CN113505923A - Regional power grid short-term load prediction method and system - Google Patents
Regional power grid short-term load prediction method and system Download PDFInfo
- Publication number
- CN113505923A CN113505923A CN202110772002.2A CN202110772002A CN113505923A CN 113505923 A CN113505923 A CN 113505923A CN 202110772002 A CN202110772002 A CN 202110772002A CN 113505923 A CN113505923 A CN 113505923A
- Authority
- CN
- China
- Prior art keywords
- day
- load
- weather
- meteorological
- days
- 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
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000007781 pre-processing Methods 0.000 claims abstract description 19
- 238000013135 deep learning Methods 0.000 claims abstract description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 10
- 230000002159 abnormal effect Effects 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 238000012935 Averaging Methods 0.000 claims description 6
- 230000009193 crawling Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000001556 precipitation Methods 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 4
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 230000002354 daily effect Effects 0.000 description 24
- 230000000694 effects Effects 0.000 description 14
- 238000004364 calculation method Methods 0.000 description 12
- 230000008859 change Effects 0.000 description 9
- 239000011159 matrix material Substances 0.000 description 7
- 230000000877 morphologic effect Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000002779 inactivation Effects 0.000 description 2
- 230000001965 increasing effect Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
Images
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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A30/00—Adapting or protecting infrastructure or their operation
-
- 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)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a method and a system for predicting short-term load of a regional power grid. The method comprises the following steps: the network terminal collects the load prediction related historical multi-element characteristic information data, the network crawls future weather prediction broadcast information, and data preprocessing is carried out on the collected characteristic information; fine-grained meteorological information, and selecting a meteorological virtual similar day of a day to be detected; selecting a virtual similar day of the daily load to be predicted; and determining a trained short-term load prediction model based on historical characteristic information data and a deep learning theory, and further realizing prediction of future loads. The invention can improve the accuracy of short-term load prediction.
Description
Technical Field
The invention relates to the field of short-term load prediction of power systems, in particular to a regional power grid short-term load prediction method and system.
Background
With the development of economy and the grid connection of large-scale distributed energy, the weather sensitive load base number of a regional power grid is continuously increased, the daily load peak-valley difference is continuously enlarged, and the local change condition of the load is more random and complex. How to scientifically and effectively combine fine-grained meteorological feature data and deeply mine the relation between related factors and local load change in short-term load prediction is a necessary direction for further improving the load prediction precision and realizing the refined work management of the load prediction.
The conventional research on the short-term load prediction of the regional power grid can be generally divided into two steps: firstly, selecting proper similar days; and secondly, based on the similar days, performing prediction modeling on the days to be predicted by combining various characteristic components. In the past, similar day selection is performed only for loads, and actually, the similar day selection is not limited to the loads, but also can be expanded to characteristic similar days with parameters such as weather as objects. In the actual load prediction work, only coarse-grained meteorological prediction data can be used as characteristic parameters to participate in prediction modeling on the day to be predicted, and the data have larger deviation compared with real meteorological sequence data, so that the prediction effect is greatly influenced. Therefore, how to make the coarse-grained meteorological data fine-grained so as to enable the coarse-grained meteorological data to correspond to the load prediction curves one by one is one of the problems that need to be solved in the short-term load prediction.
For the existing load similar days, most current load similar day selection methods still use the coarse-grained day characteristic meteorology as the similar day selection characteristic quantity, and are difficult to map to the local change of the load. In addition, most methods select load similar days from historical days, and the historical days and the similar days are often over-locally similar in characteristics, so that the effect of load prediction is greatly influenced. Therefore, trying to construct and select a load similarity day with higher similarity is a necessary way to improve the short-term load prediction accuracy.
Disclosure of Invention
The invention aims to provide a method and a system for predicting short-term load of a regional power grid, which can improve the accuracy of short-term load prediction.
In order to achieve the purpose, the invention provides the following scheme:
a method for predicting short-term load of a regional power grid comprises the following steps:
acquiring multi-element characteristic information data of a historical day and weather forecast broadcasting information of a day to be measured; the historical multivariate characteristic information data comprises: load of 96 points of a historical day, date type and fine-grained weather of the historical day; the date types include: weekdays, weekends, and holidays; the fine-grained meteorological data of the historical day comprises: fine-grained real-time temperature, humidity and wind speed data; the weather forecast broadcasting information of the day to be measured is obtained through network crawling; weather prediction broadcast information of the day to be measured comprises: highest and lowest temperature of the day, highest and lowest humidity of the day, wind direction, wind speed, air quality, precipitation and weather conditions;
preprocessing the multivariate characteristic information data of the historical days and the weather forecast broadcasting information of the days to be detected; the pretreatment comprises the following steps: completing missing values, correcting abnormal values, dividing four seasons by a temperature-averaging sliding method, carrying out LabelEncorder coding processing, carrying out standardization processing and calculating a comprehensive meteorological index;
determining weather historical similar days by adopting a grey correlation algorithm according to the weather forecast broadcast information of the preprocessed to-be-detected day and the multivariate characteristic information data of the preprocessed historical days; the weather history similar days are the first three types of history days with the highest correlation degree with the multi-type coarse-grained day characteristic weather data of the day to be predicted;
determining a fine-grained weather sequence of the virtual weather similar day corresponding to the day to be tested according to the association degree corresponding to the weather history similar day and the multi-element characteristic information data of the history day corresponding to the weather history similar day;
determining load similarity days by adopting a weighted gray correlation algorithm based on MIC weighting according to a fine-grained weather sequence of a virtual weather similarity day corresponding to a day to be detected and the preprocessed multivariate characteristic information data of a historical day;
determining a load sequence of the virtual load similar day corresponding to the day to be tested according to the similarity degree corresponding to the load similar day and the multi-element characteristic information data of the historical day corresponding to the meteorological historical similar day;
determining a trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested and the fine-grained meteorological data of the day to be tested; the trained short-term load prediction model takes fine-grained meteorological data or a fine-grained meteorological sequence as input and takes the load of each time point as output;
acquiring weather forecast broadcasting information of a future day, and preprocessing the weather forecast broadcasting information of the future day;
determining a fine-grained weather sequence of a virtual weather similar day corresponding to the future day according to the preprocessed weather prediction broadcast information; and determining the load of each time point of the future day by adopting a trained short-term load prediction model according to the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the future day.
Optionally, the preprocessing the multivariate characteristic information data of the historical day and the weather forecast report information of the day to be measured specifically includes:
performing missing value completion and abnormal value correction on the multivariate characteristic information data of the historical days and the weather prediction broadcast information of the days to be detected;
dividing the processed multi-element characteristic information data of the historical days and the weather forecast broadcasting information of the days to be measured in a sliding manner by adopting a temperature-equalizing method to obtain divided data;
converting mutually independent class labels into single operated numbers by adopting a LabelEncorder coding function according to the divided data;
carrying out normalization processing on the converted data;
and calculating the comprehensive meteorological index according to the data after the normalization processing.
