CN108846528A - A kind of big industrial electrical user long term load forecasting method - Google Patents
A kind of big industrial electrical user long term load forecasting method Download PDFInfo
- Publication number
- CN108846528A CN108846528A CN201811097409.4A CN201811097409A CN108846528A CN 108846528 A CN108846528 A CN 108846528A CN 201811097409 A CN201811097409 A CN 201811097409A CN 108846528 A CN108846528 A CN 108846528A
- Authority
- CN
- China
- Prior art keywords
- data
- prediction
- model
- electricity
- load
- 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
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
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
A kind of big industrial electrical user long term load forecasting method, includes the following steps:Step S10 selects the object extraction predicted to go out electricity consumption data, and cleans to data;Step S20 carries out time series analysis to the electricity data cleaned;Step S30 determines the electricity consumption with days past using time series analysis;Step S40 rises the data input gradient after completion Feature Engineering in decision Tree algorithms model;Step S50 is added the business rule of some prediction particular times, carries out fine tuning to the last model trained and obtain final prediction result.The present invention solves strong dependence of the load prediction to meteorological data, and has very big promotion that can carry out long-term forecast on predicted time, has clear thinking, and economic value is high, is suitble to the advantages of promoting the use of.
Description
Technical field
The invention belongs to the different big industrial user's electricity consumptions of electrical power system analysis and computing field more particularly to electric system are pre-
Survey technology field.
Background technique
The load prediction of big industrial user at present mainly by being accomplished manually, relies on the electricity consumption that large user oneself declares, needle
Need to put into many human resources to different users to predict respectively, and the effect is unsatisfactory.Many enterprises in recent years
It all attempts to predict big industrial user's load using machine learning algorithm with universities and colleges, but big industrial user's load is related
Data characteristics it is excessively complicated and can not acquire mostly, such as consumer products inventory, big industrial user produce order and product city
Price fluctuation etc., meteorological data is that current only pair of loading effects are larger and collectable dynamic data, so these sides
Method has all used weather prognosis data when choosing dynamic prediction data characteristics, and weather prognosis can only reach in short-term one week
Compared with high-accuracy, can not following one month meteorology of Accurate Prediction, and the electricity charge at present are paid or generation schedule formulation be all with
The moon is unit, therefore this short-term load forecasting enables to reach higher level and also do not have practical value.Based on time sequence
The load prediction of column parser will guarantee that high-accuracy just needs to analyze object and has apparent periodicity and stability, and not
Stablizing residual error cannot be excessive, but the electricity consumption behavior of most big industrial users is comprehensive depending on product sales volume and product order etc.
Conjunction factor, there is no periodical and extremely unstable.Method proposes a kind of big industrial head of a household without using meteorological data
Prediction unit not only can be extended to the moon, and reach very high level in accuracy rate by phase load forecasting method.
Summary of the invention
In conclusion it is necessory to provide a kind of new intelligentized big industrial user's load forecasting method, it can be with the moon
Big industrial user's load is carried out for a long time and Accurate Prediction for unit, improves the stability of operation of power networks.
