CN110245798A - A kind of monthly electricity demand forecasting method and system of office building electric system - Google Patents
A kind of monthly electricity demand forecasting method and system of office building electric system Download PDFInfo
- Publication number
- CN110245798A CN110245798A CN201910527955.5A CN201910527955A CN110245798A CN 110245798 A CN110245798 A CN 110245798A CN 201910527955 A CN201910527955 A CN 201910527955A CN 110245798 A CN110245798 A CN 110245798A
- Authority
- CN
- China
- Prior art keywords
- data
- electricity consumption
- temperature
- model
- office building
- 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
- 230000005611 electricity Effects 0.000 title claims abstract description 125
- 238000013277 forecasting method Methods 0.000 title claims abstract description 14
- 230000002123 temporal effect Effects 0.000 claims abstract description 19
- 238000004140 cleaning Methods 0.000 claims abstract description 14
- 238000005457 optimization Methods 0.000 claims abstract description 10
- 238000013500 data storage Methods 0.000 claims abstract description 8
- 241001269238 Data Species 0.000 claims abstract description 6
- 210000000352 storage cell Anatomy 0.000 claims abstract description 4
- 238000010276 construction Methods 0.000 claims abstract description 3
- 238000000034 method Methods 0.000 claims description 26
- 238000010438 heat treatment Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 12
- 238000003066 decision tree Methods 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000002790 cross-validation Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 241000208340 Araliaceae Species 0.000 claims description 4
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 4
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 4
- 235000008434 ginseng Nutrition 0.000 claims description 4
- 238000009825 accumulation Methods 0.000 claims description 3
- 238000012217 deletion Methods 0.000 claims description 3
- 230000037430 deletion Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000008676 import Effects 0.000 claims description 2
- 210000004027 cell Anatomy 0.000 claims 1
- 238000012549 training Methods 0.000 abstract description 3
- 230000000875 corresponding effect Effects 0.000 description 12
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000001932 seasonal effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- 208000025174 PANDAS Diseases 0.000 description 1
- 208000021155 Paediatric autoimmune neuropsychiatric disorders associated with streptococcal infection Diseases 0.000 description 1
- 240000000220 Panda oleosa Species 0.000 description 1
- 235000016496 Panda oleosa Nutrition 0.000 description 1
- 238000004378 air conditioning Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003471 mutagenic agent Substances 0.000 description 1
- 231100000707 mutagenic chemical Toxicity 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000000714 time series forecasting Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This patent discloses a kind of monthly electricity demand forecasting method and system of office building electric system, include the following steps: 1) input prediction date creation time sequence；2) predicted temperature data are obtained and clean predicted temperature data；3) electricity consumption GBDT nonlinear model is imported, calculates every daily power consumption and monthly electricity consumption total amount, and export prediction result；It further include building using discrete data building electricity consumption GBDT nonlinear model, construction step includes: 1) electric power initial data acquisition, and history singly amount data storage cell is saved in after cleaning；2) raw temperature data obtains, and historical temperature data storage unit is saved in after cleaning；3) temperature information determines；4) festivals or holidays determine；5) temporal information is extracted；6) data merge；7) model foundation and optimization；8) model saves.This patent is using the training of the discrete datas such as history electricity consumption information, temperature information, holiday information, temporal information and constructs electricity consumption GBDT nonlinear model, by to this electricity consumption of office building not quietly the prediction of the every daily power consumption of state electric system to obtaining monthly electricity consumption Prediction of Total as a result, larger improve prediction precision.
Description
Technical field
The present invention relates to the electricity usages unstable states such as the monthly electricity demand forecasting technology of electric system more particularly to office building
Determine the monthly electricity demand forecasting technology of electric system in situation.
Background technique
The monthly electricity demand forecasting of electric power belongs to time series forecasting type.Time series is one group and depends on the random of time
Variable, has dependence between this group of stochastic variable, and correlation properties show the continuity of prediction object development.To wherein institute
The autocorrelation performance contained with mathematical model be depicted come, so that it may utilize time series past value and now value predict future
Value.
In the prior art, it is pre- that predicting means mostly uses greatly the recurrence means of multivariable to carry out to the monthly electricity demand forecasting of electric power
It surveys, by carrying out the modeling of multivariable with living standard to local resident's amount, so that the residual sum of prediction is minimum and obtains one
A model.Or approached using SVM algorithm progress is certain, by the kernel function of SVM, the impact factor of prediction is projected not
On same dimension, so as to predict the monthly electricity consumption system more unstable compared with multivariate regression.Also have when passing through
Between sequence autoregressive moving-average model (ARMA) model, since arma modeling only focuses on the shadow of time series bring itself
It rings, has ignored many interference factors and limit the validity in practical monthly electricity demand forecasting.
These methods are higher to historical data accuracy requirement, and bad data is larger to predicted impact, more stable in power consumption
Regional prediction effect is preferable, when the exogenous influence factor such as Changes in weather, festivals or holidays cause electricity consumption not quietly in the case where, deposit
In biggish prediction error, and predict that step number is longer, precision of prediction is poorer.
