CN109376972A - A kind of wisdom Power Network Short-Term Electric Load Forecasting method based on block cluster - Google Patents
A kind of wisdom Power Network Short-Term Electric Load Forecasting method based on block cluster Download PDFInfo
- Publication number
- CN109376972A CN109376972A CN201910000269.2A CN201910000269A CN109376972A CN 109376972 A CN109376972 A CN 109376972A CN 201910000269 A CN201910000269 A CN 201910000269A CN 109376972 A CN109376972 A CN 109376972A
- Authority
- CN
- China
- Prior art keywords
- power
- block
- day
- sample
- 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
- 238000013277 forecasting method Methods 0.000 title claims abstract description 6
- 230000005611 electricity Effects 0.000 claims abstract description 26
- 238000001514 detection method Methods 0.000 claims description 42
- 238000012549 training Methods 0.000 claims description 42
- 238000000034 method Methods 0.000 claims description 31
- 238000013527 convolutional neural network Methods 0.000 claims description 30
- 239000012634 fragment Substances 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 13
- 238000012935 Averaging Methods 0.000 claims description 7
- 238000013139 quantization Methods 0.000 claims description 3
- 235000006508 Nelumbo nucifera Nutrition 0.000 claims description 2
- 240000002853 Nelumbo nucifera Species 0.000 claims description 2
- 235000006510 Nelumbo pentapetala Nutrition 0.000 claims description 2
- 230000007423 decrease Effects 0.000 abstract description 2
- 230000003203 everyday effect Effects 0.000 abstract 1
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 206010033799 Paralysis Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 230000009885 systemic effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Primary Health Care (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Probability & Statistics with Applications (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of wisdom Power Network Short-Term Electric Load Forecasting methods based on block cluster, pass through clustering, electricity unit similar in the essential laws of Power system load data variation in region is divided into same block, block power load forecasting module group is uniformly established by block, it avoids model lengthy and jumbled, promotes the generalization ability and precision of prediction of power load forecasting module；Meanwhile with the Zhou Zuowei period, in analysis one week every day electric load fluctuation pattern, multiple stable state periods were divided by one day, each period establishes power load forecasting module respectively, further decreases the fluctuation interference of electric load time series itself, greatly improves precision of prediction and robustness.
Description
Technical field
The invention belongs to Techniques for Prediction of Electric Loads fields, and in particular to a kind of wisdom power grid based on block cluster is short-term
Methods of electric load forecasting.
Background technique
Power grid short-term electric load is the fluctuation pattern by historical load, in conjunction with social environmental factor to coming few hours
Or several days loads are predicted.Timely and effectively load forecast is to the arrangement scheduling of grid power, Power System Intelligent
The raising of change level plays the role of directiveness.On July 30th, 2012 and July 31, India's power grid is in the continuous two days time
It is interior that super large area blackout twice occurs, it makes a very bad impression, surpass 6.8 hundred million people can be used without electricity, surpass the stoppage in transit of 500 column trains, bank
Systemic breakdown is affected by power failure more than half territory, India's power grid without early warning overload operation be accident occur important original
One of because.On August 14th, 2003, Northeastern United States and eastern Canada interconnected power system system unstability, cause include New York and
More than ten areas including the state of Michigan have a power failure, and public transportation system paralysis, economy also receives serious blow, according to New York City Hall
Estimation, power failure cause 7.5 hundred million dollars of the New York finance reduction of income, and Canada's aspect economic loss is up to 2,300,000,000 Canadian dollars, electric load
The low precision and non intelligentization of prediction are the one of the major reasons to cause the accident.
Summary of the invention
For technical problem of the existing technology, the present invention proposes that a kind of wisdom power grid based on block cluster is electric in short term
Power load forecasting method improves the precision of prediction and robustness of electric load.
