CN112418476A - Ultra-short-term power load prediction method - Google Patents
Ultra-short-term power load prediction method Download PDFInfo
- Publication number
- CN112418476A CN112418476A CN201910786061.8A CN201910786061A CN112418476A CN 112418476 A CN112418476 A CN 112418476A CN 201910786061 A CN201910786061 A CN 201910786061A CN 112418476 A CN112418476 A CN 112418476A
- Authority
- CN
- China
- Prior art keywords
- data
- vector machine
- power load
- neural network
- model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 claims abstract description 36
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 238000010276 construction Methods 0.000 claims abstract description 8
- 230000008859 change Effects 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 11
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 230000008571 general function Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 230000005611 electricity Effects 0.000 claims description 2
- 238000013277 forecasting method Methods 0.000 claims 1
- 238000000605 extraction Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012360 testing method 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Biophysics (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Primary Health Care (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
The invention relates to the technical field of ultra-short-term power load prediction, in particular to an ultra-short-term power load prediction method, which comprises the following steps: s1, relevant vector machine-neural network feature engineering and construction: the relevance vector machine-neural network consists of a relevance vector machine and a neural network, and iterative learning is carried out by constructing sample data on the basis of an active relevance decision theory under the structure of a prior parameter, so that the posterior distribution of partial parameters tends to 0; and S2, acquisition of input data: collecting power load data and regional temperature change data; s3, preprocessing data; s4, characteristics are normalized, and the short-term power load prediction method provided by the invention is designed, so that the prediction precision is high, the noise error resistance and the generalization capability are good, and the requirements of a power grid on short-term and ultra-short-term power load prediction can be better met.
Description
Technical Field
The invention relates to the technical field of ultra-short-term power load prediction, in particular to an ultra-short-term power load prediction method.
Background
The current ultra-short term power load prediction has low prediction precision, poor noise error resistance and poor generalization capability, and can not well meet the requirements of a power grid on short term and ultra-short term power load prediction.
In summary, the present invention provides an ultra-short term power load prediction method to solve the existing problems.
Disclosure of Invention
The present invention is directed to a method for predicting an ultra-short term power load, so as to solve the problems in the background art.
In order to achieve the purpose, the invention provides the following technical scheme:
an ultra-short-term power load prediction method comprises the following steps:
s1, relevant vector machine-neural network feature engineering and construction: the relevance vector machine-neural network consists of a relevance vector machine and a neural network, and iterative learning is carried out by constructing sample data on the basis of an active relevance decision theory under the structure of a prior parameter, so that the posterior distribution of partial parameters tends to 0;
and S2, acquisition of input data: collecting power load data and regional temperature change data;
s3, preprocessing data;
s4, feature normalization;
s5, model training, namely, dividing the sample data into 80% of training samples and 20% of verification samples, and inputting the training samples into a relevance vector machine-neural network model for model training and relevant parameter adjustment;
s6, forecasting and verifying, namely inputting a verification sample into the trained model, outputting a corresponding forecasting result, and comparing the corresponding forecasting result with a real value to obtain the generalization ability evaluation of the model;
and S7, drawing a power load curve of a maximum deviation similarity criterion clustering algorithm according to the actual use condition and by combining the predicted value and the true value of the system, comparing the power load prediction conditions of different time periods and different areas, performing iterative optimization, analyzing according to the prediction result, performing hyper-parameter setting according to a Bayesian theory and an optimization theory, readjusting a correlation vector machine kernel function, and adjusting a neural network structure, a learning rate and an optimizer.