A kind of big industrial user's long-term load of the scrolling windows shape of the mouth as one speaks rising decision Tree algorithms and autocorrelation sequence based on gradient
Prediction technique mainly includes the following steps that:Step S10 selects the large user predicted to extract electricity consumption data, is
Guarantee data volume is sufficient and calculating process consumption is not too large, and 15 minutes original electricity datas are fused to daily electricity knot
Structure, and abnormal data, feelings of the data in normal distribution are filtered out using the normal distribution-test method in statistical probability distribution
The probability occurred under condition isWherein μ is data desired value, and σ is data variance value, according to every group
The different setting probability threshold values of data object, are then considered as abnormal data when the probability that the data occur is less than the threshold value, use
Newton interpolating method is replaced, and a small number of in order to prevent greatly exceptional value it is expected data and variance has an immense impact on, first will be big
It is replaced in 10 times of median or more of data using median;Step S20 carries out time series phase to the electricity data modified
Analysis is closed, electricity data sequence is become using calculus of finite differences by stationary sequence according to the precondition of time series analysis first, is led to
A normal time series is made of long-term trend T, seasonal move S, cyclical variations C, erratic variation I, due to time series point
The property of algorithm, T, C are analysed, I will lose a part of data of Data Frontend and end during the extraction process, so for data letter
Complete fusion is ceased into metadata, only extracts the history seasonal variations coefficient S of prediction object one of training pattern as after
Data characteristics;Step S30 carries out Feature Engineering to subjects history electricity consumption data, due to not using meteorological data, so base
In the theoretical thought of time series forecasting, using the difference of a few days electricity consumptions of history and electricity consumption in the past few days as new
Data characteristics, according to the object auto-correlation coefficient in time series analysis result, auto-correlation coefficient is described for the selection of number of days
The degree that this section of sequence interacts between two different periods, calculation formula areWherein μs
Expectation for data in the s period, σsVariance for data in the s period, μtExpectation for data in the t period, σtFor data
It is maximum for the different situations optimum selecting history electricity consumption features interplay coefficient of every month in the variance of t period
Number of days;Step S40 constructs gradient ascent algorithm model, and the data input gradient after completion Feature Engineering is risen decision tree and is calculated
In method model, and tuning is carried out to model parameter, by taking turns iteration, every wheel iteration generates a Weak Classifier, Mei Gefen more
Class device is trained on the basis of the residual error of last round of classifier, and the parameter maximum of the loss function of each Weak Classifier is quasi- right
The method is asked to beFm-1It (x) is current model, T (xi,θm) change for each round
The new Weak Classifier that generation generates finally trains the classifier of every wheel iteration and newly-generated Weak Classifier linear superposition negative
Lotus prediction model;Model is become rolling window prediction model by step S50, can be into the model theory that previous step trains
The load prediction of any number of days of row, but predict that number of days excessively still can be inaccurate, in order to accomplish accurate long-term forecast, it is added
Rolling forecast window, only the model before carries out load prediction in one day, and training dataset then is added in new prediction data,
Training data distance data farthest at present are removed into out training set, the length of training set is the length of rolling window, in this way
Not only training dataset length can be kept constant, the influence of data too remote to now forecast can also be rejected, then weighed
New training pattern, then prediction in next day is carried out, prediction data was this moon after one month by such rolling forecast altogether
Load prediction;According to the unique characteristics of different user the business rule of some prediction particular times is added, to upper in step S60
One model for training carries out fine tuning, such as the fixation annual test time of large user, and employee has a holiday arrangement etc., finally obtains
Month load prediction results.
The present invention is based on gradient rising decision Tree algorithms and for the scrolling windows shape of the mouth as one speaks electric power enterprise of autocorrelation sequence, originally
The characteristics of invention does not need external data feature, only relies on itself curve data carries out aperiodicity instability data pre-
It surveys.Of the invention is mainly characterized by:
1. not needing outside weather data, the historical load data of large user is relied only on;
2. determining that data characteristics selects length by the auto-correlation coefficient of autocorrelation sequence;
3. predicting daily load data one by one using rolling window, it is finally summed into a moon load prediction.
The invention has the advantages that being constructed in the case where being limited departing from meteorological data using auto-correlation coefficient
New data characteristics, and the very high long term load forecasting of accuracy rate is realized by rolling window, broken can only carry out it is short
The data of phase load prediction limit, and the long term load forecasting method has general applicability.The present invention has clear thinking, leads to
Preferable with property, economic value is high, is suitble to the advantages of promoting the use of.
Detailed description of the invention
Fig. 1 is that a kind of scrolling windows shape of the mouth as one speaks for rising decision Tree algorithms and autocorrelation sequence based on gradient provided by the invention is big
Industrial user's long term load forecasting method flow diagram;
Fig. 2 is the abnormal data cleaning process figure provided by the invention based on probability statistics normal distribution;
Fig. 3 is that gradient provided by the invention rises decision Tree algorithms code flow diagram;
Fig. 4 is rolling window model training flow chart provided by the invention;
Fig. 5 is prediction result of the moon load forecasting method provided by the invention to certain big industrial user's the first half in 2018
Figure.
Specific embodiment
It is further stated in detail below according to Figure of description and in conjunction with specific embodiments to technical solution of the present invention.