To realize the above-mentioned technical purpose, the technical solution adopted by the present invention are as follows:
A kind of wisdom Power Network Short-Term Electric Load Forecasting method based on block cluster, comprising the following steps:
Step 1, obtain target prediction region in each electricity unit in continuous N1The voltage and current history number of a cycle
According to, the voltage and current historical data of each electricity unit is handled to obtain corresponding power sequence, all electricity units
The power training sample in power sequence composition target prediction region；
Step 2, each electricity unit is clustered according to the power sequence of each electricity unit, by target prediction region division
For N3A block；
Step 3, using power training sample, the power load forecasting module group based on convolutional neural networks is established
Step 3.1, it using all power sequences in each block, establishes respectively and N3The corresponding N of a block3A block electricity
Power load forecasting model group；
Each block power load forecasting module group includes 7 days time power load forecasting module group, wherein 7 days
Electric load model group is corresponding with 7 days one week respectively；Each day time electric load model group includes N4It is a to be based on convolutional Neural
The power load forecasting module of network, wherein N4A power load forecasting module respectively with one day N4A period is corresponding；
Step 3.2, all power sequences in each block are handled, each block obtains corresponding power load
Lotus training sample；
Step 3.3, successively with the b days of the continuous cycle of any two in the electric load training sample of a block times the
n2The power load charge values of period as input data, with the b days of next cycle time n-th2The power load charge values of period are as output
Data, the b days times n-th of training a block2The power load forecasting module based on convolutional neural networks of period；
Wherein, a=1,2 ..., N3, b=1,2 ..., 7, n2=1,2 ..., N4；
Step 4, using step 3 acquisition respectively based on the power load forecasting module of convolutional neural networks, to target prediction
The electric load that cycle is w+m, target prediction day is d is predicted；Nearest cycle wherein comprising day time d is w；
Step 4.1, the w-1 cycle of all electricity units and the voltage and current data of w cycle are obtained, and presses step
1 is handled to obtain the power prediction sample in target prediction region；
Step 4.2, the power prediction sample that step 4.1 obtains is handled by step 3.2, each block is corresponded to
Load forecast sample；
Step 4.3, select each block day time for the day time electric load model group of d；Successively by load forecast sample
Middle correspondence w-1 cycle d days secondary and d days each block day parts of w cycle power load charge values correspond to day as corresponding block
Secondary d corresponds to the input data of the power load forecasting module based on convolutional neural networks of period, obtains w+1 cycle d days times
The Electric Load Forecasting measured value of each block day part；
Step 4.4, select each block day time for the day time electric load model group of d；Successively by load forecast sample
The electric power of d days each block day parts of power load charge values and w+1 cycle of d days each block day parts of middle correspondence w cycle
Predicted load correspond to the power load forecasting module based on convolutional neural networks that day secondary d corresponds to the period as corresponding block
Input data obtains the Electric Load Forecasting measured value of d days each block day parts of w+2 cycle；
Step 4.5, and so on, the Electric Load Forecasting measured value until obtaining d days each block day parts of w+m cycle,
By the Electric Load Forecasting measured value summation of d days each block day parts of w+m cycle, d days Electric Load Forecastings of w+m cycle are obtained
Measured value.
Further, the voltage and current historical data of each electricity unit is by the electric power detection device that is respectively arranged with identical
Frequency collection obtains, described to be handled voltage and current historical data to obtain the power training sample in target prediction region
Detailed process are as follows:
With sample moment time interval T1Voltage data and current data are divided, each sample moment time interval T is calculated1
Interior average voltage and current average, by each sample moment time interval T1Intermediate time as sample moment, will
Each sample moment time interval T1Voltage value and current value of the interior average voltage and current average as sample moment,
Using the product of each sample moment voltage value and current value as the performance number of sample moment；
The corresponding one section of power sequence being made of the performance number of all sample moments of each electric power detection device, all electric power
The power sequence group success rate training sample of detection device.
Further, the detailed process of the step 2 are as follows:
Using with the zero point in 1 day to 24 points for the secondary time interval T in day2, each electric power detection device N is successively obtained at random2
A length is T2Power sequence segment, 10≤N2≤15；
Successively by the N of each electric power detection device2A power sequence segment is averaged, and obtains the power of each electric power detection device
Mean sequence；
Using each electric power detection device as element p to be clusteredi, i=1,2,3 ..., n1, n1For the number of electric power detection device
Amount, the coordinate of element to be clustered are the power averaging sequence of corresponding electric power detection device；
Select cluster centre number for N3, k-means cluster is carried out, according to cluster result, by target prediction region division
For N3A block, 4≤N3≤ 8, each block includes several electric power detection devices.
Further, the process for carrying out k-means cluster are as follows:
Step A1 randomly chooses N using the power averaging sequence of each electric power detection device as the coordinate of element to be clustered3
For a element as cluster centre, each cluster centre represents 1 clustering cluster；
Step A2, successively calculate all elements with each cluster centre distance, and successively by each element be assigned to therewith away from
From clustering cluster representated by nearest cluster centre；
Using Euclidean distance quantization element with the distance of each cluster centre, formula is as follows:
In formula, distance (pi,kj) indicate element piWith cluster centre kjDistance,Indicate the n-th of i-th of element
A coordinate,Indicate that n-th of coordinate of j-th of cluster centre, N indicate the sample moment time interval T for including in 1 day1Number
Amount；
Step A3 calculates the average coordinates of all elements in each clustering cluster using Euclidean distance, calculates according to this each
Average coordinates in clustering cluster with each cluster centre distance, if the average coordinates of each clustering cluster are equal with the distance of each cluster centre
Less than or equal to threshold valueCluster is completed, and current cluster result is obtained；The same cluster centre of the average coordinates of any clustering cluster if it exists
Distance be greater than threshold valueUsing the average coordinates of each clustering cluster cluster centre brand new as each cluster, step A2 is gone to.
Further, threshold valueValue be 0.05.