Preferably, the correlation vector machine in S1 is a sparse probability model, is a supervised learning method with wide application, and training is performed under a bayesian framework, the neural network in S1 is a supervised general function fitting method and the correlation vector machine fits the following formal equation:
preferably, the relevant vector machine-neural network feature engineering and construction in S1 specifically includes the following steps:
step 1, constructing a correlation vector machine, namely: constructing an RVM algorithm framework based on Bayesian theory, wherein the likelihood function of the data set can be expressed as follows under the assumption of positive distribution:
after derivation, the posterior distribution of weights can be rewritten as:
wherein the a-posteriori covariance matrix, ∑ ═ σ-2ΦTΦ+A)-1,A＝diag(α0，α1，...，αn-14) The posterior mean value is:
step 2, selecting a kernel function which is not limited by the Meixi theorem and can select any kernel function according to the characteristics of the sample;
step 3, initializing a penalty factor, and setting the penalty factor as an automatic assignment;
step 4, determining the input number and the output number of the vector machine according to the data type, and determining the sample form of the vector machine;
step 5, copying the constructed structure of the correlation vector machine model to obtain four prediction models with consistent structures and independent from each other;
step 6, transmitting the preprocessed data features into a first correlation vector machine model for training, wherein model labels are corresponding predicted load values;
step 7, performing wavelet decomposition on the preprocessed data to obtain three different frequency bands, respectively transmitting data characteristics into the last three relevant vector machine models for training, wherein model labels correspond to decomposition frequency band values of load decomposition;
step 8, connecting the four model prediction results with the original information to form multi-dimensional feature information, determining the input dimension of the neural network through the feature dimension, and taking the final prediction result as the output dimension;
step 9, setting a neural network hidden layer structure, including the number of network layers, the number of nodes in each layer and a network connection mode;
step 10, setting a parameter setting strategy, including determination of learning rate and selection of an optimizer;
step 11, determining a model anti-overfitting strategy, which comprises setting of a batch normalization layer, setting of a Dropout layer, a data enhancement scheme and the like;
and step 12, connecting the prediction result of the relevant vector machine with an input layer of the neural network to obtain a complete characteristic engineering structure.
Preferably, the specific steps in the acquisition of the input data in S2 are as follows:
step 13, constructing a label based on historical power load data, and classifying the data according to time sequence;
step 14, calculating a correlation coefficient table among data according to the time sequence of the data, and constructing a correlation coefficient matrix;
the maximum information coefficient equation is:
step 15, analyzing the correlation coefficient matrix, selecting historical correlation characteristics, and extracting historical influence characteristics;
step 16, determining the type of the sample and the input and output range according to the type of the influence characteristics;
step 17, constructing date marking characteristics according to date types, wherein the marking comprises time information and time additional information, and constructing current state influence marking characteristics;
step 18, time information;
step 19, marking the temperature characteristics as temperature characteristic dimensions;
and 20, finally combining the features together to form a final feature, and constructing a series of data sample sets { X } on all data in a sliding window modei，yi}。
Preferably, the data preprocessing in S3 includes the following specific steps:
step 21, performing wavelet filtering processing on the power load data to weaken the white noise influence in the data;
step 22, performing Kalman filtering processing on the power load data to weaken the influence of Gaussian white noise in the data;
and step 23, abnormal data are removed, short-time data distribution is counted in a sliding window mode, obvious data burrs are removed, and data are supplemented by a linear interpolation method.
Preferably, the characteristic normalization in S4 specifically includes the following steps:
step 24, selecting a proper obtaining range for the characteristic dimension of the obtained characteristic sample;
step 25, normalizing the characteristic quantity layer, and converting all data into the measuring range;
and step 23, performing characteristic dimension normalization on the whole data.
Preferably, the three different frequency bands in step 7 include high frequency, medium frequency and low frequency.
Preferably, the step 18 time information includes: electricity and units of minutes; the special time information includes whether the time is a weekend or a weekday, and whether the time is a holiday.
Compared with the prior art, the invention has the beneficial effects that:
according to the method, the preprocessing steps of wavelet transformation and feature selection are integrated, various types of relevant features such as historical state information, current state information and time information are considered for combined prediction, feature extraction is carried out by using a sliding window, a temperature prediction module is introduced, the predicted current temperature information is added into the feature of load prediction, the precision of model prediction is improved, and the model prediction is comprehensively fused into a relevant vector machine algorithm for training, so that the test is carried out on the New York Independent system operator, the very high prediction precision is achieved, the prediction average error RMSE fluctuates by about 1%, and further the short-term power load prediction method provided by the invention has the advantages of high prediction precision, good noise error resistance and generalization capability, and can better meet the requirements of a power grid on short-term and ultra-short-term power load prediction.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the protection scope of the present invention.