Office building electric system is a complicated real-time dynamic system, all kinds of novel mechanical equipment, high-power electric appliance
It uses, so that stability, safety substantially reduce in the process of running for office building electric system.Meanwhile the electricity consumption of office building is not
It is same as ordinarily resident's electricity consumption, it is influenced by multiple factors such as season, temperature, festivals or holidays and external environments, such as in season alternation
When, it may appear that there are heating demands in larger fluctuation, winter, and opposite summer needs cold air to supply, and require to expend more electricity,
In addition many Mutagens influence whether the prediction in some month, such as switch the specific time of air-conditioning, and the increase and decrease etc. of personnel all can
Influence electricity consumption.Therefore, office building electric system, it is a variety of by season, festivals or holidays, working day, office building personnel amount etc.
Factor influences, and electricity consumption fluctuation is larger, and the monthly electricity consumption of electric system is unstable, uses the monthly electricity demand forecasting system of conventional electric power
System, obtained prediction result and practical electricity consumption deviation is larger, can not achieve and precisely estimates.
Summary of the invention
(1) technical problems to be solved
Effectively and accurately predict that the monthly electricity consumption of office building electric power is a highly important thing.The present invention mainly solves
Certainly the problem of is to cause office building electric system power consumption larger fluctuation situation in many factors such as season, festivals or holidays, times
Under, how to predict the monthly electricity consumption of office building electric system, improves prediction precision.
(2) technical solution
To solve the above-mentioned problems, the present invention by extract influence the monthly electricity consumption of office building electric system it is main because
Element establishes GBDT nonlinear model according to the every daily power consumption data of history, thus predict the daily electricity consumption of office building, final
To the monthly electricity consumption total amount of office building.I.e. the present invention provides a kind of method of monthly electricity demand forecasting of office building electric system, comprising:
S1, input inquiry time create predicted time sequence；
S2 obtains predicted temperature data, cleans predicted temperature data；
S3 calls electricity consumption GBDT nonlinear model, calculates every daily power consumption and monthly electricity consumption total amount, and exports prediction knot
Fruit；
It further include constructing electricity consumption GBDT nonlinear model using discrete data, step includes:
1) acquisition of electric power initial data and cleaning step.The original electricity consumption data of office building are obtained, and to initial data
It is cleaned, the parameter and format needed according to system is saved in history electricity consumption data storage cell；
2) raw temperature data acquisition and cleaning step.Raw temperature data is crawled, 3 temperature ginseng that the system that obtains needs
Number: minimum temperature, to maximum temperature and mean temperature is stored in historical temperature storage unit；
3) temperature information determines；
4) festivals or holidays determine；
5) temporal information is extracted；
6) data merge, and the discrete data obtained after processing is merged according to time series, is stored in csv file
In；
7) model foundation and optimization, using electricity consumption as output valve, temperature information, holiday information, temporal information etc. are made
For input value, GBDT model will be calculated using the method for web search Optimal Parameters using 10 folding cross validations as verification mode
Method optimization, obtains final electricity consumption GBDT nonlinear model；
8) model saves；
Preferably, the original electric data of electric power include the every 5 minutes power consumption datas of distinct device, clean electric power initial data
Include: to be handled using big data correlation technique, initial data is subjected to duplicate checking, suppressing exception value fills up missing values, extracts
The power consumption total quantity index of high-grade equipment extracts parameter and corresponding format that this system needs, and saves by the period of day.
Preferably, raw temperature data is obtained, comprising: crawl using crawler to temperature data, configure by obtaining
Different cities code in file is come the temperature data in city where obtaining office building；Clean raw temperature data, comprising: to scarce
Lose data the deletion filled up with repeated data, extract this system need 3 temperature parameters: minimum temperature, maximum temperature with
And mean temperature, all temperature units are degree Celsius, and are stored in temperature information storage unit.
Preferably, raw temperature data is obtained, further includes: while determine to obtain by the control of function variable is pre-
Survey the predicted temperature or historical temperature on date.
Preferably, temperature information determines, step includes:
1) judge whether that selection judges whether out heating using temperature；
2) if it is, continuing to determine whether to reach out heating standard；
If 3) reach out heating standard, heating index is assigned a value of 1, if not reaching out heating standard, heating index
It is assigned a value of 0, is terminated；
If 4) judge whether heating without using temperature, judge whether to reach spring, if it is, season index assignment
It is 0, terminates；
If 5) do not reach spring, judge whether to reach summer, be, then season index is assigned a value of 1, terminates；
If 6) do not reach summer, judge whether to reach autumn, be, then season index is assigned a value of 2, terminates；
If 7) do not reach autumn, judge whether to reach winter, be, then season index is assigned a value of 3, terminates；If do not reached
To winter, then " not meeting default standard, request artificial setting " is prompted.
Wherein, festivals or holidays determine, the section on corresponding date is judged including using two assignment of holiday attribute and Spring Festival attribute
Holiday attribute:
If weekend, then holiday attribute assignment is 1, and if three days or more vacations, holiday attribute assignment was 2, if
Seven days or more vacations, holiday attribute assignment are 3, and if during the New Year, then holiday attribute assignment is 4, if working day, then
The holiday attribute assignment is 0；
For example holiday in the Spring Festival, then Spring Festival attribute assignment is 1, such as non-holiday in the Spring Festival, then Spring Festival attribute assignment is 0.
Preferably, festivals or holidays determine, two assignment of holiday attribute and Spring Festival attribute can not correspond, holiday attribute tool
Have a holiday or vacation legal festivals and holidays of basis of time current year of body determine, and Spring Festival attribute has fractional unit to start to have a holiday or vacation i.e. during the Spring Festival
Can start setting up is 1, which is arranged by a library.
