 provided in an embodiment of the present invention;
Fig. 3 is cluster result schematic diagram one provided in an embodiment of the present invention;
Fig. 4 is cluster result schematic diagram two provided in an embodiment of the present invention;
Fig. 5 is classification results schematic diagram provided in an embodiment of the present invention;
Fig. 6 is training result schematic diagram provided in an embodiment of the present invention.
Embodiment
To make the technical problem to be solved in the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and tool
Body embodiment is described in detail.
There is provided a kind of side for predicting user power utilization load the problem of low for the existing accuracy rate that predicts the outcome by the present invention
Method.
Embodiment one
As shown in figure 1, the method for prediction user power utilization load provided in an embodiment of the present invention, including:
Step 101, the power load historical data of multiple users is obtained, the power load historical data to acquisition is divided
Class, regard the partial history data in every class result as training set;
Step 102, the power load influence index of user is built;
Step 103, using the power load influence index of structure as feature, the training set as object set, using certainly
Plan tree builds grader;
Step 104, user to be predicted is obtained for the desired value of the power load influence index built, and desired value is inputted
The grader predicts the classification belonging to the user to be predicted, and the electricity consumption for user's generic to be predicted is born
The power load that lotus historical data passes through user to be predicted described in Neural Network Prediction.
The method of prediction user power utilization load described in the embodiment of the present invention, obtains the power load history number of multiple users
According to the power load historical data to acquisition is classified, and regard the partial history data in every class result as training set;Build
The power load influence index of user;Using the power load influence index of structure as feature, the training set as object set,
Grader is built using decision tree;User to be predicted is obtained for the desired value of the power load influence index built, will be referred to
Scale value inputs the grader to predict the classification belonging to the user to be predicted, for user's generic to be predicted
The power load historical data power load that passes through user to be predicted described in Neural Network Prediction.So, basis is passed through
The power load historical data of classification belonging to user to be predicted, passes through the use of user to be predicted described in Neural Network Prediction
Electric load, compared with the customer charge using the power load information of all users got to predict single user, prediction knot
Fruit is more accurate.
In the present embodiment, the user refers generally to the big user of power load, for example, industrial user.Grid company can be with
According to the power load of the user to be predicted predicted, Forewarning Measures are taken in advance.
In the embodiment of the method for foregoing prediction user power utilization load, further, the use of described pair of acquisition
Electric load historical data, which carries out classification, to be included:
The power load historical data of acquisition is classified using hierarchical clustering algorithm.
In the embodiment of the method for foregoing prediction user power utilization load, further, the utilization level gathers
Class algorithm carries out classification to the power load historical data of acquisition to be included:
The power load historical data of acquisition is analyzed using Agglomerative Hierarchical Clustering algorithm, walked by daily power load
Gesture is clustered, divides classification.
In the present embodiment, the power load historical data of acquisition can be subjected to visualization and shown, and using cohesion level
Clustering algorithm is analyzed the power load historical data of acquisition, is clustered to divide class by daily power load tendency
Not, classification results can also be refined by week.
It is further, described to build user's in the embodiment of the method for foregoing prediction user power utilization load
Power load influence index includes:
Primarily determine that the power load influence index of user;
The degree of association that power load is determined between each power load influence index for primarily determining that;
Judge whether the degree of association is more than predetermined threshold value;
If more than default threshold value, the primarily determine that and degree of association to be more than to the power load influence index of predetermined threshold value
It is used as final power load influence index.
In the present embodiment, the power load influence index of user is primarily determined that, and try to achieve power load and primarily determine that
The degree of association between each power load influence index, judgement obtains whether the degree of association is more than predetermined threshold value, for example, described default
Threshold value can be 0.5;It regard the power load influence index that the primarily determine that and degree of association is more than 0.5 as final power load
Influence index.