Optionally, the determining, according to the association degree corresponding to the weather history similar day and the multi-feature information data of the history day corresponding to the weather history similar day, the fine-grained weather sequence of the virtual weather similar day corresponding to the day to be measured specifically includes:
determining the weight corresponding to the similar days of the weather history according to the relevance corresponding to the similar days of the weather history;
and performing weighted combination on the multivariate characteristic information data of the corresponding historical days according to the corresponding weights of the different meteorological historical similar days, and determining the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the day to be measured.
Optionally, the determining the load similarity day by using a weighted gray relevance algorithm based on MIC weighting according to the fine-grained weather sequence of the virtual weather similarity day corresponding to the day to be measured and the preprocessed multivariate characteristic information data of the historical day specifically includes:
using formulas
Determining the degree of non-linear correlation between the load and the meteorological features;
using formulas
K-th real-time meteorological feature f of grey-fixed correlation degreekThe matched weights;
using formulas
Determining the weighted gray correlation degree of the two characteristic quantity sequences;
wherein the content of the first and second substances,
p (l, f) is the joint probability between the daily load sequence l and the real-time meteorological features f, p (l) is the probability of the daily load sequence l, p (f) is the probability of the real-time meteorological features f, a, B are the number of intervals divided in the direction of l, f, B is a variable, wkFor the kth real-time meteorological feature fkThe matched weight, γ (x)i,xi-h) Is xiAnd xi-hTwo feature quantity sequences weight grey correlation degree xii-bAnd (r) is the gray correlation coefficient corresponding to each characteristic factor.
Optionally, the determining, according to the similarity degree association degree corresponding to the load similarity day and the multivariate characteristic information data of the historical day corresponding to the meteorological historical similarity day, a load sequence of the virtual load similarity day corresponding to the day to be measured, and then further includes:
extracting a fine-grained meteorological sequence corresponding to the virtual load similarity day;
respectively calculating the association degree between the fine-grained meteorological sequence of the virtual load similar day and the fine-grained meteorological sequence of the day to be predicted according to a grey association degree algorithm and the association degree between the fine-grained meteorological sequence of the load similar day with the largest association degree and the fine-grained meteorological sequence of the day to be predicted;
judging the degree of association of the two;
if the association degree of the fine-grained meteorological sequence on the virtual load similar day is large, the load on the day is still kept as the load on the virtual load similar day; otherwise, the load with the highest relevance degree is replaced by the load with the similar day.
Optionally, the determining a trained short-term load prediction model based on deep learning according to the fine-grained weather sequences of the virtual weather similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested, and the fine-grained weather data of the day to be tested specifically includes:
dividing a data set formed by the fine-grained meteorological sequences of the virtual meteorological similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested and the fine-grained meteorological data of the day to be tested into a training set and a verification set according to a set proportion;
and constructing input and output of the short-term load prediction model from the training set and the verification set by adopting a sliding window method.
Optionally, the determining a trained short-term load prediction model based on deep learning according to the fine-grained weather sequences of the virtual weather similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested, and the fine-grained weather data of the day to be tested also includes:
and constructing a short-term load prediction model according to the convolutional neural network CNN, the gated cyclic unit GRU and the time attention mechanism.
A regional power grid short term load prediction system, comprising:
the data acquisition module is used for acquiring multi-element characteristic information data of historical days and weather prediction broadcast information of days to be measured; the historical multivariate characteristic information data comprises: load of 96 points of a historical day, date type and fine-grained weather of the historical day; the date types include: weekdays, weekends, and holidays; the fine-grained meteorological data of the historical day comprises: fine-grained real-time temperature, humidity and wind speed data; the weather forecast broadcasting information of the day to be measured is obtained through network crawling; weather prediction broadcast information of the day to be measured comprises: highest and lowest temperature of the day, highest and lowest humidity of the day, wind direction, wind speed, air quality, precipitation and weather conditions;
the preprocessing module is used for preprocessing the multivariate characteristic information data of the historical days and the weather forecast broadcasting information of the days to be detected; the pretreatment comprises the following steps: completing missing values, correcting abnormal values, dividing four seasons by a temperature-averaging sliding method, carrying out LabelEncorder coding processing, carrying out standardization processing and calculating a comprehensive meteorological index;
the weather history similar day determining module is used for determining weather history similar days by adopting a grey correlation algorithm according to weather forecast broadcast information of the preprocessed to-be-detected day and the multi-element characteristic information data of the preprocessed historical days; the weather history similar days are the first three types of history days with the highest correlation degree with the multi-type coarse-grained day characteristic weather data of the day to be predicted;
the fine-grained weather sequence determination module of the virtual weather similar day is used for determining the fine-grained weather sequence of the virtual weather similar day corresponding to the day to be detected according to the correlation degree corresponding to the weather history similar day and the multi-element characteristic information data of the history day corresponding to the weather history similar day;
the load similarity day determining module is used for determining the load similarity day by adopting a weighted gray correlation algorithm based on MIC weighting according to a fine-grained weather sequence of the virtual weather similarity day corresponding to the day to be detected and the preprocessed multivariate characteristic information data of the historical day;
the load sequence determination module of the virtual load similar day is used for determining the load sequence of the virtual load similar day corresponding to the day to be tested according to the similarity correlation degree corresponding to the load similar day and the multi-element characteristic information data of the historical day corresponding to the meteorological historical similar day;
the trained short-term load prediction model determining module is used for determining a trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days, the load sequences of the virtual load similar days, the 96-point loads of the days to be tested and the fine-grained meteorological data of the days to be tested corresponding to all the days to be tested; the trained short-term load prediction model takes fine-grained meteorological data or a fine-grained meteorological sequence as input and takes the load of each time point as output;
the data acquisition and preprocessing module is used for acquiring the weather forecast broadcasting information of the future day and preprocessing the weather forecast broadcasting information of the future day;