Preferably, temporal information is extracted, and belongs to this year including using the library datetime of Python to extract this day
Which week and this week in which day, and save it in temporal information storage element.
Preferably, input inquiry time creation time sequence includes that system generation institute's inputting date is following trimestral daily
Predicted time sequence.
Preferably, electricity consumption GBDT nonlinear model is a kind of Boosting model based on tree, using serial side
Formula, each iteration selects a weak learner, while establishing a new decision tree on the gradient direction for reducing residual error, uses
The result of each decision tree is obtained final prediction result by weighted accumulation, the model mathematical expression by linear adder model
Formula are as follows:
(1) in formula, x is to input total sample, ht(x；wt) it is every post-class processing, wtFor the ginseng of each post-class processing
Number, αtIt is the weight of each tree, T is the number of decision tree；
In the learning process of each round, a weak learner h is generatedt(x；wt), the loss function of weak learner are as follows:
(2) in formula, Ft-1(xi；wt) it is current model, GBDT determines Weak Classifier by minimizing loss function value
Parameter, loss function select quadratic loss function, corresponding mathematical formulae are as follows:
(3) in formula, yiFor true value, h (xi) it is model estimate value, corresponding difference, that is, residual error.
Preferably, electricity consumption GBDT nonlinear model, prediction evaluation index use mean absolute percentage error (MAPE),
Expression formula are as follows:
(4) X in formulaiFor practical daily power consumption, YiTo predict daily power consumption；The total electricity consumption of prediction in last statistical forecast month
The accuracy evaluation index the most final of amount and practical total electricity consumption, corresponding mathematic(al) representation are as follows:
Wherein, YSFor this month practical total electricity consumption in prediction month, PSFor the of that month prediction total electricity consumption in prediction month.
Preferably, input inquiry time creation time sequence includes that system generation institute's inputting date is following trimestral with day
For the time series in period.
To solve the above problems, the present invention also provides a kind of monthly electricity demand forecasting systems of office building electric system, comprising:
Time series creating unit, input prediction date simultaneously generate predicted time sequence；
Predicted temperature acquiring unit obtains the temperature of forecast date, and cleaning temperature data；
Electricity consumption GBDT nonlinear model unit is called, model is imported, calculates daily electricity consumption data and monthly electricity consumption total amount,
Export prediction result.
It further include electricity consumption GBDT nonlinear model construction unit, step includes:
1) acquisition of electric power initial data and cleaning unit, store history daily power consumption data and correlated characteristic；
2) raw temperature data acquisition and cleaning unit, store historical temperature data and correlated characteristic；
3) temperature information determination module；
4) festivals or holidays determination module；
5) temporal information extraction module；
6) discrete data obtained after processing is merged according to time series, is stored in csv by data combiners block
In file；
7) model foundation and optimization module, using electricity consumption as output valve, temperature information, holiday information, temporal information
Deng be used as input value, to GBDT model use web search Optimal Parameters method, using 10 folding cross validations as verification mode,
By algorithm optimization, final electricity consumption GBDT nonlinear model is obtained；
8) model preserving module.
(3) beneficial effect
This patent is used using the building of the discrete datas such as history electricity consumption information, temperature information, holiday information, temporal information
Electricity GBDT nonlinear model, to this electricity consumption of office building, quietly the monthly electricity consumption of state electric system does not implement prediction, leads to
It crosses and predicts every daily power consumption to obtain monthly electricity consumption, shorten predetermined period, improve the precision of prediction.
This prediction technique analyzes the various factors for influencing electricity consumption, and monthly electricity consumption is predicted as unit of day,
To implement precisely prediction to the monthly electricity consumption of office building electric system, be more advantageous to the manager of office building to building electric energy into
The effective control of row, avoid the occurrence of power consumption with the situation estimating electricity and differ too big, rationally estimate purchase, can reach energy saving
The effect of emission reduction can also be such that the electric energy expense of office building manager and administration fee substantially reduces, rationally reduce financial expenditures, together
When be also beneficial to the sale of electricity arrangement of power department or sale of electricity company, have good Social benefit and economic benefit.
Detailed description of the invention
Fig. 1 is GBDT modular concept figure；
Fig. 2 this system electricity consumption GBDT Building Nonlinear Model procedure chart；
Fig. 3 daily power consumption distribution map
Fig. 4 is this system temperature information decision flowchart；
Fig. 5 this system model calculation and the monthly electricity demand forecasting procedure chart of electric power；
Fig. 6 is the monthly electricity demand forecasting result figure compared with truthful data of certain office building electric system.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.
Fig. 1 shows GBDT modular concept.
Experience have shown that one group of weak learner better than random guess can be promoted by integrated as a strong learner.
In machine learning field, according to this thought, the Boosting learning algorithm that has been born more classical.Present patent application is selected
Algorithm the most reliable and the most stable in Boosting: Gradient Boosting Decision Tree (GBDT) nonlinear model,
As the core algorithm of prediction electricity consumption, carried out by collecting history electric power data and historical temperature data, and to these data
Cleaning and creation time sequence data merge data, extracted valid data, import GBDT nonlinear model, pass through model
Operation exports the monthly electricity consumption data of electric power of prediction.GBDT is based on the Boosting thought in integrated study, using serial
Mode, each iteration selects a weak learner (generally decision stub), while being on the gradient direction for reducing residual error
A new decision tree is established, principle is as shown in Fig. 1 GBDT modular concept figure.