In the present embodiment, by the correlation analysis between power load and power load influence index, obtain 8 with
The related index of electro-load forecast, including:Temperature, week, whether it is that festivals or holidays, one day festivals or holidays, two days festivals or holidays, section are false
Three days days, four days festivals or holidays, five days festivals or holidays;That is, the power load influence index finally built includes:Temperature, week, whether
It is festivals or holidays, one day festivals or holidays, two days festivals or holidays, three days festivals or holidays, four days festivals or holidays and five days festivals or holidays.
In the present embodiment, the festivals or holidays include:The Ching Ming Festival, the Dragon Boat Festival, May Day, 11, the festivals or holidays such as Spring Festival.
In the embodiment of the method for foregoing prediction user power utilization load, further, the determination electricity consumption is born
The degree of association between lotus and each power load influence index primarily determined that includes:
Using gray relative analysis method, power load is determined between each power load influence index for primarily determining that
The degree of association.
In the embodiment of the method for foregoing prediction user power utilization load, further, the use by structure
Electric load influence index, as object set, is included as feature, the training set using decision tree to build grader:
S1, using the power load influence index of structure as feature, all training samples regard one as in the training set
Node;
S2, travels through each partitioning scheme of each power load influence index, according to default determination the best cutting point
Condition, determine the best cutting point;
The target variable of decision tree can have two kinds:
1) numeric type:Types of variables is " temperature " in integer or floating number, such as index, can use ">=", ">”,“<”
Or "<=" etc. symbol be used as partitioning scheme;
2) title type:Whether " being festivals or holidays " in the enumeration type in similar programming language, such as index, can only " be
Festivals or holidays ", " not being festivals or holidays " can use "=" as partitioning scheme;
If current all data (training sample) can be divided into two classes by a cut-point, then, the cut-point
Exactly meet the best cutting point of the condition of default determination the best cutting point.
S3, according to the best cutting point of determination, by a node allocation into two nodes N1 and N2
S4, S2-S3 is continued executing with to node N1 and N2 respectively, until each node meets default class condition.
In the present embodiment, after S4 terminates, the structure of grader is completed, afterwards, it is necessary to be tested to the grader of structure
And correct.
In the embodiment of the method for foregoing prediction user power utilization load, further, by the electricity consumption of structure
Loading effects index is built after grader, methods described is also wrapped as feature, training set as object set using decision tree
Include:
It regard the remaining historical data in every class result as test set;
Grader is verified using power load influence index and the test set, by the result record mixed
Confuse in matrix;
According to confusion matrix modified result grader.
Embodiment two
For a better understanding of the present invention embodiment provide prediction user power utilization load method, to the embodiment of the present invention
The method of the prediction user power utilization load of offer is described in detail:
B1, obtains the power load historical data of multiple users.
In the present embodiment, the power load historical data of acquisition is the power load data in the whole year in 1997, by acquisition
Power load historical data carries out visualization and shown, as shown in Figure 2;
B2, is analyzed the power load historical data of acquisition using Agglomerative Hierarchical Clustering algorithm, negative by daily electricity consumption
Lotus tendency is clustered, divides classification, and obtained cluster result is as shown in figure 3, in Fig. 3, cluster dendrogram's contains
Justice is cluster, and Height represents the distance between cluster and cluster, and it is the date of bottom branch that abscissa is corresponding.
According to the distance between each cluster, the power load historical data of acquisition is divided into 6 classes, as shown in Fig. 4, Fig. 5, table 1.
Numeral in table 1 represents the numeral 15 in the quantity in each week in each class, such as first digit cell, represents the subordinate of class 1
In Monday number of days be 15.