the prediction module is used for determining a fine-grained weather sequence of the virtual weather similar day corresponding to the future day according to the preprocessed weather prediction broadcast information; and determining the load of each time point of the future day by adopting a trained short-term load prediction model according to the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the future day.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the regional power grid short-term load forecasting method and system provided by the invention, a meteorological virtual similar day selection algorithm is constructed from the characteristic meteorological data of the coarse-grained days reported by meteorological forecasting, fine-grained real-time meteorological data matched with a daily load curve are obtained, and complete and reliable data support is provided for the characteristic between the subsequent deep mining of the meteorological load and the local load change. Based on the sensitive response of human bodies to weather, original single weather factors such as temperature, humidity and the like are converted into comprehensive weather factors formed by coupling a plurality of single weather factors, the influence characteristic information of loads is widened, and the potential relation between weather and loads is favorably and deeply excavated. And (3) quantizing the nonlinear correlation degree between the load and the fine-grained meteorological features by adopting a maximum information coefficient MIC, taking fine-grained real-time meteorological data as input of load similar day selection, and selecting by using a weighted gray correlation algorithm weighted by the MIC and a correlation weighting method to obtain a load virtual similar day curve. The method can integrate different local similar characteristics of the historical similar daily load sequences on the basis of keeping the daily load sequence change characteristics, thereby obtaining higher similarity compared with the historical load similar days. According to the method, coarse-grained meteorological prediction data are refined in granularity, and the relation between fine-grained meteorological factors and local load change is deeply excavated, so that the short-term load prediction accuracy in the day-ahead is further improved, and reliable decision guidance is provided for load prediction refined work management
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a schematic flow chart of a method for predicting short-term load of a regional power grid according to the present invention;
FIG. 2 is a schematic structural diagram of a DA-LSTPNet short-term load prediction model;
FIG. 3 is a graph showing comparison results of curves of similar days 2017-08-28;
FIG. 4 is a graph showing a comparison of predicted result curves for typical summer days before and after a weather-like day;
fig. 5 is a schematic structural diagram of a system for predicting short-term load of a regional power grid according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in 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 of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for predicting short-term load of a regional power grid, which can improve the accuracy of short-term load prediction.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a schematic flow chart of a method for predicting a short-term load of a regional power grid provided by the present invention, and as shown in fig. 1, the method for predicting a short-term load of a regional power grid provided by the present invention includes:
s101, acquiring multi-element characteristic information data of a historical day and weather forecast broadcasting information of a day to be measured; the historical multivariate characteristic information data comprises: load of 96 points of a historical day, date type and fine-grained weather of the historical day; the date types include: weekdays, weekends, and holidays; the fine-grained meteorological data of the historical day comprises: fine-grained real-time temperature, humidity and wind speed data; the weather forecast broadcasting information of the day to be measured is obtained through network crawling; weather prediction broadcast information of the day to be measured comprises: highest and lowest temperature of the day, highest and lowest humidity of the day, wind direction, wind speed, air quality, precipitation and weather conditions;
s102, preprocessing the multivariate characteristic information data of the historical days and the weather forecast broadcasting information of the days to be detected; the pretreatment comprises the following steps: completing missing values, correcting abnormal values, dividing four seasons by a temperature-averaging sliding method, carrying out LabelEncorder coding processing, carrying out standardization processing and calculating a comprehensive meteorological index;
s102 specifically comprises the following steps:
performing missing value completion and abnormal value correction on the multivariate characteristic information data of the historical days and the weather prediction broadcast information of the days to be detected;
dividing the processed multi-element characteristic information data of the historical days and the weather forecast broadcasting information of the days to be measured in a sliding manner by adopting a temperature-equalizing method to obtain divided data;
in meteorology, five days are taken as one time, the moving average temperature of five days is taken as the waiting average temperature, and the specific calculation formula is as follows:
Tave＝(Td-4,Td-3,Td-2,Td-1,Td)/5；
Tavefor waiting for the mean temperature, TdTo calculate the average temperature of the day of the average temperature.
The calculated temperature is divided into four seasons each year with 15 ℃ and 23 ℃ as boundaries, and the division mode is shown in table 1.
TABLE 1
Dividing the calculated average temperature into four seasons of a year by taking 15 ℃ and 23 ℃ as boundaries, and considering that spring comes when the average temperature rises to more than 15 ℃ but is less than 23 ℃; when the average temperature continuously rises and exceeds 23 ℃, the summer comes; when the average temperature is reduced to below 23 ℃ and above 15 ℃, the autumn comes; and finally, if the average temperature is continuously reduced to below 15 ℃, entering winter.
Converting mutually independent class labels into single operated numbers by adopting a LabelEncorder coding function according to the divided data; on the basis of avoiding increasing data dimension, the LabeleEncorder coding function of the sklern machine learning package directly converts independent category labels such as seasons, week types and the like into single numbers available for operation.
In order to eliminate dimension difference caused by different data types and ensure that various gradient problems do not occur in the backward propagation process of the prediction model, the converted data is subjected to normalization processing; normalized to (0,1), the calculation formula is:
in the formula: x is data to be normalized in the sample; x is the number of*The normalized data is obtained; x is the number ofmaxAnd xminThe maximum and minimum values in the sample data.
And calculating the comprehensive meteorological index according to the data after the normalization processing.
The comprehensive weather is obtained by interweaving and synthesizing various basic weather indexes, and can effectively represent the sensitive reaction of a human body to external weather. The four comprehensive meteorology obtained by calculation are respectively: the feeling of excess temperature, warm and humid, cold and humid index, and the comfort of human body. The calculation method is as follows:
effective Temperature (ET):
temperature Humidity Index (THI):
THI＝1.8T+32-0.55(1-W)(1.8T-26)；
cold wetness Index (Chillness Humidity Index, CHI):
human Comfort (Comfort Index, CI):
wherein T is temperature (DEG C), W is relative humidity (%), and V is wind speed (m/s).
S103, determining weather history similar days by adopting a grey correlation algorithm according to the weather forecast broadcast information of the preprocessed to-be-detected day and the multi-element characteristic information data of the preprocessed historical days; the weather history similar days are the first three types of history days with the highest correlation degree with the multi-type coarse-grained day characteristic weather data of the day to be predicted;
the daily multi-day characteristic weather is constructed into a day characteristic sequence vector required by selecting weather similar days, the day characteristic weather factors contained in the vector are specifically shown in table 2, and the quantification of partial factors is shown in table 3.
TABLE 2
TABLE 3
Selecting weather history similar days for the days to be predicted of the test set by adopting the grey correlation degree, wherein a specific calculation formula is as follows:
xi-h＝{xi-h(k) n, and xi＝{xi(k) And k is 1,2.. n is a weather day characteristic sequence of the historical day and the day to be predicted respectively. Wherein n is the number of the included meteorological day characteristic factors, and is taken by referring to the table 2A value of 11; selecting the range of the target day within 30 days before, so that h is 30; xii-h(k) The grey correlation value between the historical day and the day to be predicted.