The invention provides a technical scheme that:
an ultra-short-term power load prediction method comprises the following steps:
s1, relevant vector machine-neural network feature engineering and construction: the relevance vector machine-neural network consists of a relevance vector machine and a neural network, and the posterior distribution of partial parameters tends to 0 by constructing sample data for iterative learning based on an active relevance decision theory under the structure of prior parameters;
and S2, acquisition of input data: collecting power load data and regional temperature change data;
s3, preprocessing data;
s4, feature normalization;
s5, model training, namely, dividing the sample data into 80% of training samples and 20% of verification samples, and inputting the training samples into a relevance vector machine-neural network model for model training and relevant parameter adjustment;
s6, forecasting and verifying, namely inputting a verification sample into the trained model, outputting a corresponding forecasting result, and comparing the corresponding forecasting result with a real value to obtain the generalization ability evaluation of the model;
and S7, drawing a power load curve of a maximum deviation similarity criterion clustering algorithm according to the actual use condition and by combining the predicted value and the true value of the system, comparing the power load prediction conditions of different time periods and different areas, performing iterative optimization, analyzing according to the prediction result, performing hyper-parameter setting according to a Bayesian theory and an optimization theory, readjusting a correlation vector machine kernel function, and adjusting a neural network structure, a learning rate and an optimizer.
The process flow of the invention is as follows:
s1, relevant vector machine-neural network feature engineering and construction: the relevance vector machine-neural network is composed of a relevance vector machine and a neural network, wherein the relevance vector machine is a sparse probability model and is a supervised learning method with wide application. The training is carried out under a Bayes framework, and iterative learning is carried out by constructing sample data based on an active relevance decision theory (automatic relevance determination) under the structure of prior parameters, so that the posterior distribution of partial parameters tends to 0 and is irrelevant to a predicted value. The neural network is a supervised general function fitting method, has good fitting capability to nonlinearity, and can remarkably improve the robustness and stability of the algorithm by constructing an integrated learning model through the neural network. The following formal equation was fitted by a correlation vector machine:
a) construction of a relevance vector machine first step: constructing an RVM algorithm framework based on Bayesian theory, wherein the likelihood function of the data set can be expressed as follows under the assumption of positive distribution:
after derivation, the posterior distribution of weights can be rewritten as:
wherein the posterior covariance matrix:
∑＝(σ-2ΦTΦ+A)-1，A＝diag(α0，α1，…，αn-14) The posterior mean value is
b) Selecting a kernel function which is not limited by the Meixi theorem and can select any kernel function according to the characteristics of the sample;
c) a penalty factor is initialized, and can be set to be automatically assigned;
d) determining the input number and the output number of the vector machine according to the data type, and determining the sample form of the vector machine;
e) copying the constructed structure of the correlation vector machine model to obtain four prediction models with consistent structures and independence;
f) and transmitting the preprocessed data features into a first correlation vector machine model for training, wherein the model label is a corresponding predicted load value.
g) The data characteristics of three different frequency bands (high frequency, medium frequency and low frequency) obtained after wavelet decomposition of the preprocessed data are respectively transmitted into the three subsequent relevant vector machine models for training, and model labels correspond to the decomposition frequency band values of load decomposition;
h) connecting the four model prediction results with the original information to form multi-dimensional characteristic information, determining the input dimension of the neural network through the characteristic dimension, and taking the final prediction result as the output dimension;
i) setting a neural network hidden layer structure, wherein the neural network hidden layer structure comprises the number of network layers, the number of nodes in each layer and a network connection mode;
j) setting a parameter setting strategy, including the determination of learning rate and the selection of an optimizer;
k) determining a model anti-overfitting strategy, which comprises setting of a batch normalization layer, setting of a Dropout layer, a data enhancement scheme and the like;
l) connecting the prediction result of the relevant vector machine with an input layer of a neural network to obtain a complete characteristic engineering structure;
and S2, acquisition of input data: and collecting power load data and regional temperature change data.