Further, the detailed process of the step 3.2 are as follows:
Step B1 successively sums the power sequence of all electric power detection devices in block same in power training sample,
Acquisition and N3The corresponding N of a block3A block power sequence；
Step B2 successively selects the N in the block power sequence of a block by Zhou Tianci for b1A block power sequence piece
Subsample is trained for the power of b in section, composition a block day time；Wherein a=1,2 ..., N3, b=1,2 ..., 7；
Step B3 randomly chooses 20 block power sequence pieces in the power training subsample of b from a block day time
Section is averaged to obtain the block power sequence average fragment that a block day time is b；Wherein, block power sequence average fragment
In each sample moment correspond to a power average value；
A block day time is divided into N in one day for b using block power sequence average fragment by step B44A period,
Division methods are as follows: in block power sequence average fragment, if meeting latter sample moment power average value and current sample moment
The absolute difference of power average value is greater than 30% of the minimum value in two sample moment power average values, then when current sample
Carve be critical sample moment, and using critical sample moment be used as the finish time of present period, critical sample moment it is rear as
At the beginning of this moment is used as subsequent period；When block power sequence average fragment end sample moment is unsatisfactory for critical sample
Quarter condition；
Step B5 successively calculates the according to Time segments division as a result, being the power training subsample of b using a block day time
Subsample is trained for the electric load of b in a block day time；
By the performance number of each sample moment in the c period and sample moment time interval T1The sum of product, when as c
The power load charge values of section；Wherein, c=1,2 ..., N4, then 1 day N4The power load charge values composition a block day time of a period is b's
1 electric load sequence, time span N1Then correspondence obtains N for the power training subsample of a continuous cycle1A electric load sequence
Column, N1A electric load sequence composition a block day time is the electric load training subsample of b, successively takes b by 7 days of week
=1,2 ..., the electric load training sample of 7 composition a blocks.
Further, time span N1Value are as follows: N1≥50。
Further, sample moment time interval T1Value be 20 minutes.
Further, the voltage and current historical data is the data after unscented kalman filter.
Beneficial effect
Firstly, collection voltages, current history data, calculating form power time series, sufficiently reflect this programme simultaneously
The essential laws of Power system load data variation, are obtained Power system load data by power time series and are predicted, effectively
Improve the accuracy and validity of load forecast.
Secondly, this programme is for Power system load data to the electric loads influence factor such as temperature, wind speed, humidity and red-letter day
Low sensitivity characteristic proposes the classification method based on Power system load data self attributes, analyzes the Power x Time of each electricity unit
Sequence is clustered by unit similar in fluctuation pattern of the clustering algorithm by power sequence, ultimately forms basicly stable block electricity
Power Load Time Series reduce the complexity of Power system load data and non-linear, reduce the mottled degree of data, based on block point
Block power load forecasting module group is not established and is predicted, the accuracy and stabilization of load forecast are further improved
Property, meanwhile, by clustering, reduce the quantity of model needed for load forecast, improves prediction timeliness and method
Operability.
Again, this programme established day time electric load series model group for 7 days one week respectively, improved Electric Load Forecasting
Precision is surveyed, meanwhile, according to block power time series, the time for concentrating electricity consumption in block in unit one day is analyzed, according to block
The degree of fluctuation of power time series will be divided into multiple stable states by the block power time series of length of day, if corresponding
Dry stable state period, power load forecasting module is established respectively for each period, further reduced Power system load data
Complexity and unstability greatly improve load forecast precision and robustness.
Detailed description of the invention
Fig. 1 is short-term electric load prediction flow chart of the present invention.
Specific embodiment
Elaborate below to the embodiment of the present invention, the present embodiment with the technical scheme is that according to development,
The detailed implementation method and specific operation process are given, is further explained explanation to technical solution of the present invention.
Due to the complexity of Power system load data and non-linear so as to the prediction difficulty of future electrical energy load data compared with
Height, for the precision of prediction and robustness for improving electric load, it is short-term that the present invention provides a kind of wisdom power grid based on block cluster
Methods of electric load forecasting carries out clustering by the power sequence to each electricity unit, by electric load in estimation range
Unit similar in the essential laws of data variation gathers same block, uniformly establishes power load forecasting module, avoids model
It is lengthy and jumbled, promote the generalization ability and precision of prediction of power load forecasting module；Meanwhile with the Zhou Zuowei period, analyze every in one week
The fluctuation pattern of one day electric load was divided into multiple stable state periods for one day, and each period establishes load forecast respectively
Model further decreases the fluctuation interference of electric load time series itself, greatly improves precision of prediction and robustness.
The present embodiment will further explain the present invention program in conjunction with attached drawing 1, specifically includes the following steps:
Step 1: installation electric power detection device obtains history voltage and current data.