And finally, taking the three types of historical days with the highest correlation corresponding to each day to be predicted as weather similar days. The real-time weather sequence matrix of the three types of historical weather similar days is as follows:
wherein f isfir,fsec,fthiThe data base station respectively represents the optimal, 2 nd optimal and 3 rd optimal historical similar daily load sequences, wherein s is 0, 1.. 671 represents real-time weather under each historical daily record, and comprises daily real-time temperature, humidity, wind speed and four types of comprehensive weather indexes, and m is the day number to be predicted.
S104, determining a fine-grained weather sequence of the virtual weather similar day corresponding to the day to be tested according to the association degree corresponding to the weather history similar day and the multi-element characteristic information data of the history day corresponding to the weather history similar day;
s104 specifically comprises the following steps:
determining the weight corresponding to the similar days of the weather history according to the relevance corresponding to the similar days of the weather history;
the gray correlation degrees respectively corresponding to the 3 types of historical weather similar days are selected as
Taking the percentage of gray correlation value as the weight corresponding to the correlation
The calculation method is as follows:
and performing weighted combination on the multivariate characteristic information data of the corresponding historical days according to the corresponding weights of the different meteorological historical similar days, and determining the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the day to be measured.
Finally, each virtual weather similar day real-time weather sequence corresponding to the day to be predicted
The calculation method is as follows:
s105, determining load similarity days by adopting a weighted gray correlation algorithm based on MIC weighting according to a fine-grained weather sequence of a virtual weather similarity day corresponding to the to-be-detected day and the preprocessed multivariate characteristic information data of the historical days;
and (3) introducing more specific characteristics such as 96-point fine-grained real-time meteorological characteristic indexes (including real-time comprehensive meteorological) and the like every day to construct a daily real-time characteristic quantity sequence required by selecting load similar days. The specific characteristic factors contained in the selected characteristic vector for determining the load similarity days are shown in the following table 4.
TABLE 4
S105 specifically comprises the following steps:
before selecting on the day with similar load, a plurality of meteorological features with different influence degrees need to be endowed with matched weights, and the weight distribution is usually derived from the percentage of the absolute value of the correlation characteristic coefficient. In order to fully characterize the degree of nonlinear correlation between the load and the meteorological features, a Maximum Information Coefficient (MIC) is introduced to perform correlation analysis on the load and the meteorological features.
Using formulas
Determining the degree of non-linear correlation between the load and the meteorological features;
using formulas
Determining kth real-time meteorological feature f of grey correlation degreekThe matched weights; as shown in table 5, if the characteristic factor is a real-time meteorological factor, the weight value of each real-time meteorological factor is 1/96 corresponding to the calculated weight value.
And only selecting load similar days from the historical days with the same day type as the days to be predicted, wherein the selection range is 20 historical days of the same type before the days to be predicted. Respectively setting day real-time characteristic quantity sequences y of days to be predicted and historical days of the same typei＝{yi(r), r ═ 1,2.. p } and yi-h＝{yi-b(r), r 1,2.. p } such as, wherein p is the number of real-time feature factors, and b is the search range of historical days of the same type; using formulas
Determining the weighted gray correlation degree of the two characteristic quantity sequences;
wherein the content of the first and second substances,
ρ is the resolution factor, and the present invention is set to 0.5.
Wherein the content of the first and second substances,
p (l, f) is the joint probability between the daily load sequence l and the real-time meteorological features f, p (l) is the probability of the daily load sequence l, p (f) is the probability of the real-time meteorological features f, a, B are the number of intervals divided in the direction of l, f, B is a variable, wkFor the kth real-time meteorological feature fkThe matched weight, γ (x)i,xi-h) Is xiAnd xi-hTwo feature quantity sequences weight grey correlation degree xii-bAnd (r) is the gray correlation coefficient corresponding to each characteristic factor. The invention is set to the power of 0.6 of the number of samples. MIC is calculated based on normalized data, and the value obtained by calculation is [0,1 ]]In the meantime. Table 5 below shows MIC correlation characteristics of loads and various real-time weather conditions in different seasonsThe situation is.
TABLE 5
S106, determining a load sequence of the virtual load similar day corresponding to the day to be tested according to the similarity degree corresponding to the load similar day and the multi-element characteristic information data of the historical day corresponding to the meteorological historical similar day;
setting the selected 3 types of historical similar daily load sequence matrixes as follows:
wherein L isfir,Lsec,LthiRepresent best, 2 nd best, 3 rd best historical similar daily load sequences respectively, d is 0, 1.. 95 represents 96 time points of day, and m is the number of days of the day to be predicted. The 3 similar days respectively correspond to gray correlation degrees of
And taking the percentage of the gray correlation value as the weight corresponding to the correlation.
Finally, the virtual similar daily load corresponding to each day to be predicted
The calculation formula is as follows:
in order to prevent the load of the virtual similar day formed by combining the optimal historical similar day and the rest 2 nd and 3 rd optimal similar days by weight combination from possibly decreasing the similarity, the method further includes, after S106:
extracting a fine-grained meteorological sequence corresponding to the virtual load similarity day;
respectively calculating the association degree between the fine-grained meteorological sequence of the virtual load similar day and the fine-grained meteorological sequence of the day to be predicted according to a grey association degree algorithm and the association degree between the fine-grained meteorological sequence of the load similar day with the largest association degree and the fine-grained meteorological sequence of the day to be predicted;
judging the degree of association of the two;
if the association degree of the fine-grained meteorological sequence on the virtual load similar day is large, the load on the day is still kept as the load on the virtual load similar day; otherwise, the load with the highest relevance degree is replaced by the load with the similar day.
As a specific example, let
Day real-time characteristic components corresponding to various historical similar day loads of the ith virtual similar day load are as follows:
wherein f isfir,fsec,fthiThe day real-time feature component sequences corresponding to the optimal, 2 nd optimal and 3 rd optimal historical similarity days respectively, where k is 1,2, 3.
Combining weights
With the aforementioned daily real-time feature component
Multiplying to obtain real-time characteristic weather of the virtual similar day corresponding to the load of the ith virtual similar day
The calculation method is as follows:
and (3) calculating the association degrees between the real-time weather of the virtual similar day and the best similar day and the real-time weather of the day to be predicted respectively by using a grey association degree algorithm again, judging the association degrees of the virtual similar day and the best similar day, if the association degree of the real-time weather of the virtual similar day is large, keeping the load of the similar day of the day as the load of the virtual similar day, and if not, replacing the load of the virtual similar day with the load of the best historical similar day.
S107, determining a trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested and the fine-grained meteorological data of the day to be tested; the trained short-term load prediction model takes fine-grained meteorological data or a fine-grained meteorological sequence as input and takes the load of each time point as output;
s107 specifically comprises the following steps:
dividing a data set formed by the fine-grained meteorological sequences of the virtual meteorological similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested and the fine-grained meteorological data of the day to be tested into a training set and a verification set according to a set proportion;
and constructing input and output of the short-term load prediction model from the training set and the verification set by adopting a sliding window method.