a) Constructing a label based on historical power load data, and classifying the data according to time sequence;
b) calculating a correlation coefficient table among data according to the time sequence of the data, and constructing a correlation coefficient matrix;
maximum information coefficient equation:
c) analyzing the correlation coefficient matrix, selecting historical correlation characteristics, and extracting historical influence characteristics;
d) determining the type and input/output range of the sample according to the type of the influence characteristics;
e) according to the date type, constructing date marking characteristics, wherein the marking comprises time information and time additional information, and constructing current state influence marking characteristics;
f) the time information includes 1. power consumption (in minutes) 2. special time information (including whether the time is weekend or working day, whether the time is holiday or holiday);
g) marking the temperature characteristics as temperature characteristic dimensions;
h) and finally, combining the features together to form a final feature, and constructing a series of data sample sets { X ] on all data in a sliding window modei，yi}。
S3, preprocessing data:
a) the wavelet filtering processing is carried out on the power load data, and the white noise influence in the data is weakened;
b) performing Kalman filtering processing on the power load data to weaken the influence of Gaussian white noise in the data;
c) and (4) abnormal data elimination, namely counting the short-time data distribution in a sliding window mode, eliminating obvious data burrs, and supplementing data by using a linear interpolation method.
S4, feature normalization;
a) selecting a proper obtaining range for the characteristic dimension of the obtained characteristic sample;
b) normalizing the characteristic quantity layer, and converting all data into the measuring range;
c) and the characteristic dimension normalization is carried out on the whole data, so that the huge influence brought by different data input formats is avoided.
S5, model training, namely dividing the Sample data into 80% of training samples (Train Sample) and 20% of verification samples (valid Sample), and inputting the training samples (Train Sample) into a Relevance Vector Machine (RVM) -Neural Network (NN) model for model training and relevant parameter adjustment;
s6, forecasting and verifying, namely inputting a verification sample into the trained model, outputting a corresponding forecasting result, and comparing the corresponding forecasting result with a real value to obtain the generalization ability evaluation of the model;
and S7, drawing a power load curve of a maximum deviation similarity criterion clustering algorithm according to the actual use condition and by combining the predicted value and the true value of the system, comparing the power load prediction conditions of different time periods and different areas, performing iterative optimization, analyzing according to the prediction result, performing hyper-parameter setting according to a Bayesian theory and an optimization theory, readjusting a correlation vector machine kernel function, and adjusting a neural network structure, a learning rate and an optimizer.
The process integrates the algorithm with the preprocessing steps of wavelet transformation and feature selection by design, takes historical state information into consideration, the current state information, the time information and other various types of relevant characteristics are combined and predicted, the characteristic extraction is carried out by utilizing a sliding window, and a temperature prediction module is introduced to add the predicted current temperature information into the load prediction characteristic, the short-term power load prediction method provided by the invention has the advantages of high prediction precision, good noise error resistance and generalization capability, and can better meet the requirements of a power grid on short-term and ultra-short-term power load prediction.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (8)
1. An ultra-short-term power load prediction method is characterized by comprising the following steps:
s1, relevant vector machine-neural network feature engineering and construction: the relevance vector machine-neural network consists of a relevance vector machine and a neural network, and iterative learning is carried out by constructing sample data on the basis of an active relevance decision theory under the structure of a prior parameter, so that the posterior distribution of partial parameters tends to 0;
and S2, acquisition of input data: collecting power load data and regional temperature change data;
s3, preprocessing data;
s4, feature normalization;
s5, model training, namely, dividing the sample data into 80% of training samples and 20% of verification samples, and inputting the training samples into a relevance vector machine-neural network model for model training and relevant parameter adjustment;
s6, forecasting and verifying, namely inputting a verification sample into the trained model, outputting a corresponding forecasting result, and comparing the corresponding forecasting result with a real value to obtain the generalization ability evaluation of the model;
and S7, drawing a power load curve of a maximum deviation similarity criterion clustering algorithm according to the actual use condition and by combining the predicted value and the true value of the system, comparing the power load prediction conditions of different time periods and different areas, performing iterative optimization, analyzing according to the prediction result, performing hyper-parameter setting according to a Bayesian theory and an optimization theory, readjusting a correlation vector machine kernel function, and adjusting a neural network structure, a learning rate and an optimizer.