The quantity in each week in each class of table 1
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
Monday | 15 | 8 | 2 | 5 | 12 | 10 |
Tuesday | 12 | 11 | 5 | 2 | 15 | 7 |
Wednesday | 13 | 11 | 5 | 3 | 17 | 4 |
Thursday | 10 | 13 | 4 | 4 | 14 | 7 |
Friday | 12 | 9 | 6 | 4 | 13 | 8 |
Saturday | 14 | 0 | 5 | 10 | 8 | 15 |
Sunday | 5 | 0 | 9 | 13 | 5 | 20 |
As it can be seen from table 1 class 2 and class 5 are mainly working day power load, class 4 and class 6 are mainly weekend electricity consumption and born
Lotus, class 1 be one week in addition to Sunday load, class 3 be all-round power load.
Secondly 6 dimensions are extended for festivals or holidays single dimension, whether be respectively is festivals or holidays, one day festivals or holidays, festivals or holidays
Two days, three days festivals or holidays, four days festivals or holidays, five days festivals or holidays.Obtain the relation between class categories and festivals or holidays, such as table 2-5
Shown in table:
Class categories of table 2 and whether be relation between festivals or holidays
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
It is no | 79 | 52 | 33 | 37 | 82 | 67 |
It is | 2 | 0 | 3 | 4 | 2 | 4 |
Relation between the class categories of table 3 and one day festivals or holidays
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
It is no | 75 | 50 | 30 | 36 | 74 | 63 |
It is | 6 | 2 | 6 | 5 | 10 | 8 |
Relation between the class categories of table 4 and two days festivals or holidays
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
It is no | 71 | 47 | 29 | 35 | 66 | 61 |
It is | 10 | 5 | 7 | 6 | 18 | 10 |
Relation between the class categories of table 5 and three days festivals or holidays
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
It is no | 66 | 46 | 28 | 35 | 59 | 58 |
It is | 15 | 6 | 8 | 6 | 25 | 13 |
Relation between the class categories of table 6 and four days festivals or holidays
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
It is no | 61 | 45 | 27 | 35 | 52 | 57 |
It is | 20 | 7 | 9 | 6 | 32 | 14 |
Relation between the class categories of table 7 and five days festivals or holidays
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
It is no | 57 | 44 | 26 | 33 | 48 | 54 |
It is | 24 | 8 | 10 | 8 | 36 | 17 |
B3, primarily determines that the power load influence index of user;Determine power load and each class electricity consumption primarily determined that
The degree of association between loading effects index, as shown in table 8.Calculation of relationship degree formula is:
Wherein, rjRepresent the degree of association of jth day;εj(k) incidence coefficient of jth day and k-th of index is represented;N represents index
Sum;ρ represents resolution ratio, typically between 0~1, generally takes 0.5;Δ (max) represents the data sequence of system action feature
The data sequence maximum difference that row are constituted with index;Δ (min) represents what the data sequence of system action feature and index were constituted
Data sequence minimal difference;Δj(k) k-th of index of jth day that the data sequence of system action feature is constituted with index is represented
The absolute difference of data sequence.In the present embodiment, Δ (max)=2.076483, Δ (min)=0.002054705.
Table 7 associates angle value
Power load influence index of the degree of association more than 0.5 is selected as final power load influence index, then finally
Power load influence index it is as shown in table 9.
The final power load influence index of table 9
Numbering | fuheMean21 | Description |
D01 | temperature21 | Temperature |
D02 | weekNum21 | Week |
D03 | isholiday | Whether it is festivals or holidays |
D04 | isholiday_1 | At one day festivals or holidays |
D05 | isholiday_2 | At two days festivals or holidays |
D06 | isholiday_3 | At three days festivals or holidays |
D07 | isholiday_4 | At four days festivals or holidays |
D08 | isholiday_5 | At five days festivals or holidays |
B4, grader is built using decision tree
Using 70% in every class result as training set, using cart algorithms to power load influence index and training set
It is trained, as a result as shown in Figure 6.
B5, using 30% in every class result as test set, utilizes power load influence index and the test set pair
Grader is verified, by the result record in confusion matrix (table 10), according to confusion matrix modified result grader.