The input of the model is 1 two-dimensional characteristic matrix, and each column of the two-dimensional characteristic data matrix represents a class of characteristic components. Taking the feature components corresponding to the predicted load at time t of the day as an example, specific feature elements included in these feature components are shown in table 6 below. In order to enable a subsequent prediction model to learn time sequence correlation information between loads and features from wide feature data, an input matrix of load prediction is constructed from a two-dimensional feature data matrix sliding window according to the time step size of 96 multiplied by the time window size of feature dimension 21, the input matrix of each time window corresponds to a load at a time point to be predicted, a plurality of time window matrixes are overlapped, and then a three-dimensional matrix of the load prediction model to be input is obtained.
TABLE 6
Before S107, the method further includes:
and constructing a short-term load prediction model according to the convolutional neural network CNN, the gated cyclic unit GRU and the time attention mechanism. Namely, a Double-stage Time Attention mechanism neural Network (DA-LSTPNet) capable of effectively mining Long-term macroscopic and Short-term local change characteristics of Time sequence data is constructed. The model structure is shown in fig. 2 and described in detail as follows:
a convolutional neural network (SA-CNN) layer under the Single-stage TAM. The SA-CNN is one of the branch network layers, a one-dimensional convolution layer and a one-dimensional maximum pooling layer are arranged to extract short-term local features of input information, and a TAM layer is used for endowing different local feature information with key prediction weights: the first layer is a one-dimensional convolution layer, the number of convolution kernels is set to be 64, the size of each convolution kernel is 4, and a function relu is activated; the second layer is a one-dimensional maximum pooling layer, and the size of the pool is set to be 2; the third layer is a TAM layer, and the output dimension parameter is set to be 96.
A GRU neural network (SA-GRU) layer under the Single-stage TAM. The SA-GRU is another branch network layer, a double-layer GRU network layer is arranged for extracting long-term macroscopic features of input information, and a TAM layer is arranged for enhancing the memory of the model to the information for a long time: the number of neurons in the double-layer GRU network layer is 64; inserting a random inactivation layer in the two GRU networks, wherein the random inactivation rate is 0.2; the last layer is also the TAM layer, and the output dimension parameter is 96.
and an add layer. And fusing the characteristic information extracted from the two branch networks by adopting the add layer.
And fully connecting the Dense layers. Two layers of Dense network layers are used for reducing the dimension of the information fused and output by the add layer: the number of neurons in the first full-junction layer is 21, the second full-junction layer serves as an output layer, the number of neurons is set to be 1, and a Prelu activation function is adopted to add nonlinear components to output data. The whole model is optimized by adopting an adam optimizer.
Model training and prediction: and inputting a training set and a verification set to train model parameters, checking the training effect of the model in a sample of the verification set after each iterative training, setting the iteration times for 50 times, setting a ModelCheckPoint optimal model preservation function by taking the minimum loss function of the verification set as a target, and preserving the model parameters with the best effect in the training process. And finally, inputting future day characteristic data containing the virtual real-time meteorological sequence to complete the day-ahead short-term load prediction of the future day to be predicted.
S108, acquiring weather forecast broadcasting information of a future day, and preprocessing the weather forecast broadcasting information of the future day;
s109, determining a fine-grained weather sequence of the virtual weather similar day corresponding to the future day according to the preprocessed weather prediction broadcast information; and determining the load of each time point of the future day by adopting a trained short-term load prediction model according to the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the future day.
Example simulations are divided into two major parts: 1) performing contrast simulation on the similarity of different similar days; 2) comparing prediction results of DA-LSTPNet models before and after a weather similar day; 3) and comparing the load prediction results under different load similar day types. And the date to be predicted from 2017 to 12 months in the data set is taken as the future date, which is different from the date to be predicted in the training and verification set, and the real-time weather on the current day of the future date is unknown, so that the virtual weather similar day needs to be selected.
In order to evaluate the similarity degree of similar day selection, the invention introduces a morphological similarity distance DMSDAnd selecting the evaluation indexes of the advantages and the disadvantages as similar days. The morphological similarity distance can measure not only the magnitude between curve values, but also the curve shape similarity. The more similar the shape, the closer the numerical value, the similar distance DMSDThe smaller the value of (c). The calculation formula is as follows:
taking the daily 96-point load sequence as an example, in the formula: l isiAnd LjFor similar daily load sequences to be compared with daily load sequences to be predicted, where Li＝(li1,...,lin)TAnd Lj＝(lj1,...,ljn)TWhere n is 96.
In order to evaluate the prediction precision of the load prediction model, the invention selects the root mean square error RMSE, the average percentage error MAPE and the accuracy AiThree indexes are used for measurement. The calculation formula is as follows:
in the formula: n is the total number of predicted elements, the invention is the prediction of the load before the day, and n is 96; y istrueAnd ypredThe actual load value and the predicted load value of the ith sampling point are obtained. Smaller values of RMSE and MAPE in short term load prediction represent better prediction, and vice versa for accuracy.
Similar day selection is carried out on the days to be predicted 2017-8-28 by respectively adopting a historical similar day method (namely a method for selecting the optimal similar day from historical days by using the weighted gray relevance) and the virtual similar day method (including the meteorological and load virtual similar day method) provided by the invention. The weighted gray relevance is selected from the historical days to obtain three types of similar days with the relevance from large to small, wherein the three types of similar days are 2017-8-10, 2017-8-18 and 2017-8-25 respectively, the ratio of load sequence curves of the three types of historical similar days, the virtual similar days and the day to be predicted is shown in fig. 3, and a comparison table 7 of evaluation indexes is shown. In addition, the difference between the weather virtual similar day and the temperature sequence of the day to be predicted 2017-8-28 is also shown in fig. 3, and the last row of table 7 is attached with the temperature selection result of the weather virtual similar day.
TABLE 7
The following two points can be found by combining table 7 and fig. 3:
1) because the random fluctuation component contained in the load and the consideration of the characteristic factors are not comprehensive, the day with the highest characteristic similarity does not mean that the daily load sequence similarity is necessarily the highest, and the correction and the improvement are needed.
2) According to the virtual similar day, the correlation weight is calculated according to the gray correlation degree value, the historical similar day load sequences are combined in a self-adaptive weighting mode, the similar characteristics of different historical similar day load sequences are effectively converged, and the purpose of improving the similarity degree of the similar day load sequences is achieved. The same is true for weather virtual similar days.