Electric power detection device is successively installed respectively to electricity unit all in target prediction region；
The voltage and current historical data in each electric power detection device is acquired simultaneously with identical sample frequency, successively to each electricity
Voltage and current historical data in power detection device carries out unscented kalman filter, with sample moment time interval T1Divide filter
Voltage and current historical data after wave calculates each sample moment time interval T1The average value of interior voltage and current, will be every
A sample moment time interval T1Intermediate time as sample moment, by each sample moment time interval T1Interior voltage and electricity
Voltage and current value of the average value of stream as sample moment, using the product of each sample moment voltage value and current value as sample
The performance number at this moment obtains the power sequence of each electric power detection device；
It is acquisition starting with Monday zero point from each electric power detection device voltage and current historical data collected
Point is acquisition terminating point with 24 points of weekend, includes N1The voltage and current historical data of a continuous cycle, N1Value is at least
It is 50；
Sample moment time interval T1Value be 20 minutes；
The corresponding one section of power sequence comprising all sample moments of each electric power detection device, all electric power detection devices
Power sequence group success rate training sample；
In power training sample, each power sequence contains 3 power datas, respectively the 10th minute, the per hour
30 minutes and the 50th minute performance numbers；
Step 2: power training sample is utilized, by target prediction region division block；
Using with the zero point in 1 day to 24 points for the secondary time interval T in day2, each electric power detection device N is successively obtained at random2
A length is T2Power sequence segment, 10≤N2≤15；
Successively by the N of each electric power detection device2A power sequence segment is averaged, and obtains the power of each electric power detection device
Mean sequence；
Using each electric power detection device as element to be clustered, it is denoted as pi, i=1,2,3..., n1, n1It is set for electric power detection
Standby quantity, the coordinate of element to be clustered are the power averaging sequence of corresponding electric power detection device；
Select cluster centre number for N3, k-means cluster is carried out, according to cluster result, by target prediction region division
For N3A block, 4≤N3≤ 8, each block contains multiple electric power detection devices, and same block may not phase on geographical location
Even；
Cluster process is as follows:
Step A1: using the power averaging sequence of each electric power detection device as the coordinate of element to be clustered, N is randomly choosed3
For a element as cluster centre, each cluster centre represents 1 clustering cluster；
Step A2: successively calculate all elements with each cluster centre distance, and successively by each element be assigned to therewith away from
From clustering cluster representated by nearest cluster centre；
Using Euclidean distance quantization element with the distance of each cluster centre, formula is as follows:
In formula, distance (pi,kj) indicate element piWith cluster centre kjDistance,Indicate the n-th of i-th of element
A coordinate,Indicate n-th of coordinate of j-th of cluster centre；
Step A3: calculating the average coordinates of all elements in each clustering cluster using Euclidean distance, calculates according to this each
Average coordinates in clustering cluster with cluster centre distance, if the average coordinates of each clustering cluster are respectively less than with the distance of cluster centre
Equal to threshold valueCluster is completed, and current cluster result is obtained；If it exists the average coordinates of any clustering cluster with cluster centre away from
From greater than threshold valueUsing the average coordinates of each clustering cluster cluster centre brand new as each cluster, step A2 is gone to；
This example, threshold valueIt is selected as 0.05.
Step 3: utilizing power training sample, establish the power load forecasting module group based on convolutional neural networks；
Successively the power sequence of all electric power detection devices of block same in power training sample is summed, obtains N3A area
Block power sequence；
Utilize N3A block power sequence, establishes N respectively3A block electric load model group, each block Electric Load Forecasting
Surveying model group includes 7 days time power load forecasting module group, 7 days time power load forecasting module groups respectively with one week seven
It is corresponding；Total N3* 7 respectively correspond the different blocks not secondary power load forecasting module based on convolutional neural networks on the same day
Group；
It is below the power load forecasting module based on convolutional neural networks of 7 (being assumed to be Monday) with day in block 2 time
For the building process of group, as shown in Figure 1, to the electric load based on convolutional neural networks in each day in one week of each block time
The building process of prediction model group is illustrated:
The N that selection current block power sequence includes1The block power sequence piece of all corresponding Mondays in a continuous cycle
Section, composition current block correspond to the power training subsample of Monday, and the power training subsample that current block corresponds to Monday includes
N1A time span is the block power sequence segment of 1 day and corresponding Monday；
20 block power sequence segments are randomly choosed from the power training subsample that current block corresponds to Monday, it will be upper
It states 20 block power sequence segments and takes 1 block power sequence average fragment of average acquisition, block power sequence average fragment
Comprising 72 (24*3) a sample moments, the corresponding power average value of each sample moment is averaged piece using block power sequence
Section, is divided into N for the Monday of current block4A period, division methods are as follows:
Remember in block power sequence average fragment, meets latter sample moment power average value and current sample moment power
30% sample moment that the absolute difference of average value is greater than the minimum value of above-mentioned two sample moment power average value is critical
Sample moment, finish time of the critical sample moment as present period, under the latter sample moment of critical sample moment is used as
At the beginning of one period；Block power sequence average fragment end sample moment is unsatisfactory for critical sample moment condition；
According to Time segments division as a result, successively calculating current block using the power training subsample that current block corresponds to Monday
The electric load training subsample of corresponding Monday, the electric load training subsample that current block corresponds to Monday includes N1A electric power
Load sequence, each electric load sequence is corresponding with the Monday in a certain week, and each electric load sequence includes N4Power load
Charge values, N4A same N of power load charge values4A period is corresponding；
In the power load charge values of a certain period i.e. period performance number Yu T product of each sample moment and；
Using current block correspond to Monday electric load training subsample as current block correspond to Monday based on convolution
The training data of the power load forecasting module group of neural network；
The power load forecasting module group based on convolutional neural networks that current block corresponds to Monday includes N4It is a to be based on volume
The power load forecasting module of product neural network, same to N4A period is corresponding；
Successively with any two continuous cycle Monday n-th in electric load training subsample2The power load charge values of period are made
For the input data of the power load forecasting module based on convolutional neural networks of corresponding period, n2=1,2,3 ..., N4, below
One cycle Monday n-th2Load forecast based on convolutional neural networks of the power load charge values of period as the corresponding period
The output data of model, training convolutional neural networks obtain corresponding n-th2The electric load based on convolutional neural networks of period
Prediction model；
Successively by n2From 1 value until N4, N is obtained altogether4A electric power based on convolutional neural networks corresponding with the period respectively
Load forecasting model, to obtain the power load forecasting module group based on convolutional neural networks that current block corresponds to Monday；
All blocks are successively selected, each block successively selects all days, is analogous to aforementioned current block and corresponds to Monday
The power load forecasting module group based on convolutional neural networks building process, obtain N3* 7 based on convolutional neural networks
Power load forecasting module group.