Using linear adder model, the result of each decision tree is obtained into final prediction result by weighted accumulation.It should
Model mathematic(al) representation are as follows:
(1) in formula, x is to input total sample, ht(x；wt) it is every post-class processing, wtFor the ginseng of each post-class processing
Number, αtIt is the weight of each tree, T is the number of decision tree.
In the learning process of each round, a weak learner h is generatedt(x；wt), the loss function of weak learner are as follows:
(2) in formula, Ft-1(xi；wt) it is current model, GBDT determines Weak Classifier by minimizing loss function value
Parameter, the selection of loss function itself can have a L1, and Squared Error Loss, 0-1 loss function etc., this patent application is to return to ask
Topic, therefore select quadratic loss function, corresponding mathematical formulae are as follows:
(3) in formula, yiFor true value, h (xi) it is model estimate value, corresponding difference, that is, residual error.
Specific algorithm is realized as shown in table 1:
1 GDBT model algorithm of table realizes process
Also, electricity consumption GBDT nonlinear model, prediction evaluation index use mean absolute percentage error (MAPE), table
Up to formula are as follows:
(4) X in formulaiFor practical daily power consumption, YiTo predict daily power consumption；The total electricity consumption of prediction in last statistical forecast month
The accuracy evaluation index the most final of amount and practical total electricity consumption, corresponding mathematic(al) representation are as follows:
Wherein, YSFor this month practical total electricity consumption in prediction month, PSFor the of that month prediction total electricity consumption in prediction month.
Fig. 2 shows this system electricity consumption GBDT nonlinear model building process.
A total of four office buildings of this experimental data the collected ammeter numerical value every 5 minutes, innovated with Beijing
Mansion data are sample, and the process of entire data prediction is described in detail.Pass through the history day of Beijing that crawler acquires
Destiny evidence.The range of training data is on November 9th, 2017 by 06 31st, 2018, and initial data is as shown in table 2.
The initial data of 2 Experiment Training of table
Data above is subjected to data cleansing first, groundwork is to merge the data in each month, and by data
It is separated into according to equipment in different tables, finally weeds out some unreasonable dirty datas.Then data prediction is carried out, including
(this experiment fills those electricity consumptions lacked using mean value) is filled up for missing values, since power data is for this reality
It tests and does not act on, selection is rejected, and finally crawls 2345 weather forecasts and the two nets of Chinese weather using scrapy crawler frame
It stands, obtains historical weather data.It is as shown in table 3 after data processing.
Experimental data after 3 data prediction of table
Wherein, T_MAX indicates that daily maximum temperature, T_AVG represent mean daily temperature, and T_MIN represents daily minimal tcmperature, YL generation
Table daily power consumption.Corresponding daily power consumption distribution is as shown in Figure 3.
As seen in Figure 3, the fluctuation of electricity consumption is bigger, but there is also stronger between electricity consumption and date simultaneously
Periodicity, to this present embodiment to timestamp extraction time feature.Simultaneously by statisticalling analyze the related coefficient between variable,
As shown in table 4, it can be seen that there are stronger correlations for daily power consumption and daily temperature.
The correlation analysis of table 4 daily power consumption and temperature
| YL | T_MIN | T_MAX | T_AVG | |
| YL | 1.000000 | 0.475377 | 0.404250 | 0.443254 |
| T_MIN | 0.475377 | 1.000000 | 0.969911 | 0.992489 |
| T_MAX | 0.404250 | 0.969911 | 1.000000 | 0.992409 |
| T_AVG | 0.443254 | 0.992489 | 0.992409 | 1.000000 |
By signature analysis, the present embodiment is extracted the temporal characteristics such as table 5, not due to the season information in the library pandas
The seasonal conditions in specific place can be objectively responded out.In this regard, we are by analyzing the temperature variations in each area, selection compared with
For reasonable seasonal variations index, to determine seasonal characteristic.The corresponding different degree coefficient of each feature, numerical value have been obtained simultaneously
Show that the different degree of feature is higher more greatly.
5 feature description of table and significant coefficient
| Feature | Explanation | Characteristic importance |
| Year | Year information | 0.0000002 |
| Month | Month information | 0.0008050 |
| Day | Day information | 0.108818 |
| Dayofweek | Week information | 0.004686 |
| Quarter | Season information | 0.050045 |
| Dayofyear | In 1 year how many days | 0.511821 |
| Weekofyear | Which week in 1 year | 0.069484 |
| Holidays | It whether is festivals or holidays | 0.175032 |
| 1_m_mean | First 1 month electricity consumption mean value | 0.002069 |
| 1_m_std | First 1 month electricity consumption variance | 0.000797 |
| 2_m_mean | First 2 months electricity consumption mean values | 0.008490 |
| 2_m_std | First 2 months electricity consumption variances | 0.007784 |
| 1_t_mean | First 1 month temperature mean value | 0.005912 |
| 1_t_std | First 1 month temperature variance | 0.001338 |
| 2_t_mean | First 2 months temperature mean values | 0.004759 |
| 2_t_std | First 2 months temperature variances | 0.000259 |
The present embodiment has chosen the principal element of the above-mentioned every daily power consumption of influence office building electric system, and utilizes above-mentioned
Discrete data constructs electricity consumption GBDT nonlinear model, predicts the every daily power consumption of office building electric system and monthly electricity consumption total amount.