Obtained error rate of being tested according to test set is 48.35%;
The confusion matrix of table 10
Class 1 | Class 2 | Class 3 | Class 4 | Class 5 | Class 6 | |
Class 1 | 12 | 4 | 1 | 2 | 1 | 0 |
Class 2 | 4 | 9 | 0 | 0 | 0 | 0 |
Class 3 | 6 | 0 | 0 | 1 | 2 | 0 |
Class 4 | 4 | 0 | 1 | 4 | 0 | 1 |
Class 5 | 2 | 0 | 0 | 0 | 10 | 9 |
Class 6 | 0 | 0 | 0 | 0 | 6 | 12 |
Correct for the first time:
Reason is checked by confusion matrix, for example, in table 10 first cell represent be class 1 originally, it is accurate pre-
Survey is that the number of class 1 is 12, and the number of error prediction is 8.
It can be seen that class 1 and class 4 are more by mutual prediction error by table 10, what is be equally mutually easily predicted has the He of class 1
Class 3, class 1 and class 2, class 5 and 6, therefore the class 6 of class 5 is classified as 1 class, remaining class 2, class 3, class 4 are respectively divided into 1 class, and class 1 is removed, finally examined
Worry gathers for four classes again.
Once grader finds that error rate is 37.36% to re -training, although reduction, but still higher.
Second of amendment:
Now add index:Month and day are placed in grader, and now error rate will be 28.57%;
Optimized algorithm:
The method for being now considered as random forest is predicted, and now error rate is reduced to 19.7%.
Four classes obtained after cluster are predicted respectively using neutral net:
Assuming that power load is replaced with h, t represents certain day, then anticipation function is as follows:
ht=f (temperaturet,weekNumt,isholidayt,isholiday_1t,isholiday_2t,
isholiday_3t,isholiday_4t,isholiday_5t,Month,Day)
For the first classification:
For clustering the power load historical data for first category, therefrom choose last 10 groups of data to build prediction BP
Neutral net, (wherein, predicted value is the predicted value of power load historical data, actual value to the result of the last 10 groups of data of prediction
For power load historical data actual value) be:
[1] " predicted value:699.9431967012 actual value:692 errors:1.14786079497107% "
[1] " predicted value:699.943182748405 actual value:683 errors:2.48070025598908% "
[1] " predicted value:699.943178436882 actual value:681 errors:2.7816708424202% "
[1] " predicted value:699.943176087034 actual value:637 errors:9.88118933862379% "
[1] " predicted value:699.943183092937 actual value:635 errors:10.2272729280215% "
[1] " predicted value:699.943376758066 actual value:641 errors:9.19553459564209% "
[1] " predicted value:699.944323182212 actual value:648 errors:8.01609925651419% "
[1] " predicted value:700.133389768538 actual value:677 errors:3.41704427895682% "
[1] " predicted value:701.058473529607 actual value:700 errors:0.15121050422955% "
[1] " predicted value:707.179937916174 actual value:697 errors:1.46053628639506"
For second of classification, predict the outcome for:
[1] " predicted value:701.298430735931 actual value:612 errors:14.5912468522763% "
[1] " predicted value:701.298430735931 actual value:649 errors:8.05830982063648% "
[1] " predicted value:701.298430735931 actual value:658 errors:6.58030862248188% "
[1] " predicted value:701.298430735931 actual value:628 errors:11.6717246394794% "
[1] " predicted value:701.298430735931 actual value:652 errors:7.56110900857834% "
[1] " predicted value:701.298430735931 actual value:618 errors:13.4787104750697% "
[1] " predicted value:701.298430735931 actual value:616 errors:13.847147846742% "
[1] " predicted value:701.298430735931 actual value:618 errors:13.4787104750697% "
[1] " predicted value:701.298430735931 actual value:623 errors:12.5679664102618% "
[1] " predicted value:701.298430735931 actual value:604 errors:16.1090117112468% "
For the third classification, predict the outcome for:
[1] " predicted value:698.48242752988 actual value:759 errors:- 7.97332970620812% "
[1] " predicted value:698.482425888167 actual value:728 errors:- 4.05461182854852% "
[1] " predicted value:698.48242543383 actual value:681 errors:2.56716966722906% "
[1] " predicted value:698.482424576936 actual value:672 errors:3.94083699061545% "