By adopting the method similar to the load virtual similar day, the real-time weather of the weather virtual similar day is constructed, and the effect of improving the similarity of the real-time weather sequence of the similar day is realized. The following table 8 shows real-time meteorological sequences of the virtual similar days of the solar meteorological phenomena to be predicted in the test set, and the real meteorological sequences are compared with the mean values and the maximum values of the morphological similarity distances of the real meteorological sequences in temperature, humidity and wind speed.
TABLE 8
The meteorological virtual similar day can bring fine-grained real-time meteorological features to load prediction, and simulation analysis comparison is carried out between the meteorological virtual similar day and the non-meteorological virtual similar day to verify that the prediction effect brought by the fine-grained real-time meteorological features to the load prediction is improved. Wherein, the characteristic data set of the weather similar day is not contained, and the weather characteristic components are replaced by the characteristic mean value of the weather. Taking the typical summer days 2017-08-02-2017-08-04 as an example for analysis, the maximum temperature of the three days is 30 ℃ upwards, the maximum humidity is above 90%, the minimum humidity can be below 50%, the humidity change is large, the weather is accompanied by rain gust weather, the local weather is variable, the weather belongs to the typical high-temperature humid climate condition in the south, and the influence of the local weather change on the load needs to be deeply grasped for predicting the three days.
TABLE 9
By combining the graph 4 with the table 9, it can be found that the model can well capture the influence characteristics of local meteorological changes on the load after the similar weather-containing days are considered, so that the load trend and the peak valley are accurately predicted, and the prediction effect is remarkably improved.
The method has the advantage of effectively verifying the universality of effect improvement on load prediction after the weather virtual similar day is added. According to the invention, continuous half-year day-ahead load comparison prediction is carried out before and after the meteorological virtual similar day is considered, and the prediction effect is shown in a table 10.
Watch 10
And selecting a daily load curve from 2015 to 2017 according to the load similarity day, and measuring the virtual similar daily load and the first three types of historical similar daily loads with the maximum weighted gray relevance by adopting the morphological similarity distance. The average and most significant lists of morphological similarity between the three types of historical similar days and the virtual similar days and the days to be predicted are shown in the following table 11.
TABLE 11
As can be seen from the above table, compared with the method of selecting the best similar day from the historical days by using the weighted gray correlation degree, the method for selecting the similar day by using the virtual similar day provided by the invention has the advantages that the form similarity distance between the selected similar day and the day to be predicted is smaller, and the obtained similar day effect is better.
In order to verify the improvement effect brought to the load prediction by the selection of the load virtual similar days, the invention respectively utilizes the historical best similar day load and the virtual similar day load to construct a prediction feature set, the DA-LSTPNet model provided by the invention is used for prediction, the effect of the two similar days is compared by taking the next half year of 7-12 months in 2017 in the region as a test set, and the effect is shown in Table 12.
TABLE 12
Fig. 5 is a schematic structural diagram of a system for predicting short-term load of a regional power grid according to the present invention, and as shown in fig. 5, the system for predicting short-term load of a regional power grid according to the present invention includes:
the data acquisition module 401 is configured to acquire multi-element characteristic information data of a historical day and weather forecast broadcasting information of a day to be measured; the historical multivariate characteristic information data comprises: load of 96 points of a historical day, date type and fine-grained weather of the historical day; the date types include: weekdays, weekends, and holidays; the fine-grained meteorological data of the historical day comprises: fine-grained real-time temperature, humidity and wind speed data; the weather forecast broadcasting information of the day to be measured is obtained through network crawling; weather prediction broadcast information of the day to be measured comprises: highest and lowest temperature of the day, highest and lowest humidity of the day, wind direction, wind speed, air quality, precipitation and weather conditions;
the preprocessing module 402 is configured to preprocess the multivariate characteristic information data of the historical day and the weather forecast broadcasting information of the day to be detected; the pretreatment comprises the following steps: completing missing values, correcting abnormal values, dividing four seasons by a temperature-averaging sliding method, carrying out LabelEncorder coding processing, carrying out standardization processing and calculating a comprehensive meteorological index;
a weather history similar day determining module 403, configured to determine a weather history similar day by using a grey relevancy algorithm according to the weather prediction broadcast information of the preprocessed to-be-detected day and the multi-feature information data of the preprocessed historical day; the weather history similar days are the first three types of history days with the highest correlation degree with the multi-type coarse-grained day characteristic weather data of the day to be predicted;
a fine-grained weather sequence determination module 404 for the virtual weather similar day, configured to determine a fine-grained weather sequence of the virtual weather similar day corresponding to the day to be detected according to the association degree corresponding to the weather history similar day and the multi-element feature information data of the history day corresponding to the weather history similar day;
a load similarity day determination module 405, configured to determine a load similarity day by using a weighted gray relevance algorithm based on MIC weighting according to a fine-grained weather sequence of a virtual weather similarity day corresponding to a day to be detected and the preprocessed multivariate characteristic information data of a history day;
a load sequence determination module 406 for determining a load sequence of the virtual load similar day corresponding to the day to be tested according to the similarity degree corresponding to the load similar day and the multi-element characteristic information data of the historical day corresponding to the meteorological historical similar day;
the trained short-term load prediction model determining module 407 is configured to determine a trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days, the load sequences of the virtual load similar days, the 96-point loads of the days to be tested, and the fine-grained meteorological data of the days to be tested corresponding to all the days to be tested; the trained short-term load prediction model takes fine-grained meteorological data or a fine-grained meteorological sequence as input and takes the load of each time point as output;
the data obtaining and preprocessing module 408 is configured to obtain weather forecast broadcasting information of a future day and preprocess the weather forecast broadcasting information of the future day;
the prediction module 409 is used for determining a fine-grained weather sequence of the virtual weather similar day corresponding to the future day according to the preprocessed weather prediction broadcast information; and determining the load of each time point of the future day by adopting a trained short-term load prediction model according to the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the future day.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A method for predicting short-term load of a regional power grid is characterized by comprising the following steps:
acquiring multi-element characteristic information data of a historical day and weather forecast broadcasting information of a day to be measured; the historical multivariate characteristic information data comprises: load of 96 points of a historical day, date type and fine-grained weather of the historical day; the date types include: weekdays, weekends, and holidays; the fine-grained meteorological data of the historical day comprises: fine-grained real-time temperature, humidity and wind speed data; the weather forecast broadcasting information of the day to be measured is obtained through network crawling; weather prediction broadcast information of the day to be measured comprises: highest and lowest temperature of the day, highest and lowest humidity of the day, wind direction, wind speed, air quality, precipitation and weather conditions;
preprocessing the multivariate characteristic information data of the historical days and the weather forecast broadcasting information of the days to be detected; the pretreatment comprises the following steps: completing missing values, correcting abnormal values, dividing four seasons by a temperature-averaging sliding method, carrying out LabelEncorder coding processing, carrying out standardization processing and calculating a comprehensive meteorological index;
determining weather historical similar days by adopting a grey correlation algorithm according to the weather forecast broadcast information of the preprocessed to-be-detected day and the multivariate characteristic information data of the preprocessed historical days; the weather history similar days are the first three types of history days with the highest correlation degree with the multi-type coarse-grained day characteristic weather data of the day to be predicted;
determining a fine-grained weather sequence of the virtual weather similar day corresponding to the day to be tested according to the association degree corresponding to the weather history similar day and the multi-element characteristic information data of the history day corresponding to the weather history similar day;
determining load similarity days by adopting a weighted gray correlation algorithm based on MIC weighting according to a fine-grained weather sequence of a virtual weather similarity day corresponding to a day to be detected and the preprocessed multivariate characteristic information data of a historical day;
determining a load sequence of the virtual load similar day corresponding to the day to be tested according to the similarity degree corresponding to the load similar day and the multi-element characteristic information data of the historical day corresponding to the meteorological historical similar day;
determining a trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested and the fine-grained meteorological data of the day to be tested; the trained short-term load prediction model takes fine-grained meteorological data or a fine-grained meteorological sequence as input and takes the load of each time point as output;
acquiring weather forecast broadcasting information of a future day, and preprocessing the weather forecast broadcasting information of the future day;
determining a fine-grained weather sequence of a virtual weather similar day corresponding to the future day according to the preprocessed weather prediction broadcast information; and determining the load of each time point of the future day by adopting a trained short-term load prediction model according to the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the future day.