As shown in Fig. 2, electricity consumption GBDT nonlinear model building of the present invention, comprising the following steps:
1, electric power initial data obtains, and cleans electric power initial data, saves the every daily power consumption historical data of office building to going through
In history electricity consumption data storage cell.
Preferably, office building history day electricity consumption data packet described in this system includes the every 5 minutes power consumption datas of distinct device,
The cleaning office building history electricity consumption data includes: to be handled using big data correlation technique, and initial data is carried out duplicate checking,
Suppressing exception value, fills up missing values, extracts the power consumption total quantity index of high-grade equipment, extracts parameter that this system needs and corresponding
Format, and using day as the period save.
2, raw temperature data obtains, and cleans historical temperature data, is saved in historical temperature data storage unit.
Preferably, the acquisition raw temperature data, comprising: temperature data is crawled using crawler, passes through acquisition
Different cities code in configuration file is come the temperature data in city where obtaining office building；The cleaning historical temperature data,
Include: the deletion filled up with repeated data to missing data, extracts 3 temperature parameters that this system needs: minimum temperature, most
High-temperature and mean temperature, all temperature units are degree Celsius, and are stored in temperature information storage unit.
Further preferably, the acquisition historical temperature data, further includes: while determining to obtain by the control of function variable
What is taken is the predicted temperature or historical temperature of forecast date.
3, temperature information determines；
As shown in figure 4, temperature information determines, including the following steps:
1) judge whether that selection judges whether out heating using temperature；
2) if it is, continuing to determine whether to reach out heating standard；
If 3) reach out heating standard, heating index is assigned a value of 1, if not reaching out heating standard, heating index
It is assigned a value of 0, is terminated；
If 4) judge whether out heating without using temperature, judge whether to reach spring, if it is, season index is assigned
Value is 0, is terminated；
If 5) do not reach spring, judge whether to reach summer, be, then season index is assigned a value of 1, terminates；
If 6) do not reach summer, judge whether to reach autumn, be, then season index is assigned a value of 2, terminates；
If 7) do not reach autumn, judge whether to reach winter, be, then season index is assigned a value of 3, terminates, if do not reached
To winter, then " not meeting default standard, request artificial setting " is prompted.
4, holiday information determines；
Wherein, this system has done the file for containing all festivals or holidays in January, 2017 in January, 2019, file
The middle festivals or holidays attribute that the corresponding date is judged using two attribute values of holiday attribute and Spring Festival attribute:
If weekend, then holiday attribute assignment is 1, and if three days or more vacations, holiday attribute assignment was 2, if
Seven days or more vacations, holiday attribute assignment are 3, and if during the New Year, then holiday attribute assignment is 4, if working day, then
The holiday attribute assignment is 0；
For example holiday in the Spring Festival, then Spring Festival attribute assignment is 1, such as non-holiday in the Spring Festival, then Spring Festival attribute assignment is 0.
Preferably, two assignment of holiday attribute and Spring Festival attribute can not correspond, and holiday attribute is specifically had a holiday or vacation the time
It is determined according to the legal festivals and holidays of current year, and Spring Festival attribute has fractional unit to start to have a holiday or vacation and can start setting up during the Spring Festival
It is 1, which is arranged by a database.
5, temporal information is extracted；
Wherein, this system using the library datetime of Python extract this day belong to this year which week and should
Which day in week, and save it in temporal information storage element.
6, data merge；
The discrete data obtained after processing is merged according to time series, is stored in csv file；
7, model foundation and optimization, using electricity consumption as output valve, temperature information, holiday information, temporal information etc. work
For input value, GBDT model will be calculated using the method for web search Optimal Parameters using 10 folding cross validations as verification mode
Method optimization, obtains the best model of GBDT.
8, model saves.
Fig. 5 shows the prediction technique of the monthly electricity consumption of office building electric system of the present invention.
As shown in figure 5, the present invention provides a kind of prediction technique of monthly electricity consumption of office building electric system, including it is following
Step:
Step 1, input inquiry time, creation time sequence；After the present embodiment input time, the system automatically generated date
The continuous 3 months time serieses with the day period afterwards；
Step 2, predicted temperature data are obtained, predicted temperature data are cleaned；
Step 3, electricity consumption GBDT nonlinear model is called, every daily power consumption and monthly electricity consumption total amount, output prediction are calculated
As a result.
Present patent application predicts the monthly electricity consumption of certain office building electric system in 1 to 31 of August in 2018, specifically
Data are shown in Table the 6 monthly electricity demand forecasting results of certain office building electric system and truthful data list.
The monthly electricity demand forecasting result of certain the office building electric system of table 6 and truthful data list
Fig. 6 prediction result figure compared with truthful data shows the effect of the office building prediction result.
The purpose of the application, technical scheme and beneficial effects are described in detail in particular embodiments described above,
It should be understood that described above, be not used to limit the protection scope of the application, it is all spirit herein and principle it
Any modifications, equivalent replacements, and improvements etc. that is interior, being done, should be included within the scope of protection of this application.