Step 4: using step 3 acquisition respectively based on the power load forecasting module of convolutional neural networks, carrying out short-term electricity
Power load prediction；
The voltage and current data that each electric power detection device includes nearest 3 cycles are obtained, according to power training in step 1
The preparation method of sample obtains power prediction sample；Herein select 3 cycles voltage and current data, be in order to guarantee to
It less include the day time data secondary for target prediction day, to input the driving respectively load forecast mould based on convolutional neural networks
Type, the final Electric Load Forecasting measured value for obtaining target prediction cycle target prediction day time.
The k-means cluster result and power prediction sample obtained using step 2 obtains N3A block power prediction sequence,
Each block power prediction sequence, that is, time span is that the sample moment time interval of 3 continuous cycles is T1Continuous power
Value；
The identical sequence fragment in day time in each block power prediction sequence is successively selected, each block is obtained respectively and corresponds to each day
Secondary total N3* 7 power prediction subsamples, each power prediction subsample include the block function that 3 time spans are one day
Rate forecasting sequence segment；
The Time segments division result and N obtained using step 33* 7 power prediction subsamples successively calculate power prediction increment
Power load charge values in this in day part obtain N3* 7 load forecast subsamples, all load forecast subsamples
Form load forecast sample；
Note target prediction cycle is w+m, and it includes target prediction day in load forecast sample that target prediction day time, which is d,
The nearest cycle of secondary d is w, and each block is selected to correspond to the power load forecasting module group based on convolutional neural networks of day time d, according to
It is secondary w-1 cycle to be corresponded in load forecast sample d days times and the electric load of d days each block day parts of w cycle
It is worth the input data for corresponding to the power load forecasting module group based on convolutional neural networks of day time d as each block, obtains w
The Electric Load Forecasting measured value of each block day part of+1 cycle d days time；
Each block is selected to correspond to the power load forecasting module group based on convolutional neural networks of day time d, successively by electric power
D days each blocks of the power load charge values of d days each block day parts of corresponding w cycle and w+1 cycle in load prediction sample
The Electric Load Forecasting measured value of day part corresponds to the power load forecasting module based on convolutional neural networks of day time d as each block
The input data of group obtains the Electric Load Forecasting measured value of d days each block day parts of w+2 cycle；
The rest may be inferred, the Electric Load Forecasting measured value until obtaining d days each block day parts of w+m cycle, by the w+m weeks
The Electric Load Forecasting measured value summation of secondary d days each block day parts obtains d days Electric Load Forecasting measured values of w+m cycle.
Above embodiments are preferred embodiment of the present application, those skilled in the art can also on this basis into
The various transformation of row or improvement these transformation or improve this Shen all should belong under the premise of not departing from the application total design
Within the scope of please being claimed.
Claims (9)
1. a kind of wisdom Power Network Short-Term Electric Load Forecasting method based on block cluster, which comprises the following steps:
Step 1, obtain target prediction region in each electricity unit in continuous N1The voltage and current historical data of a cycle is right
The voltage and current historical data of each electricity unit is handled to obtain corresponding power sequence, the power sequence of all electricity units
The power training sample in column composition target prediction region；
Step 2, each electricity unit is clustered according to the power sequence of each electricity unit, is N by target prediction region division3
A block；
Step 3, using power training sample, the power load forecasting module group based on convolutional neural networks is established
Step 3.1, it using all power sequences in each block, establishes respectively and N3The corresponding N of a block3A block power load
Lotus prediction model group；
Each block power load forecasting module group includes 7 days time power load forecasting module group, wherein 7 day time electric power
Load model group is corresponding with 7 days one week respectively；Each day time electric load model group includes N4It is a to be based on convolutional neural networks
Power load forecasting module, wherein N4A power load forecasting module respectively with one day N4A period is corresponding；
Step 3.2, all power sequences in each block are handled, each block obtains corresponding electric load instruction
Practice sample；
Step 3.3, successively with the b days of the continuous cycle of any two in the electric load training sample of a block times n-th2Period
Power load charge values as input data, with the b days of next cycle time n-th2The power load charge values of period as output data,
The b days times n-th of training a block2The power load forecasting module based on convolutional neural networks of period；
Wherein, a=1,2 ..., N3, b=1,2 ..., 7, n2=1,2 ..., N4；
Step 4, using step 3 acquisition respectively based on the power load forecasting module of convolutional neural networks, to target prediction cycle
The electric load for being d for w+m, target prediction day is predicted；Nearest cycle wherein comprising day time d is w；
Step 4.1, obtain the w-1 cycle of all electricity units and the voltage and current data of w cycle, and by step 1 into
Row processing obtains the power prediction sample in target prediction region；
Step 4.2, the power prediction sample that step 4.1 obtains is handled by step 3.2, each block obtains corresponding electricity
Power load prediction sample；
Step 4.3, select each block day time for the day time electric load model group of d；It successively will be right in load forecast sample
It answers w-1 cycle d days times and the power load charge values of d days each block day parts of w cycle corresponds to d pairs of day time as corresponding block
The input data of the power load forecasting module based on convolutional neural networks of period is answered, d days each blocks of w+1 cycle are obtained
The Electric Load Forecasting measured value of day part；
Step 4.4, select each block day time for the day time electric load model group of d；It successively will be right in load forecast sample
Answer the electric load of d days each block day parts of power load charge values and w+1 cycle of d days each block day parts of w cycle
Predicted value corresponds to the input that day time d corresponds to the power load forecasting module based on convolutional neural networks of period as corresponding block
Data obtain the Electric Load Forecasting measured value of d days each block day parts of w+2 cycle；
Step 4.5, and so on, the Electric Load Forecasting measured value until obtaining d days each block day parts of w+m cycle, by w
The Electric Load Forecasting measured value summation of d days each block day parts of+m cycle obtains d days Electric Load Forecasting measured values of w+m cycle.