Referring to Fig.1, Fig. 1 is a kind of rolling for rising decision Tree algorithms and autocorrelation sequence based on gradient provided by the invention
The big industrial user's long term load forecasting method flow diagram of Window-type, mainly includes the following steps:
Step S10 selects the object extraction predicted to go out electricity consumption data, in order to guarantee data volume abundance and
Calculating process consumption is not too large, 15 minutes original electricity datas is fused to daily electricity structure, and use statistical probability point
Normal distribution-test method in cloth identifies abnormal data.
The probability that one data occurs in the case where normal distribution is
Wherein μ is data desired value, and σ is data variance value, and probability threshold values are arranged according to every group of the different of data object, when
The probability that the data occur then is considered as abnormal data when being less than the threshold value, referring to fig. 2, is replaced using equidistant Newton interpolating method
It changes, the calculation method of Newton interpolating method is:
Before calculating data expectation and variance, a small number of greatly exceptional values it is expected data in order to prevent and variance generates
Tremendous influence first will be greater than 10 times of median or more of data and be replaced using median;
Step S20 carries out time series correlation analysis to the electricity data modified, first according to time series analysis
Electricity data sequence is become stationary sequence using calculus of finite differences by precondition, and a usual time series is by long-term trend T, season
S is changed, cyclical variations C, erratic variation I composition can usually show as following form:
yt=Tt+St+Ct+It
T is time index, and due to the property of time sequence analysis algorithm, T, C, I will lose data during the extraction process
A part of data in front end and end, so only extracting the history of prediction object in order to which data information complete fusion is into metadata
One data characteristics of seasonal variations coefficient S training pattern as after;
Step S30 carries out Feature Engineering to subjects history electricity consumption data, due to not using meteorological data, so only
Once dynamic data only remaining historical load data, it is new to create that historical load data is screened by time series analysis
Data characteristics, the theoretical thought based on time series forecasting, electricity consumption by a few days electricity consumptions of history and in the past few days
Difference is as new data characteristics, and the selection of number of days is according to the object auto-correlation coefficient in time series analysis result, auto-correlation
Coefficient describes the degree that this section of sequence interacts between two different periods, that is, see the data of present period by
The influence of those periods of recent history is maximum, and correlation calculations formula is:
Wherein μsExpectation for data in the s period, σsVariance for data in the s period, μtIt is data in the t period
Expectation, σtVariance for data in the t period, it is mutual for the different situations optimum selecting history electricity consumption measure feature of each period
Mutually influence the maximum number of days of coefficient, the load datas of these days do difference processing before predicting target date, then together as
Data new feature;
Step S40, building gradient rise decision Tree algorithms model, and the data after completion Feature Engineering are inputted algorithm model
In, and tuning is carried out to model parameter, model generates a Weak Classifier, each classification by taking turns iteration, every wheel iteration more
Device is trained on the basis of the residual error that last round of iteration generates.
Gradient, which promotes decision tree, can be expressed as following form:FmIt (x) is current model, T (xi,θm) change for each round
The new Weak Classifier that generation generates, that is, the m decision tree, θmIndicate the parameter of decision tree;Strong classifier Fm(x) by multiple weak
Classifier linear, additive forms.
Iteration m times strong classifier be:Fm(x)=Fm-1(x)+T(x,θm)
Then loss function is obtained:L(Fm(x), y)=L (Fm-1(x)+T(x,θm),y)
The purpose of iteration is to find out parameter θmConstruct T (xi,θm), so that epicycle loses L (Fm(x), y) it is minimum, by experiment
It is best for selecting huber loss function.
Huber loss function:
The parameter maximum of the loss function of each Weak Classifier is quasi- so to ask the method to be:
Fm-1It (x) is current model, T (xi,θm) be each round grey iterative generation new Weak Classifier, finally train available
In the prediction model of the following month electricity consumption.This steps flow chart please refers to Fig. 3;
Step S50, rolling forecast window, which is added, by model becomes rolling window prediction model, carries out one with model before
Then training dataset is added in new prediction data by it load prediction, training data distance data farthest at present are removed
The length of training set out, training set is the length of rolling window, then uses new training dataset re -training model, then
It carrying out prediction in next day, repeats this step, training set window is just with the change rolls forward of training set, and rolling forecast one
The as load prediction of this month altogether by prediction data after month.This steps flow chart please refers to Fig. 4;
According to the unique characteristics of different user the business rule of some prediction particular times is added, to upper one in step S60
Model that portion trains carries out fine tuning, such as the fixation annual test time of large user, and employee has a holiday arrangement etc., finally obtains the moon
Load prediction results.