Claims (12)
1. a kind of monthly electricity demand forecasting method of office building electric system, comprising:
S1, input inquiry time create predicted time sequence；
S2 obtains predicted temperature initial data, cleans predicted temperature initial data；
S3 imports electricity consumption GBDT nonlinear model, calculates every daily power consumption and export monthly electricity consumption Prediction of Total result；
It is characterized in that, further including constructing electricity consumption GBDT nonlinear model using discrete data, step includes:
1) electric power initial data is obtained, electric power initial data is cleaned, saves history daily power consumption data to history electricity consumption data
In storage unit；
2) raw temperature data is obtained, raw temperature data is cleaned, is saved in historical temperature data storage unit；
3) temperature information determines；
4) festivals or holidays determine；
5) temporal information is extracted；
6) data merge, and the discrete data obtained after processing is merged according to time series, is stored in csv file；
7) model foundation and optimization, using electricity consumption as output valve, temperature information, holiday information, temporal information etc. are as defeated
Enter value, it is using 10 folding cross validations as verification mode, algorithm is excellent to the method that GBDT model uses web search Optimal Parameters
Change, obtains electricity consumption GBDT nonlinear model；
8) model saves.
2. a kind of monthly electricity demand forecasting method of office building electric system according to claim 1, which is characterized in that described
Electric power initial data includes the every 5 minutes power consumption datas of distinct device, and the cleaning electric power initial data includes: utilizing big number
It is handled according to correlation technique, initial data is subjected to duplicate checking, suppressing exception value fills up missing values, and the power consumption of tabulating equipment is total
Figureofmerit extracts parameter and corresponding format that this system needs, and saves by the period of day.
3. a kind of monthly electricity demand forecasting method of office building electric system according to claim 1, which is characterized in that described
Obtain raw temperature data, comprising: crawl using crawler to temperature data, by obtaining the different cities in configuration file
Code obtains the temperature data in city where office building；The cleaning raw temperature data, comprising: missing data is filled up
With the deletion of repeated data, 3 temperature parameters that this system needs: minimum temperature, maximum temperature and mean temperature, institute are extracted
There is the temperature unit to be degree Celsius, and is stored in historical temperature information memory cell.
4. a kind of monthly electricity demand forecasting method of office building electric system according to claim 3, which is characterized in that described
Obtain raw temperature data, further includes: while determining that is obtained is the pre- thermometric of forecast date by the control of function variable
Degree or historical temperature.
5. a kind of monthly electricity demand forecasting method of office building electric system according to claim 1, which is characterized in that described
Temperature information determines that step includes:
1) judge whether that selection judges whether out heating using temperature；
2) if it is, continuing to determine whether to reach out heating standard；
If 3) reach out heating standard, heating index is assigned a value of 1, if not reaching out heating standard, heating index assignment
It is 0, terminates；
If 4) judge whether heating without using temperature, judge whether to reach spring, if it is, season index is assigned a value of 0,
Terminate；
If 5) do not reach spring, judge whether to reach summer, be, then season index is assigned a value of 1, terminates；
If 6) do not reach summer, judge whether to reach autumn, be, then season index is assigned a value of 2, terminates；
If 7) do not reach autumn, judge whether to reach winter, be, then season index is assigned a value of 3, terminates, if not reaching the winter
Then prompt " not meeting default standard, request artificial setting " in season.
6. a kind of monthly electricity demand forecasting method of office building electric system according to claim 1, which is characterized in that described
Festivals or holidays determine, the festivals or holidays attribute on corresponding date is judged including using two attribute values of vacation attribute and Spring Festival attribute:
If weekend, then vacation attribute assignment is 1, and if three days or more vacations, vacation attribute assignment was 2, if seven days
Or more vacation, vacation attribute is assigned a value of 3, and if during the Spring Festival, then vacation attribute value is 4, if working day, the then vacation
Phase attribute assignment is 0；
The for example Spring Festival, then Spring Festival attribute assignment is 1, such as the non-Spring Festival, then Spring Festival attribute assignment is 0.
7. a kind of monthly electricity demand forecasting method of office building electric system according to claim 5, which is characterized in that described
Festivals or holidays determine that two attribute values of vacation attribute and Spring Festival attribute can not correspond, vacation attribute specifically have a holiday or vacation the time according to
Determined according to the legal festivals and holidays of current year, and Spring Festival attribute have during the Spring Festival fractional unit start to have a holiday or vacation can start setting up for
1, which is arranged by a library.
8. a kind of monthly electricity demand forecasting method of office building electric system according to claim 1, which is characterized in that described
Temporal information is extracted, including using the library datetime of Python to extract the date in the cycle information of current year and in this week
Week several information, and save it in temporal information storage element.
9. a kind of monthly electricity demand forecasting method of office building electric system described in -8 according to claim 1, which is characterized in that institute
Electricity consumption GBDT nonlinear model is stated, is a kind of Boosting model based on tree, using serial manner, each iteration selection
One weak learner, while a new decision tree is established on the gradient direction for reducing residual error, it, will using linear adder model
The result of each decision tree obtains final prediction result by weighted accumulation, the model mathematic(al) representation are as follows:
(1) in formula, x is to input total sample, ht(x；wt) it is every post-class processing, wtFor the parameter of each post-class processing, αt
It is the weight of each tree, T is the number of decision tree；
In the learning process of each round, a weak learner h is generatedt(x；wt), the loss function of weak learner are as follows:
(2) in formula, Ft-1(xi；wt) it is current model, GBDT determines the ginseng of Weak Classifier by minimizing loss function value
Number, loss function select quadratic loss function, corresponding mathematical formulae are as follows:
(3) in formula, yiFor true value, h (xi) it is model estimate value, corresponding difference, that is, residual error.