2. the method according to claim 1, wherein the voltage and current historical data of each electricity unit is by respective
The electric power detection device of setting is collected with identical frequency, described to be handled to obtain target to voltage and current historical data
The detailed process of the power training sample of estimation range are as follows:
With sample moment time interval T1Voltage data and current data are divided, each sample moment time interval T is calculated1Interior
Average voltage and current average, by each sample moment time interval T1Intermediate time as sample moment, will be each
Sample moment time interval T1Voltage value and current value of the interior average voltage and current average as sample moment, will be every
Performance number of the product of a sample moment voltage value and current value as sample moment；
The corresponding one section of power sequence being made of the performance number of all sample moments of each electric power detection device, all electric power detections
The power sequence group success rate training sample of equipment.
3. according to the method described in claim 2, it is characterized in that, the detailed process of the step 2 are as follows:
Using with the zero point in 1 day to 24 points for the secondary time interval T in day2, each electric power detection device N is successively obtained at random2It is a long
Degree is T2Power sequence segment, 10≤N2≤15；
Successively by the N of each electric power detection device2A power sequence segment is averaged, and obtains the power averaging of each electric power detection device
Sequence；
Using each electric power detection device as element p to be clusteredi, i=1,2,3..., n1, n1For the quantity of electric power detection device,
The coordinate of element to be clustered is the power averaging sequence of corresponding electric power detection device；
Select cluster centre number for N3, k-means cluster is carried out, is N by target prediction region division according to cluster result3It is a
Block, 4≤N3≤ 8, each block includes several electric power detection devices.
4. according to the method described in claim 3, it is characterized in that, the process for carrying out k-means cluster are as follows:
Step A1 randomly chooses N using the power averaging sequence of each electric power detection device as the coordinate of element to be clustered3A element
As cluster centre, each cluster centre represents 1 clustering cluster；
Step A2 successively calculates all elements with the distance of each cluster centre, and successively each element is assigned to therewith apart from most
Clustering cluster representated by close cluster centre；
Using Euclidean distance quantization element with the distance of each cluster centre, formula is as follows:
In formula, distance (pi,kj) indicate element piWith cluster centre kjDistance,Indicate n-th of seat of i-th of element
Mark,Indicate that n-th of coordinate of j-th of cluster centre, N indicate the sample moment time interval T for including in 1 day1Quantity；
Step A3, the average coordinates of all elements in each clustering cluster are calculated using Euclidean distance, calculate each cluster according to this
Average coordinates in cluster with each cluster centre distance, if the average coordinates of each clustering cluster are respectively less than with the distance of each cluster centre
Equal to threshold valueCluster is completed, and current cluster result is obtained；If it exists the average coordinates of any clustering cluster with cluster centre away from
From greater than threshold valueUsing the average coordinates of each clustering cluster cluster centre brand new as each cluster, step A2 is gone to.
5. according to the method described in claim 4, it is characterized in that, threshold valueValue be 0.05.
6. according to the method described in claim 3, it is characterized in that, the detailed process of the step 3.2 are as follows:
Step B1 successively sums the power sequence of all electric power detection devices in block same in power training sample, obtains
With N3The corresponding N of a block3A block power sequence；
Step B2 successively selects the N in the block power sequence of a block by Zhou Tianci for b1A block power sequence segment, group
It is the power training subsample of b at a block day time；Wherein a=1,2 ..., N3, b=1,2 ..., 7；
Step B3 takes from a block day time to randomly choose 20 block power sequence segments in the power training subsample of b
Average value obtains the block power sequence average fragment that a block day time is b；Wherein, every in block power sequence average fragment
A sample moment corresponds to a power average value；
A block day time is divided into N in one day for b using block power sequence average fragment by step B44A period divides
Method are as follows: in block power sequence average fragment, if meeting latter sample moment power average value and current sample moment power
The absolute difference of average value is greater than 30% of the minimum value in two sample moment power average values, then current sample moment is
Critical sample moment, and using critical sample moment as the finish time of present period, when the latter sample of critical sample moment
It carves as at the beginning of subsequent period；Block power sequence average fragment end sample moment is unsatisfactory for critical sample moment item
Part；
Step B5 successively calculates the area a according to Time segments division as a result, being the power training subsample of b using a block day time
Subsample is trained for the electric load of b in block day time；
By the performance number of each sample moment in the c period and sample moment time interval T1The sum of product, as the c period
Power load charge values；Wherein, c=1,2 ..., N4, then 1 day N4The power load charge values composition a block day time of a period is 1 of b
Electric load sequence, time span N1Then correspondence obtains N for the power training subsample of a continuous cycle1A electric load sequence,
N1A electric load sequence composition a block day time is the electric load training subsample of b, successively takes b=1 by 7 days of week,
2 ..., the electric load training sample of 7 composition a blocks.