2. The method for predicting the short-term load of the regional power grid according to claim 1, wherein the preprocessing of the multivariate characteristic information data of the historical days and the meteorological prediction report information of the days to be measured specifically comprises:
performing missing value completion and abnormal value correction on the multivariate characteristic information data of the historical days and the weather prediction broadcast information of the days to be detected;
dividing the processed multi-element characteristic information data of the historical days and the weather forecast broadcasting information of the days to be measured in a sliding manner by adopting a temperature-equalizing method to obtain divided data;
converting mutually independent class labels into single operated numbers by adopting a LabelEncorder coding function according to the divided data;
carrying out normalization processing on the converted data;
and calculating the comprehensive meteorological index according to the data after the normalization processing.
3. The method for predicting the short-term load of the regional power grid according to claim 1, wherein the determining the fine-grained weather sequence of the virtual weather similar day corresponding to the day to be measured according to the correlation degree corresponding to the weather history similar day and the multivariate characteristic information data of the history day corresponding to the weather history similar day specifically comprises:
determining the weight corresponding to the similar days of the weather history according to the relevance corresponding to the similar days of the weather history;
and performing weighted combination on the multivariate characteristic information data of the corresponding historical days according to the corresponding weights of the different meteorological historical similar days, and determining the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the day to be measured.
4. The method for predicting the short-term load of the regional power grid according to claim 1, wherein the load similarity day is determined by adopting a weighted gray relevance algorithm based on MIC weighting according to a fine-grained weather sequence of a virtual weather similarity day corresponding to a day to be measured and the preprocessed multivariate characteristic information data of a historical day, and specifically comprises the following steps:
using formulas
Determining the degree of non-linear correlation between the load and the meteorological features;
using formulas
Determining kth real-time meteorological feature f of grey correlation degreekThe matched weights;
using formulas
Determining the weighted gray correlation degree of the two characteristic quantity sequences;
wherein the content of the first and second substances,
p (l, f) is the joint probability between the daily load sequence l and the real-time meteorological features f, p (l) is the probability of the daily load sequence l, p (f) is the probability of the real-time meteorological features f, a, B are the number of intervals divided in the direction of l, f, B is a variable, wkFor the kth real-time meteorological feature fkThe matched weight, γ (x)i,xi-h) Is xiAnd xi-hTwo feature quantity sequences weight grey correlation degree xii-bAnd (r) is the gray correlation coefficient corresponding to each characteristic factor.
5. The method for predicting the short-term load of the regional power grid according to claim 1, wherein the method for predicting the load sequence of the virtual load similar day corresponding to the day to be measured is determined according to the similarity degree correlation degree corresponding to the load similar day and the multivariate characteristic information data of the historical day corresponding to the meteorological historical similar day, and then further comprises:
extracting a fine-grained meteorological sequence corresponding to the virtual load similarity day;
respectively calculating the association degree between the fine-grained meteorological sequence of the virtual load similar day and the fine-grained meteorological sequence of the day to be predicted according to a grey association degree algorithm and the association degree between the fine-grained meteorological sequence of the load similar day with the largest association degree and the fine-grained meteorological sequence of the day to be predicted;
judging the degree of association of the two;
if the association degree of the fine-grained meteorological sequence on the virtual load similar day is large, the load on the day is still kept as the load on the virtual load similar day; otherwise, the load with the highest relevance degree is replaced by the load with the similar day.
6. The method for predicting the short-term load of the regional power grid according to claim 1, wherein the method for determining the trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days, the load sequences of the virtual load similar days, the 96-point loads of the days to be tested and the fine-grained meteorological data of the days to be tested, which correspond to all the days to be tested, specifically comprises the following steps:
dividing a data set formed by the fine-grained meteorological sequences of the virtual meteorological similar days corresponding to all the days to be tested, the load sequences of the virtual load similar days, the 96-point load of the day to be tested and the fine-grained meteorological data of the day to be tested into a training set and a verification set according to a set proportion;
and constructing input and output of the short-term load prediction model from the training set and the verification set by adopting a sliding window method.
7. The method for predicting the short-term load of the regional power grid according to claim 1, wherein the trained short-term load prediction model is determined based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days, the load sequences of the virtual load similar days, the 96-point loads of the day to be tested and the fine-grained meteorological data of the day to be tested corresponding to all the days to be tested, and the method further comprises the following steps:
and constructing a short-term load prediction model according to the convolutional neural network CNN, the gated cyclic unit GRU and the time attention mechanism.