10. a kind of monthly electricity demand forecasting method of office building electric system described in -8 according to claim 1, which is characterized in that
The electricity consumption GBDT nonlinear model, prediction evaluation index use mean absolute percentage error (MAPE), expression formula are as follows:
(4) X in formulaiFor practical daily power consumption, YiTo predict daily power consumption；The prediction total electricity consumption in last statistical forecast month and
The accuracy of practical total electricity consumption evaluation index the most final, corresponding mathematic(al) representation are as follows:
Wherein, YSFor this month practical total electricity consumption in prediction month, PSFor the of that month prediction total electricity consumption in prediction month.
11. a kind of monthly electricity demand forecasting method of office building electric system according to claim 1, which is characterized in that institute
State creation predicted time sequence include creation institute future on defeated date it is trimestral using day as the time series in period.
12. a kind of monthly electricity demand forecasting system of office building electric system, comprising:
Time series creating unit, input prediction date simultaneously generate predicted time sequence；
Predicted temperature acquiring unit obtains forecast date temperature；
Electricity consumption GBDT nonlinear model unit is imported, calculates daily power consumption and monthly electricity consumption total amount, and export prediction result；
It is characterized in that, further including electricity consumption GBDT nonlinear model construction unit, step includes:
1) acquisition of electric power initial data and cleaning unit save history daily power consumption data to history electricity consumption data storage cell
In；
2) raw temperature data acquisition and cleaning unit, are saved in historical temperature data storage unit；
3) temperature information determination module；
4) festivals or holidays determination module；
5) temporal information extraction module；
6) data combiners block merges the discrete data obtained after processing according to time series, is stored in csv file
In；
7) model foundation and optimization module, using electricity consumption as output valve, temperature information, holiday information, temporal information etc. are made
For input value, GBDT model will be calculated using the method for web search Optimal Parameters using 10 folding cross validations as verification mode
Method optimization, obtains electricity consumption GBDT nonlinear model；
8) model preserving module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910527955.5A CN110245798A (en) | 2019-06-18 | 2019-06-18 | A kind of monthly electricity demand forecasting method and system of office building electric system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910527955.5A CN110245798A (en) | 2019-06-18 | 2019-06-18 | A kind of monthly electricity demand forecasting method and system of office building electric system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110245798A true CN110245798A (en) | 2019-09-17 |
Family
ID=67887872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910527955.5A Pending CN110245798A (en) | 2019-06-18 | 2019-06-18 | A kind of monthly electricity demand forecasting method and system of office building electric system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245798A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782094A (en) * | 2019-10-28 | 2020-02-11 | 国网江苏省电力有限公司苏州供电分公司 | Subentry load prediction method for fine-grained electricity consumption behaviors of residential users |
| CN111122963A (en) * | 2019-12-03 | 2020-05-08 | 深圳供电局有限公司 | Time-sharing electric quantity abnormity judgment and repair method for electric meter of metering automation system |
| CN111177651A (en) * | 2019-12-03 | 2020-05-19 | 深圳供电局有限公司 | Time-sharing missing code fitting method for electric meter of metering automation system |
| CN111178611A (en) * | 2019-12-23 | 2020-05-19 | 广西电网有限责任公司 | Method for predicting daily electric quantity |
| 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 |
| CN112465256A (en) * | 2020-12-08 | 2021-03-09 | 华中科技大学 | Building power consumption prediction method and system based on Stacking model fusion |
| CN112529301A (en) * | 2020-12-14 | 2021-03-19 | 南京中兴力维软件有限公司 | Power consumption prediction method, equipment and storage medium |
| CN112949928A (en) * | 2021-03-12 | 2021-06-11 | 国网能源研究院有限公司 | Method and device for analyzing influence of winter air temperature on regional power grid power load |
| CN113627682A (en) * | 2021-08-25 | 2021-11-09 | 深圳供电局有限公司 | Method and system for predicting daily electric quantity fluctuation of non-residential users |
| CN116166857A (en) * | 2023-04-26 | 2023-05-26 | 国网吉林省电力有限公司信息通信公司 | Big data-based power data processing system |
| CN116433440A (en) * | 2023-04-11 | 2023-07-14 | 北京西清能源科技有限公司 | Data autoregressive enhanced carbon emission measuring and calculating method, system and electronic equipment |
| CN116956198A (en) * | 2023-09-18 | 2023-10-27 | 湖南迈格瑞特物联科技有限公司 | Intelligent electricity consumption data analysis method and system based on Internet of things |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
| CN108876019A (en) * | 2018-05-31 | 2018-11-23 | 中国电力科学研究院有限公司 | A kind of electro-load forecast method and system based on big data |
-
2019
- 2019-06-18 CN CN201910527955.5A patent/CN110245798A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