7. according to the method described in claim 3, it is characterized in that, time span N1Value are as follows: N1≥50。
8. the method according to claim 1, wherein sample moment time interval T1Value be 20 minutes.
9. the method according to claim 1, wherein the voltage and current historical data is by unwise karr
Graceful filtered data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910000269.2A CN109376972B (en) | 2019-01-01 | 2019-01-01 | Smart grid short-term power load prediction method based on block clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910000269.2A CN109376972B (en) | 2019-01-01 | 2019-01-01 | Smart grid short-term power load prediction method based on block clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109376972A true CN109376972A (en) | 2019-02-22 |
| CN109376972B CN109376972B (en) | 2021-07-27 |
Family
ID=65372287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910000269.2A Active CN109376972B (en) | 2019-01-01 | 2019-01-01 | Smart grid short-term power load prediction method based on block clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109376972B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110874671A (en) * | 2019-10-24 | 2020-03-10 | 腾讯科技（深圳）有限公司 | Power load prediction method and device of power distribution network and storage medium |
| CN111144611A (en) * | 2019-11-22 | 2020-05-12 | 国网辽宁省电力有限公司经济技术研究院 | Spatial load prediction method based on clustering and nonlinear autoregression |
| CN111260116A (en) * | 2020-01-10 | 2020-06-09 | 河南理工大学 | Time-interval refined short-term load prediction method based on BOA-SVR and fuzzy clustering |
| CN111612227A (en) * | 2020-05-12 | 2020-09-01 | 国网河北省电力有限公司电力科学研究院 | Load prediction method based on K-means clustering and bat optimization neural network |
| CN112348282A (en) * | 2020-11-25 | 2021-02-09 | 新智数字科技有限公司 | Load prediction method and device |
| CN113011630A (en) * | 2021-01-25 | 2021-06-22 | 国网浙江省电力有限公司杭州供电公司 | Method for short-term prediction of space load in zone time of big data power distribution network |
| CN113255973A (en) * | 2021-05-10 | 2021-08-13 | 曙光信息产业(北京)有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
| CN113379005A (en) * | 2021-08-12 | 2021-09-10 | 新风光电子科技股份有限公司 | Intelligent energy management system and method for power grid power equipment |
| CN115660233A (en) * | 2022-12-26 | 2023-01-31 | 尚特杰电力科技有限公司 | Photovoltaic power prediction method and device, electronic equipment and storage medium |
| CN116701973A (en) * | 2023-08-09 | 2023-09-05 | 无锡市明通动力工业有限公司 | Motor abnormal vibration detection early warning method based on big data |
| CN116805785A (en) * | 2023-08-17 | 2023-09-26 | 国网浙江省电力有限公司金华供电公司 | Power load hierarchy time sequence prediction method based on random clustering |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102999791A (en) * | 2012-11-23 | 2013-03-27 | 广东电网公司电力科学研究院 | Power load forecasting method based on customer segmentation in power industry |
| CN106127360A (en) * | 2016-06-06 | 2016-11-16 | 国网天津市电力公司 | A kind of multi-model load forecasting method analyzed based on user personality |
| CN108304978A (en) * | 2018-05-08 | 2018-07-20 | 国网江西省电力有限公司经济技术研究院 | A kind of mid-term Electric Power Load Forecast method based on data clusters theory |
| CN108876070A (en) * | 2018-09-25 | 2018-11-23 | 新智数字科技有限公司 | A kind of method and apparatus that Load Prediction In Power Systems are carried out based on neural network |
-
2019
- 2019-01-01 CN CN201910000269.2A patent/CN109376972B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102999791A (en) * | 2012-11-23 | 2013-03-27 | 广东电网公司电力科学研究院 | Power load forecasting method based on customer segmentation in power industry |
| CN106127360A (en) * | 2016-06-06 | 2016-11-16 | 国网天津市电力公司 | A kind of multi-model load forecasting method analyzed based on user personality |
| CN108304978A (en) * | 2018-05-08 | 2018-07-20 | 国网江西省电力有限公司经济技术研究院 | A kind of mid-term Electric Power Load Forecast method based on data clusters theory |
| CN108876070A (en) * | 2018-09-25 | 2018-11-23 | 新智数字科技有限公司 | A kind of method and apparatus that Load Prediction In Power Systems are carried out based on neural network |
Non-Patent Citations (2)
| Title |
|---|
| XISHUANG DONG 等: "Short-Term Load Forecasting in Smart Grid: A Combined CNN and K-Means Clustering Approach", 《2017 IEEE INTERNATIONAL CONFERENCE ON BIG DATA AND SMART COMPUTING (BIGCOMP)》 * |
| 蔡小华 等: "日电力负荷的分时段多模型组合预测", 《计算机系统应用》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110874671A (en) * | 2019-10-24 | 2020-03-10 | 腾讯科技（深圳）有限公司 | Power load prediction method and device of power distribution network and storage medium |
| CN110874671B (en) * | 2019-10-24 | 2021-03-16 | 腾讯科技（深圳）有限公司 | Power load prediction method and device of power distribution network and storage medium |
| CN111144611A (en) * | 2019-11-22 | 2020-05-12 | 国网辽宁省电力有限公司经济技术研究院 | Spatial load prediction method based on clustering and nonlinear autoregression |
| CN111260116A (en) * | 2020-01-10 | 2020-06-09 | 河南理工大学 | Time-interval refined short-term load prediction method based on BOA-SVR and fuzzy clustering |
| CN111612227A (en) * | 2020-05-12 | 2020-09-01 | 国网河北省电力有限公司电力科学研究院 | Load prediction method based on K-means clustering and bat optimization neural network |