Not the characteristics of present invention does not need external data feature, only relies on itself curve data, to aperiodicity instability
Data are predicted.Of the invention is mainly characterized by:
1. not needing outside weather data, the historical load data of large user is relied only on;
2. determining that data characteristics selects length by the auto-correlation coefficient of autocorrelation sequence;
3. predicting daily load data one by one using rolling window, it is finally summed into a moon load prediction.
The invention has the advantages that being constructed in the case where being limited departing from meteorological data using auto-correlation coefficient
New data characteristics, and the very high long term load forecasting of accuracy rate is realized by rolling window, broken can only carry out it is short
The data of phase load prediction limit, and the long term load forecasting method has general applicability.The present invention has clear thinking, leads to
Preferable with property, economic value is high, is suitble to the advantages of promoting the use of.
Embodiment one
A kind of big industry of the scrolling windows shape of the mouth as one speaks rising decision Tree algorithms and autocorrelation sequence based on gradient of the present invention
User's long term load forecasting method, it is negative that the load data based on certain big industrial user under certain power grid carries out 2018 1 moons to June
Lotus prediction, and compared with practical electricity consumption, specific step is as follows:
(1) certain big industrial user is taken in metering automation system since 2010 to the original minus in June, 2018
15 minute datas are fused to the load data as unit of day for the 15 minutes frequency acquisitions in interval by lotus data, due to
Most data are all 0 before in October, 2013, it should be that acquisition terminal is set up not yet, therefore by data from 2013
October is initially as valid data.
(2) exceptional value is examined with normal distribution-test method, respectively brings data intoThreshold value is set
It is 0.0001, and when currently, just the probability that occurs then judges the data to be different to this data if it is less than 1/10000th too in distribution
Constant value, and substituted after being calculated with Newton interpolating method according to front and back data.
(3) time series point is carried out to modified load data by the statsmodel frame in python language library
Analysis, and tables of data is added as new data characteristics in the seasonal variations ingredient S extracted.
(4) according to the data before in the January, 2018 to be predicted, pass throughCalculated load sequence
The auto-correlation coefficient of column lag 0 rank to 10 ranks, the related coefficient of 10 ranks is maximum, is 0.673, so by before on January 1st, 2018
10 days load datas tables of data is added as new data characteristics, and do difference processing.
(5) building gradient rises decision-tree model, and by August in handle well 2013 on December 31st, 1 day 1
Data carry out parameter regulation and model training as training data input model.
(6) it is predicted using trained model, is predicted since on January 1st, 2018, rolling window is added, and
It former training dataset is added in prediction result forms new training dataset, then predict the load on January 2nd, 2018, with this
Analogize, after prediction 31 times, this 31 values are added to the moon predicted load for just obtaining in January, 2018.
(7) and so on obtain 2018 1 predicted loads to June, according to analysis, this user can have a holiday or vacation during the Spring Festival to stop
It produces, respective rule is added, and practical electricity consumption carries out error calculation, error calculation formula is:
Final result is referring to Fig. 5.
From above-mentioned calculating process it is found that a kind of scrolling windows shape of the mouth as one speaks for rising decision Tree algorithms and autocorrelation sequence based on gradient
Big industrial user's long term load forecasting method, it is pre- accurately can to provide the long-term moon load of big industrial user for power supply bureau
It surveys, Average Accuracy provides basis 90% or more, for power system stability operation.1. in addition, those skilled in the art may be used also
Make other variations in spirit of that invention, these made variations of spirit according to the present invention, should all be included in institute of the present invention certainly
In claimed range.