[1] " predicted value:698.482424246907 actual value:674 errors:3.6324071582948% "
[1] " predicted value:698.482423497784 actual value:668 errors:4.5632370505664% "
[1] " predicted value:698.482423176937 actual value:664 errors:5.19313602062306% "
[1] " predicted value:698.482422417514 actual value:668 errors:4.56323688884937% "
[1] " predicted value:698.4824220861 actual value:680 errors:2.71800324795581% "
[1] " predicted value:698.482421297923 actual value:690 errors:1.22933641998879% "
For the 4th kind of classification, predict the outcome for:
[1] " predicted value:698.745009840381 actual value:759 errors:- 7.93873388137271% "
[1] " predicted value:698.707739712033 actual value:728 errors:- 4.02366212746801% "
[1] " predicted value:698.691596108352 actual value:681 errors:2.59788489109434% "
[1] " predicted value:698.650335098196 actual value:672 errors:3.96582367532679% "
[1] " predicted value:698.631144775431 actual value:674 errors:3.65447251860998% "
[1] " predicted value:698.585528799011 actual value:668 errors:4.57867197589989% "
[1] " predicted value:698.567019253976 actual value:664 errors:5.20587639367102% "
[1] " predicted value:698.530413410825 actual value:668 errors:4.57042116928511% "
[1] " predicted value:698.518393499634 actual value:680 errors:2.72329316171086% "
[1] " predicted value:698.499121575287 actual value:690 errors:1.23175675004162% "
By predicting the outcome:The error rate of first kind prediction is 4.8717%, and the error rate of Equations of The Second Kind prediction is
10.782%, the error rate of the 3rd class prediction is 2.867%, and the error rate of Equations of The Second Kind prediction is 1.65%, and prediction error rate exists
In the range of permission, Neural Network Prediction power load can be passed through.
In the present embodiment, the desired value for obtaining power load influence index of the user to be predicted for building (is refered in particular to:It is pre-
The desired value of certain day surveyed), desired value is inputted into the grader to predict the classification belonging to the user to be predicted, for
The power load historical data of user's generic to be predicted passes through user to be predicted described in Neural Network Prediction
Power load.
It should be noted that herein, such as first and second or the like relational terms are used merely to a reality
Body or operation make a distinction with another entity or operation, and not necessarily require or imply these entities or deposited between operating
In any this actual relation or order.
Described above is the preferred embodiment of the present invention, it is noted that for those skilled in the art
For, on the premise of principle of the present invention is not departed from, some improvements and modifications can also be made, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (9)
1. a kind of method for predicting user power utilization load, it is characterised in that including:
The power load historical data of multiple users is obtained, the power load historical data to acquisition is classified, will be per class knot
Partial history data in fruit are used as training set;
Build the power load influence index of user;
Using the power load influence index of structure as feature, the training set as object set, built using decision tree point
Class device;
User to be predicted is obtained for the desired value of the power load influence index built, desired value is inputted into the grader
The classification belonging to the user to be predicted is predicted, the power load historical data for user's generic to be predicted is led to
Cross the power load of user to be predicted described in Neural Network Prediction.
2. the method for prediction user power utilization load according to claim 1, it is characterised in that the electricity consumption of described pair of acquisition is born
Lotus historical data, which carries out classification, to be included:
The power load historical data of acquisition is classified using hierarchical clustering algorithm.