2. The ultra-short term power load prediction method as claimed in claim 1, wherein the correlation vector machine in S1 is a sparse probability model, is a supervised learning method with wide usage, and training is performed under bayesian framework, the neural network in S1 is a supervised general function fitting method and the correlation vector machine fits the following form equation:
3. the method of claim 1, wherein the relevant vector machine-neural network feature engineering and construction in S1 comprises the following steps:
step 1, constructing a correlation vector machine, namely: constructing an RVM algorithm framework based on Bayesian theory, wherein the likelihood function of the data set can be expressed as follows under the assumption of positive distribution:
after derivation, the posterior distribution of weights can be rewritten as:
wherein the a-posteriori covariance matrix, ∑ ═ σ-2ΦTΦ+A)-1,A＝diag(α0，α1，...，αn-14) The posterior mean value is:
step 2, selecting a kernel function which is not limited by the Meixi theorem and can select any kernel function according to the characteristics of the sample;
step 3, initializing a penalty factor, and setting the penalty factor as an automatic assignment;
step 4, determining the input number and the output number of the vector machine according to the data type, and determining the sample form of the vector machine;
step 5, copying the constructed structure of the correlation vector machine model to obtain four prediction models with consistent structures and independent from each other;
step 6, transmitting the preprocessed data features into a first correlation vector machine model for training, wherein model labels are corresponding predicted load values;
step 7, performing wavelet decomposition on the preprocessed data to obtain three different frequency bands, respectively transmitting data characteristics into the last three relevant vector machine models for training, wherein model labels correspond to decomposition frequency band values of load decomposition;
step 8, connecting the four model prediction results with the original information to form multi-dimensional feature information, determining the input dimension of the neural network through the feature dimension, and taking the final prediction result as the output dimension;
step 9, setting a neural network hidden layer structure, including the number of network layers, the number of nodes in each layer and a network connection mode;
step 10, setting a parameter setting strategy, including determination of learning rate and selection of an optimizer;
step 11, determining a model anti-overfitting strategy, which comprises setting of a batch normalization layer, setting of a Dropout layer, a data enhancement scheme and the like;
and step 12, connecting the prediction result of the relevant vector machine with an input layer of the neural network to obtain a complete characteristic engineering structure.
4. The method of claim 1, wherein the step of collecting the input data in S2 is as follows:
step 13, constructing a label based on historical power load data, and classifying the data according to time sequence;
step 14, calculating a correlation coefficient table among data according to the time sequence of the data, and constructing a correlation coefficient matrix;
the maximum information coefficient equation is:
step 15, analyzing the correlation coefficient matrix, selecting historical correlation characteristics, and extracting historical influence characteristics;
step 16, determining the type of the sample and the input and output range according to the type of the influence characteristics;
step 17, constructing date marking characteristics according to date types, wherein the marking comprises time information and time additional information, and constructing current state influence marking characteristics;
step 18, time information;
step 19, marking the temperature characteristics as temperature characteristic dimensions;
and 20, finally combining the features together to form a final feature, and constructing a series of data sample sets { X } on all data in a sliding window modei，yi}。
5. The method of claim 1, wherein the data preprocessing of S3 comprises the following steps:
step 21, performing wavelet filtering processing on the power load data to weaken the white noise influence in the data;
step 22, performing Kalman filtering processing on the power load data to weaken the influence of Gaussian white noise in the data;
and step 23, abnormal data are removed, short-time data distribution is counted in a sliding window mode, obvious data burrs are removed, and data are supplemented by a linear interpolation method.