Claims (1)
1. a kind of big industrial electrical user long term load forecasting method, which is characterized in that include the following steps:
Step S10 selects the large user that is predicted to extract electricity consumption data, using by 15 minutes original electricity numbers
According to being fused to daily electricity structure, and abnormal data, a number are filtered out using the normal distribution-test method in statistical probability distribution
It is according to the probability occurred in the case where normal distributionWherein μ is data desired value, and σ is data
Variance yields is then regarded according to the different setting probability threshold values of every group of data object when the probability that the data occur is less than the threshold value
It for abnormal data, is replaced using Newton interpolating method, and the data that first will be greater than 10 times of median or more are replaced using median
It changes;
Step S20 carries out time series correlation analysis to the electricity data modified, first according to the premise of time series analysis
Electricity data sequence is become stationary sequence using calculus of finite differences by condition, and a time series is followed by long-term trend T, seasonal move S
Ring changes C, and erratic variation I composition only extracts history seasonal variations coefficient S training pattern as after of prediction object
One data characteristics;
Step S30 carries out Feature Engineering to subjects history electricity consumption data, by a few days electricity consumptions of history and in the past few days
The difference of electricity consumption is as new data characteristics, and the selection of number of days is according to the object auto-correlation system in time series analysis result
Number, auto-correlation coefficient describe the degree that this section of sequence interacts between two different periods, and calculation formula isWherein μsExpectation for data in the s period, σsVariance for data in the s period, μtFor
Expectation of the data in the t period, σtVariance for data in the t period, for the different situations optimum selecting history of every month
The maximum number of days of electricity consumption features interplay coefficient;
Step S40 constructs gradient ascent algorithm model, and the data input gradient after completion Feature Engineering is risen decision Tree algorithms
In model, and tuning is carried out to model parameter, by taking turns iteration, every wheel iteration generates a Weak Classifier, each classification more
Device is trained on the basis of the residual error of last round of classifier, and the parameter maximum of the loss function of each Weak Classifier is quasi- so to be asked
Method isFm-1It (x) is current model, T (xi,θm) it is each round iteration
The classifier of every wheel iteration and newly-generated Weak Classifier linear superposition are finally trained load by the new Weak Classifier generated
Prediction model;
Model is become rolling window prediction model, carries out any number of days in the model theory that previous step trains by step S50
Rolling forecast window is added in load prediction, and only the model before carries out load prediction in one day, then adds new prediction data
Enter training dataset, training data distance data farthest at present is removed into out training set, the length of training set is scrolling windows
Mouthful length, then re -training model, then carry out prediction in next day, such rolling forecast closed prediction data after one month
Get up as the load prediction of this month;
Step S60 is added the business rule of prediction particular time, trains to last according to the unique characteristics of different user
Model carry out fine tuning, finally obtain a moon load prediction results.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811097409.4A CN108846528B (en) | 2018-09-19 | 2018-09-19 | Long-term load prediction method for large-scale industrial power consumer |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811097409.4A CN108846528B (en) | 2018-09-19 | 2018-09-19 | Long-term load prediction method for large-scale industrial power consumer |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108846528A true CN108846528A (en) | 2018-11-20 |
CN108846528B CN108846528B (en) | 2021-07-09 |
Family
ID=64187947