3. the method for prediction user power utilization load according to claim 2, it is characterised in that the utilization hierarchical clustering is calculated
Method carries out classification to the power load historical data of acquisition to be included:
The power load historical data of acquisition is analyzed using Agglomerative Hierarchical Clustering algorithm, entered by daily power load tendency
Row cluster, division classification.
4. the method for prediction user power utilization load according to claim 1, it is characterised in that the power load influence refers to
Mark includes:Temperature, week, whether be festivals or holidays, at one day festivals or holidays, at two days festivals or holidays, at three days festivals or holidays, at four days festivals or holidays and section
At five days holiday.
5. the method for prediction user power utilization load according to claim 1, it is characterised in that the electricity consumption of the structure user
Loading effects index includes:
Primarily determine that the power load influence index of user;
The degree of association that power load is determined between each power load influence index for primarily determining that;
Judge whether the degree of association is more than predetermined threshold value;
If more than default threshold value, the primarily determine that and degree of association is more than the power load influence index of predetermined threshold value as
Final power load influence index.
6. it is according to claim 5 prediction user power utilization load method, it is characterised in that the determination power load with
The degree of association between each power load influence index primarily determined that includes:
Using gray relative analysis method, associating between power load and each power load influence index primarily determined that is determined
Degree.
7. the method for prediction user power utilization load according to claim 1, it is characterised in that described to bear the electricity consumption of structure
Lotus influence index, as object set, is included as feature, the training set using decision tree to build grader:
S1, using the power load influence index of structure as feature, all training samples regard a node as in the training set;
S2, travels through each partitioning scheme of each power load influence index, according to the bar of default determination the best cutting point
Part, determines the best cutting point;
S3, according to the best cutting point of determination, by a node allocation into two nodes N1 and N2
S4, S2-S3 is continued executing with to node N1 and N2 respectively, until each node meets default class condition.
8. the method for prediction user power utilization load according to claim 1, it is characterised in that by the power load of structure
Influence index is built after grader, methods described also includes as feature, training set as object set using decision tree:
It regard the remaining historical data in every class result as test set;
Grader is verified using power load influence index and the test set, the result record is being obscured into square
In battle array;
According to confusion matrix modified result grader.
9. the method for prediction user power utilization load according to claim 1, it is characterised in that obtain the prediction user's pin treated
Desired value to the power load influence index of structure is the desired value of certain day to be predicted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710292676.6A CN107145976A (en) | 2017-04-28 | 2017-04-28 | A kind of method for predicting user power utilization load |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710292676.6A CN107145976A (en) | 2017-04-28 | 2017-04-28 | A kind of method for predicting user power utilization load |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107145976A true CN107145976A (en) | 2017-09-08 |
Family
ID=59775165
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710292676.6A Pending CN107145976A (en) | 2017-04-28 | 2017-04-28 | A kind of method for predicting user power utilization load |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107145976A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107784455A (en) * | 2017-11-16 | 2018-03-09 | 鼎信信息科技有限责任公司 | A kind of power network peak load shifting method based on trade power consumption database and distribution transforming weight overload prediction |