6. The method of claim 1, wherein the step of normalizing the characteristics in S4 comprises:
step 24, selecting a proper obtaining range for the characteristic dimension of the obtained characteristic sample;
step 25, normalizing the characteristic quantity layer, and converting all data into the measuring range;
and step 23, performing characteristic dimension normalization on the whole data.
7. The ultra-short term power load forecasting method as claimed in claim 3, wherein the three different frequency bands in step 7 include high frequency, medium frequency and low frequency.
8. The ultra-short term power load prediction method as claimed in claim 4, wherein the step 18 time information comprises: electricity and units of minutes; the special time information includes whether the time is a weekend or a weekday, and whether the time is a holiday.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910786061.8A CN112418476A (en) | 2019-08-23 | 2019-08-23 | Ultra-short-term power load prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910786061.8A CN112418476A (en) | 2019-08-23 | 2019-08-23 | Ultra-short-term power load prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112418476A true CN112418476A (en) | 2021-02-26 |
Family
ID=74779074
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910786061.8A Pending CN112418476A (en) | 2019-08-23 | 2019-08-23 | Ultra-short-term power load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112418476A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113107626A (en) * | 2021-05-25 | 2021-07-13 | 浙大城市学院 | Combined cycle generator set load prediction method based on multivariable LSTM |
| CN113313049A (en) * | 2021-06-11 | 2021-08-27 | 北京百度网讯科技有限公司 | Method, device, equipment, storage medium and computer program product for determining hyper-parameters |
| CN113570207A (en) * | 2021-07-09 | 2021-10-29 | 北京淇瑀信息科技有限公司 | User policy distribution method and device and electronic equipment |
| CN115037767A (en) * | 2022-06-23 | 2022-09-09 | 辽宁宏成供电有限公司 | Data monitoring integrated platform |
| CN117571184A (en) * | 2024-01-17 | 2024-02-20 | 四川省公路规划勘察设计研究院有限公司 | Bridge structure cable force identification method and equipment based on sliding window and cluster analysis |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184453A (en) * | 2011-05-16 | 2011-09-14 | 上海电气集团股份有限公司 | Wind power combination predicting method based on fuzzy neural network and support vector machine |
| CN104992248A (en) * | 2015-07-07 | 2015-10-21 | 中山大学 | Microgrid photovoltaic power station generating capacity combined forecasting method |
| CN108171369A (en) * | 2017-12-21 | 2018-06-15 | 国家电网公司 | Short term combination forecasting method based on customer electricity differentiation characteristic |
| CN108596242A (en) * | 2018-04-20 | 2018-09-28 | 浙江大学 | Power grid meteorology load forecasting method based on wavelet neural network and support vector machines |
| CN109447332A (en) * | 2018-10-18 | 2019-03-08 | 国网天津市电力公司电力科学研究院 | A kind of Middle-long Electric Power Load Forecast method suitable for S type load curve |
| CN109670628A (en) * | 2018-11-09 | 2019-04-23 | 昆明理工大学 | A kind of micro-grid load prediction technique of the neural network based on LSSVM |
| CN109685265A (en) * | 2018-12-21 | 2019-04-26 | 积成电子股份有限公司 | A kind of prediction technique of power-system short-term electric load |
| CN110135617A (en) * | 2019-03-31 | 2019-08-16 | 中南林业科技大学 | A kind of prediction technique of improvement short term under meteorological factor influence based on part throttle characteristics |
-
2019
- 2019-08-23 CN CN201910786061.8A patent/CN112418476A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184453A (en) * | 2011-05-16 | 2011-09-14 | 上海电气集团股份有限公司 | Wind power combination predicting method based on fuzzy neural network and support vector machine |
| CN104992248A (en) * | 2015-07-07 | 2015-10-21 | 中山大学 | Microgrid photovoltaic power station generating capacity combined forecasting method |
| CN108171369A (en) * | 2017-12-21 | 2018-06-15 | 国家电网公司 | Short term combination forecasting method based on customer electricity differentiation characteristic |
| CN108596242A (en) * | 2018-04-20 | 2018-09-28 | 浙江大学 | Power grid meteorology load forecasting method based on wavelet neural network and support vector machines |