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811097409.4A Active CN108846528B (en) | 2018-09-19 | 2018-09-19 | Long-term load prediction method for large-scale industrial power consumer |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108846528B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378541A (en) * | 2019-08-08 | 2019-10-25 | 上海交通大学 | A kind of short-term multistep forecasting method of wind power and device |
CN111553516A (en) * | 2020-04-17 | 2020-08-18 | 贵州电网有限责任公司 | Short-term electric quantity load accurate prediction method |
CN111582551A (en) * | 2020-04-15 | 2020-08-25 | 中南大学 | Method and system for predicting short-term wind speed of wind power plant and electronic equipment |
CN111598302A (en) * | 2020-04-19 | 2020-08-28 | 华电郑州机械设计研究院有限公司 | Thermal power plant short-term industrial heat load prediction method based on AP-TS-SVR model |
CN111950805A (en) * | 2020-08-25 | 2020-11-17 | 润联软件系统(深圳)有限公司 | Medium-and-long-term power load prediction method and device, computer equipment and storage medium |
CN112598181A (en) * | 2020-12-23 | 2021-04-02 | 国网河北省电力有限公司经济技术研究院 | Load prediction method, device, equipment and storage medium |
CN112668765A (en) * | 2020-12-23 | 2021-04-16 | 韶关东阳光科技研发有限公司 | Method and device for load prediction and energy management of power consumers of aluminum processing enterprises |
CN114399098A (en) * | 2021-12-30 | 2022-04-26 | 昆明能讯科技有限责任公司 | High-applicability and high-precision industrial power utilization user classification peak-shifting power utilization method |
CN114638425A (en) * | 2022-03-24 | 2022-06-17 | 国网江苏省电力有限公司南通供电分公司 | Large-user monthly power consumption prediction method and system based on historical data |
CN115759236A (en) * | 2022-12-30 | 2023-03-07 | 北京德风新征程科技有限公司 | Model training method, information sending method, device, equipment and medium |
CN117592770A (en) * | 2024-01-19 | 2024-02-23 | 成都秦川物联网科技股份有限公司 | Intelligent gas industry and commercial tenant flowmeter regulation and control method, internet of things system and medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140281645A1 (en) * | 2013-03-15 | 2014-09-18 | Accenture Global Services Limited | Enchanced grid reliability through predictive analysis and dynamic action for stable power distribution |
CN104598986A (en) * | 2014-12-12 | 2015-05-06 | 国家电网公司 | Big data based power load prediction method |
CN107563539A (en) * | 2017-07-24 | 2018-01-09 | 佛山市顺德区中山大学研究院 | Short-term and long-medium term power load forecasting method based on machine learning model |
CN108510113A (en) * | 2018-03-21 | 2018-09-07 | 中南大学 | A kind of application of XGBoost in short-term load forecasting |
-
2018
- 2018-09-19 CN CN201811097409.4A patent/CN108846528B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140281645A1 (en) * | 2013-03-15 | 2014-09-18 | Accenture Global Services Limited | Enchanced grid reliability through predictive analysis and dynamic action for stable power distribution |
CN104598986A (en) * | 2014-12-12 | 2015-05-06 | 国家电网公司 | Big data based power load prediction method |
CN107563539A (en) * | 2017-07-24 | 2018-01-09 | 佛山市顺德区中山大学研究院 | Short-term and long-medium term power load forecasting method based on machine learning model |
CN108510113A (en) * | 2018-03-21 | 2018-09-07 | 中南大学 | A kind of application of XGBoost in short-term load forecasting |
Non-Patent Citations (1)
Title |
---|
毕云帆等: "基于梯度提升决策树的电力短期负荷预测模型", 《青岛大学学报》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378541A (en) * | 2019-08-08 | 2019-10-25 | 上海交通大学 | A kind of short-term multistep forecasting method of wind power and device |
CN111582551A (en) * | 2020-04-15 | 2020-08-25 | 中南大学 | Method and system for predicting short-term wind speed of wind power plant and electronic equipment |
CN111582551B (en) * | 2020-04-15 | 2023-12-08 | 中南大学 | Wind power plant short-term wind speed prediction method and system and electronic equipment |
CN111553516A (en) * | 2020-04-17 | 2020-08-18 | 贵州电网有限责任公司 | Short-term electric quantity load accurate prediction method |
CN111598302B (en) * | 2020-04-19 | 2023-09-05 | 华电郑州机械设计研究院有限公司 | AP-TS-SVR model-based thermal power plant short-term industrial heat load prediction method |
CN111598302A (en) * | 2020-04-19 | 2020-08-28 | 华电郑州机械设计研究院有限公司 | Thermal power plant short-term industrial heat load prediction method based on AP-TS-SVR model |
CN111950805A (en) * | 2020-08-25 | 2020-11-17 | 润联软件系统(深圳)有限公司 | Medium-and-long-term power load prediction method and device, computer equipment and storage medium |