CN107944495A (en) * | 2017-12-04 | 2018-04-20 | 贵州电网有限责任公司电力科学研究院 | A kind of household electricity load classification recognition methods based on deep layer forest algorithm |
CN108062560A (en) * | 2017-12-04 | 2018-05-22 | 贵州电网有限责任公司电力科学研究院 | A kind of power consumer feature recognition sorting technique based on random forest |
CN108491893A (en) * | 2018-04-13 | 2018-09-04 | 贵州电网有限责任公司 | A kind of household electricity load classification recognition methods based on deep layer forest algorithm |
CN108539738A (en) * | 2018-05-10 | 2018-09-14 | 国网山东省电力公司电力科学研究院 | A kind of short-term load forecasting method promoting decision tree based on gradient |
CN109034905A (en) * | 2018-08-03 | 2018-12-18 | 四川长虹电器股份有限公司 | The method for promoting sales volume prediction result robustness |
CN109214458A (en) * | 2018-09-19 | 2019-01-15 | 合肥工业大学 | A kind of city load quantization method based on historical data |
CN109359786A (en) * | 2018-12-05 | 2019-02-19 | 国网江苏省电力有限公司扬州供电分公司 | A kind of power station area short-term load forecasting method |
CN109376971A (en) * | 2018-12-29 | 2019-02-22 | 北京中电普华信息技术有限公司 | A kind of load curve forecasting method and system towards power consumer |
CN109583086A (en) * | 2018-11-30 | 2019-04-05 | 国网河北省电力有限公司电力科学研究院 | Distribution transformer heavy-overload prediction technique and terminal device |
CN109784748A (en) * | 2019-01-25 | 2019-05-21 | 广东电网有限责任公司 | User power utilization behavior discrimination method and device under a kind of market and competitive mechanisms |
WO2019134254A1 (en) * | 2018-01-02 | 2019-07-11 | 上海交通大学 | Real-time economic dispatch calculation method using distributed neural network |
CN110956306A (en) * | 2019-10-23 | 2020-04-03 | 广东电网有限责任公司 | Load prediction method based on load clustering |
CN111539841A (en) * | 2019-12-31 | 2020-08-14 | 远景智能国际私人投资有限公司 | Electric quantity prediction method, device, equipment and readable storage medium |
CN112734135A (en) * | 2021-01-26 | 2021-04-30 | 吉林大学 | Power load prediction method, intelligent terminal and computer readable storage medium |
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 |
-
2017
- 2017-04-28 CN CN201710292676.6A patent/CN107145976A/en active Pending
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107784455A (en) * | 2017-11-16 | 2018-03-09 | 鼎信信息科技有限责任公司 | A kind of power network peak load shifting method based on trade power consumption database and distribution transforming weight overload prediction |
CN107944495A (en) * | 2017-12-04 | 2018-04-20 | 贵州电网有限责任公司电力科学研究院 | A kind of household electricity load classification recognition methods based on deep layer forest algorithm |
CN108062560A (en) * | 2017-12-04 | 2018-05-22 | 贵州电网有限责任公司电力科学研究院 | A kind of power consumer feature recognition sorting technique based on random forest |
WO2019134254A1 (en) * | 2018-01-02 | 2019-07-11 | 上海交通大学 | Real-time economic dispatch calculation method using distributed neural network |
US11216759B2 (en) | 2018-01-02 | 2022-01-04 | Shanghai Jiao Tong University | Real-time economic dispatch method of power system |
CN108491893A (en) * | 2018-04-13 | 2018-09-04 | 贵州电网有限责任公司 | A kind of household electricity load classification recognition methods based on deep layer forest algorithm |
CN108539738A (en) * | 2018-05-10 | 2018-09-14 | 国网山东省电力公司电力科学研究院 | A kind of short-term load forecasting method promoting decision tree based on gradient |
CN108539738B (en) * | 2018-05-10 | 2020-04-21 | 国网山东省电力公司电力科学研究院 | Short-term load prediction method based on gradient lifting decision tree |
CN109034905A (en) * | 2018-08-03 | 2018-12-18 | 四川长虹电器股份有限公司 | The method for promoting sales volume prediction result robustness |
CN109214458A (en) * | 2018-09-19 | 2019-01-15 | 合肥工业大学 | A kind of city load quantization method based on historical data |
CN109214458B (en) * | 2018-09-19 | 2021-08-13 | 合肥工业大学 | Urban load quantification method based on historical data |
CN109583086A (en) * | 2018-11-30 | 2019-04-05 | 国网河北省电力有限公司电力科学研究院 | Distribution transformer heavy-overload prediction technique and terminal device |