| CN109447332A (en) * | 2018-10-18 | 2019-03-08 | 国网天津市电力公司电力科学研究院 | A kind of Middle-long Electric Power Load Forecast method suitable for S type load curve |
| CN109670628A (en) * | 2018-11-09 | 2019-04-23 | 昆明理工大学 | A kind of micro-grid load prediction technique of the neural network based on LSSVM |
| CN109685265A (en) * | 2018-12-21 | 2019-04-26 | 积成电子股份有限公司 | A kind of prediction technique of power-system short-term electric load |
| CN110135617A (en) * | 2019-03-31 | 2019-08-16 | 中南林业科技大学 | A kind of prediction technique of improvement short term under meteorological factor influence based on part throttle characteristics |
Non-Patent Citations (1)
| Title |
|---|
| 哎呦不错的温JAY: "机器学习-极大似然估计详解", 《HTTPS://BLOG.CSDN.NET/WENQIWENQI123/ARTICLE/DETAILS/80158165》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113107626A (en) * | 2021-05-25 | 2021-07-13 | 浙大城市学院 | Combined cycle generator set load prediction method based on multivariable LSTM |
| CN113107626B (en) * | 2021-05-25 | 2024-02-02 | 浙大城市学院 | Load prediction method of combined cycle generator set based on multivariable LSTM |
| CN113313049A (en) * | 2021-06-11 | 2021-08-27 | 北京百度网讯科技有限公司 | Method, device, equipment, storage medium and computer program product for determining hyper-parameters |
| CN113570207A (en) * | 2021-07-09 | 2021-10-29 | 北京淇瑀信息科技有限公司 | User policy distribution method and device and electronic equipment |
| CN113570207B (en) * | 2021-07-09 | 2024-04-26 | 北京淇瑀信息科技有限公司 | User policy allocation method and device and electronic equipment |
| CN115037767A (en) * | 2022-06-23 | 2022-09-09 | 辽宁宏成供电有限公司 | Data monitoring integrated platform |
| CN117571184A (en) * | 2024-01-17 | 2024-02-20 | 四川省公路规划勘察设计研究院有限公司 | Bridge structure cable force identification method and equipment based on sliding window and cluster analysis |
| CN117571184B (en) * | 2024-01-17 | 2024-03-19 | 四川省公路规划勘察设计研究院有限公司 | Bridge structure cable force identification method and equipment based on sliding window and cluster analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112418476A (en) | Ultra-short-term power load prediction method | |
| CN106650767B (en) | Flood forecasting method based on cluster analysis and real-time correction | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN111178611B (en) | Method for predicting daily electric quantity | |
| CN111401599B (en) | Water level prediction method based on similarity search and LSTM neural network | |
| CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
| CN111898839B (en) | Importance degree classification method and device for power users | |
| CN111369045A (en) | Method for predicting short-term photovoltaic power generation power | |
| CN113537469B (en) | Urban water demand prediction method based on LSTM network and Attention mechanism | |
| CN111460001B (en) | Power distribution network theoretical line loss rate evaluation method and system | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN106778838A (en) | A kind of method for predicting air quality | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN110738232A (en) | grid voltage out-of-limit cause diagnosis method based on data mining technology | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| Wang et al. | Big data analytics for price forecasting in smart grids | |
| CN115660182A (en) | Photovoltaic output prediction method based on maximum expected sample weighted neural network model | |
| CN116776209A (en) | Method, system, equipment and medium for identifying operation state of gateway metering device | |
| CN116070458A (en) | New wind power plant scene generation method based on RAC-GAN | |
| CN114004530B (en) | Enterprise electric power credit modeling method and system based on ordering support vector machine | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| CN115423145A (en) | Photovoltaic power prediction method based on Kmeans-VMD-WT-LSTM method | |
| CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
| CN111353523A (en) | Method for classifying railway customers | |
| CN117688367B (en) | Wind power generation ultra-short term power prediction method and device based on instant learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210226 |