CN112668765A (en) * | 2020-12-23 | 2021-04-16 | 韶关东阳光科技研发有限公司 | Method and device for load prediction and energy management of power consumers of aluminum processing enterprises |
CN112598181A (en) * | 2020-12-23 | 2021-04-02 | 国网河北省电力有限公司经济技术研究院 | Load prediction method, device, equipment and storage medium |
CN114399098A (en) * | 2021-12-30 | 2022-04-26 | 昆明能讯科技有限责任公司 | High-applicability and high-precision industrial power utilization user classification peak-shifting power utilization method |
CN114638425A (en) * | 2022-03-24 | 2022-06-17 | 国网江苏省电力有限公司南通供电分公司 | Large-user monthly power consumption prediction method and system based on historical data |
CN115759236A (en) * | 2022-12-30 | 2023-03-07 | 北京德风新征程科技有限公司 | Model training method, information sending method, device, equipment and medium |
CN115759236B (en) * | 2022-12-30 | 2024-01-12 | 北京德风新征程科技股份有限公司 | Model training method, information sending method, device, equipment and medium |
CN117592770A (en) * | 2024-01-19 | 2024-02-23 | 成都秦川物联网科技股份有限公司 | Intelligent gas industry and commercial tenant flowmeter regulation and control method, internet of things system and medium |
CN117592770B (en) * | 2024-01-19 | 2024-04-12 | 成都秦川物联网科技股份有限公司 | Intelligent gas industry and commercial tenant flowmeter regulation and control method, internet of things system and medium |
Also Published As
Publication number | Publication date |
---|---|
CN108846528B (en) | 2021-07-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108846528A (en) | A kind of big industrial electrical user long term load forecasting method | |
Koopman et al. | Predicting time-varying parameters with parameter-driven and observation-driven models | |
Chen et al. | A hybrid application algorithm based on the support vector machine and artificial intelligence: An example of electric load forecasting | |
Pirbazari et al. | Evaluating feature selection methods for short-term load forecasting | |
US20170323332A1 (en) | Methods and apparatus to incorporate saturation effects into marketing mix models | |
Koop et al. | Forecasting in large macroeconomic panels using Bayesian model averaging | |
CN104537433A (en) | Sold electricity quantity prediction method based on inventory capacities and business expansion characteristics | |
Jochmann | Modeling US inflation dynamics: A Bayesian nonparametric approach | |
Wen et al. | Daily sales forecasting for grapes by support vector machine | |
Gupta et al. | Implementation of demand forecasting–A comparative approach | |
CN110009427B (en) | Intelligent electric power sale amount prediction method based on deep circulation neural network | |
Aman et al. | Learning to reduce: A reduced electricity consumption prediction ensemble | |
Nachane | Dynamic stochastic general equilibrium (DSGE) modelling in practice: identification, estimation and evaluation | |
Afrah et al. | The Utilization of Deep Learning in Forecasting The Inflation Rate of Education Costs in Malang | |
CN113421148B (en) | Commodity data processing method, commodity data processing device, electronic equipment and computer storage medium | |
Kedrowski et al. | Short-term seasonal energy forecasting | |
CN112488421B (en) | Tracking and predicting method and device for accounts receivable of electric charge | |
Heylman et al. | Forecasting energy trends and peak usage at the University of Virginia | |
CN112465260A (en) | Student teaching management system based on campus data | |
Donaldson et al. | Improving forecast accuracy using a synthetic weather station: An incremental approach and BFCom2018 lessons learned | |
Khumaidi et al. | Forecasting of Sales Based on Long Short Term Memory Algorithm with Hyperparameter | |
Wang et al. | EMD based value at risk estimate algorithm for electricity markets | |
Peters et al. | Choosing Technologies: Benefits of Developing Fourth Industrial Revolution Technologies | |
Yakubu et al. | SEASONAL AND CYCLIC FORECASTING FOR HYDRO-ELECTRIC POWER GENERATING STATION | |
Wei et al. | Dynamic Bayesian predictive model for box office forecasting |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220214 Address after: 650034 No. 73, Tuodong Road, Kunming, Yunnan Patentee after: INFORMATION CENTER OF YUNNAN GRID POWER Co.,Ltd. Address before: 650217 No. 105 Yunda West Road, Kunming Economic and Technological Development Zone, Yunnan Province Patentee before: INFORMATION CENTER OF YUNNAN GRID POWER Co.,Ltd. Patentee before: Kunming Nengxun Technology Co., Ltd |