| CN108876019A (en) * | 2018-05-31 | 2018-11-23 | 中国电力科学研究院有限公司 | A kind of electro-load forecast method and system based on big data |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782094B (en) * | 2019-10-28 | 2022-06-21 | 国网江苏省电力有限公司苏州供电分公司 | Subentry load prediction method for fine-grained electricity consumption behaviors of residential users |
| CN110782094A (en) * | 2019-10-28 | 2020-02-11 | 国网江苏省电力有限公司苏州供电分公司 | Subentry load prediction method for fine-grained electricity consumption behaviors of residential users |
| CN111122963A (en) * | 2019-12-03 | 2020-05-08 | 深圳供电局有限公司 | Time-sharing electric quantity abnormity judgment and repair method for electric meter of metering automation system |
| CN111177651A (en) * | 2019-12-03 | 2020-05-19 | 深圳供电局有限公司 | Time-sharing missing code fitting method for electric meter of metering automation system |
| CN111178611A (en) * | 2019-12-23 | 2020-05-19 | 广西电网有限责任公司 | Method for predicting daily electric quantity |
| CN111178611B (en) * | 2019-12-23 | 2022-09-23 | 广西电网有限责任公司 | Method for predicting daily electric quantity |
| EP4085387A4 (en) * | 2019-12-31 | 2023-06-07 | 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 |
| CN112465256A (en) * | 2020-12-08 | 2021-03-09 | 华中科技大学 | Building power consumption prediction method and system based on Stacking model fusion |
| CN112529301A (en) * | 2020-12-14 | 2021-03-19 | 南京中兴力维软件有限公司 | Power consumption prediction method, equipment and storage medium |
| CN112529301B (en) * | 2020-12-14 | 2023-11-14 | 南京中兴力维软件有限公司 | Power consumption prediction method, device and storage medium |
| CN112949928A (en) * | 2021-03-12 | 2021-06-11 | 国网能源研究院有限公司 | Method and device for analyzing influence of winter air temperature on regional power grid power load |
| CN113627682A (en) * | 2021-08-25 | 2021-11-09 | 深圳供电局有限公司 | Method and system for predicting daily electric quantity fluctuation of non-residential users |
| CN116433440A (en) * | 2023-04-11 | 2023-07-14 | 北京西清能源科技有限公司 | Data autoregressive enhanced carbon emission measuring and calculating method, system and electronic equipment |
| CN116166857A (en) * | 2023-04-26 | 2023-05-26 | 国网吉林省电力有限公司信息通信公司 | Big data-based power data processing system |
| CN116166857B (en) * | 2023-04-26 | 2023-08-15 | 国网吉林省电力有限公司信息通信公司 | Big data-based power data processing system |
| CN116956198A (en) * | 2023-09-18 | 2023-10-27 | 湖南迈格瑞特物联科技有限公司 | Intelligent electricity consumption data analysis method and system based on Internet of things |
| CN116956198B (en) * | 2023-09-18 | 2023-12-26 | 湖南迈格瑞特物联科技有限公司 | Intelligent electricity consumption data analysis method and system based on Internet of things |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245798A (en) | A kind of monthly electricity demand forecasting method and system of office building electric system | |
| CN105260803B (en) | A kind of system power consumption prediction technique | |
| Babel et al. | A multivariate econometric approach for domestic water demand modeling: an application to Kathmandu, Nepal | |
| Cao et al. | A data-driven hybrid optimization model for short-term residential load forecasting | |
| González-Vidal et al. | Data driven modeling for energy consumption prediction in smart buildings | |
| de Santiago et al. | The generation of domestic hot water load profiles in Swiss residential buildings through statistical predictions | |
| CN109376936A (en) | Value of house prediction technique, device, computer equipment and storage medium | |
| CN106557835B (en) | Power consumption prediction method and system based on scenic gas index | |
| CN104951840A (en) | Analysis method for overall electricity utilization trend of electric power market | |
| JP2006011715A (en) | Estimation method for resource consumption, and device | |
| CN114239292B (en) | Comprehensive evaluation method and system for low-carbon economic operation-oriented multifunctional demand park | |
| CN115796915A (en) | Electricity price prediction method and system for electricity trading market | |
| Gerossier et al. | A novel method for decomposing electricity feeder load into elementary profiles from customer information | |
| CN104618133B (en) | A kind of grid voltage quality Monitoring Data acquisition method and system | |
| Zhang et al. | Quantile huber function guided TCN for short-term consumer-side probabilistic load forecasting | |
| Kedrowski et al. | Short-term seasonal energy forecasting | |
| Deng et al. | Medium-term rolling load forecasting based on seasonal decomposition and long short-term memory neural network | |
| Schmidt | Forecasting state retail sales: econometric vs. time series models | |
| Figueiredo et al. | A Data Mining Decision Support Tool to Study Electricity Retail Contracts | |
| CN113313533A (en) | Method for performing macroscopic economy prediction and monitoring by using electric power data | |
| Surur et al. | Determinants of Job Opportunities in Yogyakarta Special Regional Province | |
| Sinitsyna et al. | Some practical aspects of electric power consumption time series analysis | |
| Skribans | Construction demand: a model of research and forecast for Latvia from 2002 to 2025 | |
| CN107798124A (en) | Search system and method based on prediction modeling technique | |
| Li et al. | Research and application of electricity anti-stealing system based on neural network |
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 |
Application publication date: 20190917 |
|
| RJ01 | Rejection of invention patent application after publication |