CN109359786A (en) * | 2018-12-05 | 2019-02-19 | 国网江苏省电力有限公司扬州供电分公司 | A kind of power station area short-term load forecasting method |
CN109376971A (en) * | 2018-12-29 | 2019-02-22 | 北京中电普华信息技术有限公司 | A kind of load curve forecasting method and system towards power consumer |
CN109784748A (en) * | 2019-01-25 | 2019-05-21 | 广东电网有限责任公司 | User power utilization behavior discrimination method and device under a kind of market and competitive mechanisms |
CN110956306A (en) * | 2019-10-23 | 2020-04-03 | 广东电网有限责任公司 | Load prediction method based on load clustering |
WO2021137759A1 (en) * | 2019-12-31 | 2021-07-08 | Envision Digital International Pte. Ltd. | Method and apparatus for predicting power consumption, device and readiable storage medium |
CN111539841A (en) * | 2019-12-31 | 2020-08-14 | 远景智能国际私人投资有限公司 | Electric quantity prediction method, device, equipment and readable storage medium |
US20230049089A1 (en) * | 2019-12-31 | 2023-02-16 | Envision Digital International Pte. Ltd. | Method and Apparatus For Predicting Power Consumption, Device and Readiable Storage Medium |
CN112734135A (en) * | 2021-01-26 | 2021-04-30 | 吉林大学 | Power load prediction method, intelligent terminal and computer readable storage medium |
CN112734135B (en) * | 2021-01-26 | 2022-07-15 | 吉林大学 | Power load prediction method, intelligent terminal and computer readable storage medium |
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 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107145976A (en) | A kind of method for predicting user power utilization load | |
CN110634080B (en) | Abnormal electricity utilization detection method, device, equipment and computer readable storage medium | |
CN108491969A (en) | Spatial Load Forecasting model building method based on big data | |
CN110097297A (en) | A kind of various dimensions stealing situation Intellisense method, system, equipment and medium | |
CN111860979A (en) | Short-term load prediction method based on TCN and IPSO-LSSVM combined model | |
Enke et al. | A hybrid neuro-fuzzy model to forecast inflation | |
CN107169633A (en) | A kind of gas line network, gas storage peak regulating plan integrated evaluating method | |
CN107203774A (en) | The method and device that the belonging kinds of data are predicted | |
Chai et al. | A decomposition–integration model with dynamic fuzzy reconstruction for crude oil price prediction and the implications for sustainable development | |
CN105929690B (en) | A kind of Flexible Workshop Robust Scheduling method based on decomposition multi-objective Evolutionary Algorithm | |
Månsson et al. | A machine learning approach to fault detection in district heating substations | |
CN107679676A (en) | A kind of city based on numerical simulation is low to influence exploitation Optimal Configuration Method | |
Nijkamp et al. | Multi-objective programming models: New ways in regional decision-making | |
Lahdelma et al. | Stochastic multicriteria acceptability analysis (SMAA) | |
CN106447075B (en) | Trade power consumption needing forecasting method and system | |
Aghayi et al. | Efficiency measurement of DMUs with undesirable outputs under uncertainty based on the directional distance function: Application on bank industry | |
CN109472003A (en) | A kind of arithmetic of linearity regression applied to cost analysis | |
Wang et al. | A multicriteria decision making approach based on fuzzy theory and credibility mechanism for logistics center location selection | |
CN108241901A (en) | A kind of transformer Forewarn evaluation method and device based on prediction data | |
CN107067341A (en) | A kind of RBFNN electrical power distribution automatization system state operation evaluation methods based on multistage entropy weight | |
CN107464017A (en) | Based on the adaptive soft-sensor Forecasting Methodology with time difference Bayesian network | |
CN114595861A (en) | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method | |
Chen et al. | Does stricter Command-and-Control environmental regulation promote total factor productivity? Evidence from China’s industrial enterprises | |
CN109345021A (en) | A method of using LSTM modeling and forecasting labour demand increment | |
Mu et al. | Multistage iterative fully automatic partitioning in water distribution systems |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170908 |