| CN112348282A (en) * | 2020-11-25 | 2021-02-09 | 新智数字科技有限公司 | Load prediction method and device |
| CN113011630A (en) * | 2021-01-25 | 2021-06-22 | 国网浙江省电力有限公司杭州供电公司 | Method for short-term prediction of space load in zone time of big data power distribution network |
| CN113011630B (en) * | 2021-01-25 | 2024-01-23 | 国网浙江省电力有限公司杭州供电公司 | Short-term prediction method for space-time load of big data distribution network area |
| CN113255973A (en) * | 2021-05-10 | 2021-08-13 | 曙光信息产业(北京)有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
| CN113379005A (en) * | 2021-08-12 | 2021-09-10 | 新风光电子科技股份有限公司 | Intelligent energy management system and method for power grid power equipment |
| CN115660233A (en) * | 2022-12-26 | 2023-01-31 | 尚特杰电力科技有限公司 | Photovoltaic power prediction method and device, electronic equipment and storage medium |
| CN116701973A (en) * | 2023-08-09 | 2023-09-05 | 无锡市明通动力工业有限公司 | Motor abnormal vibration detection early warning method based on big data |
| CN116701973B (en) * | 2023-08-09 | 2023-10-03 | 无锡市明通动力工业有限公司 | Motor abnormal vibration detection early warning method based on big data |
| CN116805785A (en) * | 2023-08-17 | 2023-09-26 | 国网浙江省电力有限公司金华供电公司 | Power load hierarchy time sequence prediction method based on random clustering |
| CN116805785B (en) * | 2023-08-17 | 2023-11-28 | 国网浙江省电力有限公司金华供电公司 | Power load hierarchy time sequence prediction method based on random clustering |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109376972B (en) | 2021-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109376972A (en) | A kind of wisdom Power Network Short-Term Electric Load Forecasting method based on block cluster | |
| CN109508835A (en) | A kind of wisdom Power Network Short-Term Electric Load Forecasting method of integrated environment feedback | |
| Yang et al. | A hybrid forecasting approach applied in wind speed forecasting based on a data processing strategy and an optimized artificial intelligence algorithm | |
| Wang et al. | Research and application of a hybrid forecasting framework based on multi-objective optimization for electrical power system | |
| CN102663513B (en) | Utilize the wind power combined prediction modeling method of grey relational grade analysis | |
| CN111080032A (en) | Load prediction method based on Transformer structure | |
| Yu et al. | Switching ARIMA model based forecasting for traffic flow | |
| Lu et al. | A hybrid model based on convolutional neural network and long short-term memory for short-term load forecasting | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN111537938A (en) | Error short-time prediction method for electronic transformer based on intelligent algorithm | |
| CN105160423A (en) | Photovoltaic power generation prediction method based on Markov residual error correction | |
| CN104463356A (en) | Photovoltaic power generation power prediction method based on multi-dimension information artificial neural network algorithm | |
| CN111092451B (en) | Photovoltaic consumption early warning method based on power distribution network grid | |
| CN110535149A (en) | A kind of area Gong Biantai electric load three-phase imbalance prediction technique | |
| CN109636054A (en) | Solar energy power generating amount prediction technique based on classification and error combination prediction | |
| CN112307675B (en) | Neural network-based temperature-sensitive load separation identification method and system | |
| CN107293118A (en) | A kind of traffic speed motion interval Forecasting Approach for Short-term | |
| Wang et al. | Electric vehicle charging load clustering and load forecasting based on long short term memory neural network | |
| CN112990597A (en) | Ultra-short-term prediction method for industrial park factory electrical load | |
| CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
| CN114285086B (en) | Electrochemical energy storage power station energy scheduling method and system based on improved ant colony algorithm | |
| CN105279582B (en) | Super short-period wind power prediction technique based on dynamic correlation feature | |
| CN109934396A (en) | A kind of regional network for the load fining prediction technique based on two stages reduction | |
| CN109784562A (en) | A kind of wisdom grid power load forecasting method based on big data space-time cluster | |
| CN109615142A (en) | A kind of wind farm wind velocity combination forecasting method based on wavelet analysis |
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 |
Effective date of registration: 20240105 Address after: 430061 Room 26, Floor 8, Unit 1, Building 8, Donghu Chunshuli, No.166, Zhongbei Road, Guoguohu Street, Wuchang District, Wuhan City, Hubei Province Patentee after: Wuhan Pusidi Power Technology Co.,Ltd. Address before: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee before: Dragon totem Technology (Hefei) Co.,Ltd. Effective date of registration: 20240105 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Address before: Yuelu District City, Hunan province 410083 Changsha Lushan Road No. 932 Patentee before: CENTRAL SOUTH University |
|
| TR01 | Transfer of patent right |