8. A system for predicting short-term load on a regional power grid, comprising:
the data acquisition module is used for acquiring multi-element characteristic information data of historical days and weather prediction broadcast information of days to be measured; the historical multivariate characteristic information data comprises: load of 96 points of a historical day, date type and fine-grained weather of the historical day; the date types include: weekdays, weekends, and holidays; the fine-grained meteorological data of the historical day comprises: fine-grained real-time temperature, humidity and wind speed data; the weather forecast broadcasting information of the day to be measured is obtained through network crawling; weather prediction broadcast information of the day to be measured comprises: highest and lowest temperature of the day, highest and lowest humidity of the day, wind direction, wind speed, air quality, precipitation and weather conditions;
the preprocessing module is used for preprocessing the multivariate characteristic information data of the historical days and the weather forecast broadcasting information of the days to be detected; the pretreatment comprises the following steps: completing missing values, correcting abnormal values, dividing four seasons by a temperature-averaging sliding method, carrying out LabelEncorder coding processing, carrying out standardization processing and calculating a comprehensive meteorological index;
the weather history similar day determining module is used for determining weather history similar days by adopting a grey correlation algorithm according to weather forecast broadcast information of the preprocessed to-be-detected day and the multi-element characteristic information data of the preprocessed historical days; the weather history similar days are the first three types of history days with the highest correlation degree with the multi-type coarse-grained day characteristic weather data of the day to be predicted;
the fine-grained weather sequence determination module of the virtual weather similar day is used for determining the fine-grained weather sequence of the virtual weather similar day corresponding to the day to be detected according to the correlation degree corresponding to the weather history similar day and the multi-element characteristic information data of the history day corresponding to the weather history similar day;
the load similarity day determining module is used for determining the load similarity day by adopting a weighted gray correlation algorithm based on MIC weighting according to a fine-grained weather sequence of the virtual weather similarity day corresponding to the day to be detected and the preprocessed multivariate characteristic information data of the historical day;
the load sequence determination module of the virtual load similar day is used for determining the load sequence of the virtual load similar day corresponding to the day to be tested according to the similarity correlation degree corresponding to the load similar day and the multi-element characteristic information data of the historical day corresponding to the meteorological historical similar day;
the trained short-term load prediction model determining module is used for determining a trained short-term load prediction model based on deep learning according to the fine-grained meteorological sequences of the virtual meteorological similar days, the load sequences of the virtual load similar days, the 96-point loads of the days to be tested and the fine-grained meteorological data of the days to be tested corresponding to all the days to be tested; the trained short-term load prediction model takes fine-grained meteorological data or a fine-grained meteorological sequence as input and takes the load of each time point as output;
the data acquisition and preprocessing module is used for acquiring the weather forecast broadcasting information of the future day and preprocessing the weather forecast broadcasting information of the future day;
the prediction module is used for determining a fine-grained weather sequence of the virtual weather similar day corresponding to the future day according to the preprocessed weather prediction broadcast information; and determining the load of each time point of the future day by adopting a trained short-term load prediction model according to the fine-grained meteorological sequence of the virtual meteorological similar day corresponding to the future day.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110772002.2A CN113505923B (en) | 2021-07-08 | 2021-07-08 | Regional power grid short-term load prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110772002.2A CN113505923B (en) | 2021-07-08 | 2021-07-08 | Regional power grid short-term load prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113505923A true CN113505923A (en) | 2021-10-15 |
| CN113505923B CN113505923B (en) | 2022-06-17 |
Family
ID=78012331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110772002.2A Active CN113505923B (en) | 2021-07-08 | 2021-07-08 | Regional power grid short-term load prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113505923B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114257521A (en) * | 2021-12-17 | 2022-03-29 | 北京沃东天骏信息技术有限公司 | Flow prediction method, device, electronic equipment and storage medium |
| CN114819318A (en) * | 2022-04-18 | 2022-07-29 | 广西电网有限责任公司南宁供电局 | Load processing method based on ATT-CNN-BiGRU |
| CN115375002A (en) * | 2022-07-12 | 2022-11-22 | 国网江苏省电力有限公司电力科学研究院 | Short-term power load prediction method, system, storage medium and computing device |
| CN116632841A (en) * | 2023-07-26 | 2023-08-22 | 国网山东省电力公司信息通信公司 | Power distribution area short-term electricity load prediction method and system integrating multiple time sequence characteristics |
| CN117374973A (en) * | 2023-12-04 | 2024-01-09 | 国网西藏电力有限公司 | Meteorological data-based power grid load prediction method and system |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170286487A1 (en) * | 2015-01-16 | 2017-10-05 | International Business Machines Corporation | Database statistical histogram forecasting |
| CN108229754A (en) * | 2018-01-31 | 2018-06-29 | 杭州电子科技大学 | Short-term load forecasting method based on similar day segmentation and LM-BP networks |
| CN108345954A (en) * | 2017-11-16 | 2018-07-31 | 北京四方继保自动化股份有限公司 | Distribution network line short-term load forecasting method and device |
| CN109726858A (en) * | 2018-12-21 | 2019-05-07 | 新奥数能科技有限公司 | Heat load prediction method and device based on dynamic time warping |
| CN110383292A (en) * | 2017-04-07 | 2019-10-25 | 英特尔公司 | The method and system through budget and simplified training for deep neural network |
| CN110443417A (en) * | 2019-07-05 | 2019-11-12 | 上海电力大学 | Multiple-model integration load forecasting method based on wavelet transformation |
| CN110472777A (en) * | 2019-07-29 | 2019-11-19 | 上海电力大学 | A kind of micro-capacitance sensor short-term load forecasting method based on empirical mode decomposition |
| CN111860979A (en) * | 2020-07-01 | 2020-10-30 | 广西大学 | Short-term load prediction method based on TCN and IPSO-LSSVM combined model |
| CN111932402A (en) * | 2020-08-13 | 2020-11-13 | 河南理工大学 | Short-term power load bidirectional combined prediction method based on similar day and LSTM |
| CN111985719A (en) * | 2020-08-27 | 2020-11-24 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN112465251A (en) * | 2020-12-08 | 2021-03-09 | 上海电力大学 | Short-term photovoltaic output probability prediction method based on simplest gated neural network |
| CN112669173A (en) * | 2021-01-07 | 2021-04-16 | 云南电网有限责任公司电力科学研究院 | Short-term load prediction method based on multi-granularity features and XGboost model |
-
2021
- 2021-07-08 CN CN202110772002.2A patent/CN113505923B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170286487A1 (en) * | 2015-01-16 | 2017-10-05 | International Business Machines Corporation | Database statistical histogram forecasting |
| CN110383292A (en) * | 2017-04-07 | 2019-10-25 | 英特尔公司 | The method and system through budget and simplified training for deep neural network |
| CN108345954A (en) * | 2017-11-16 | 2018-07-31 | 北京四方继保自动化股份有限公司 | Distribution network line short-term load forecasting method and device |
| CN108229754A (en) * | 2018-01-31 | 2018-06-29 | 杭州电子科技大学 | Short-term load forecasting method based on similar day segmentation and LM-BP networks |
| CN109726858A (en) * | 2018-12-21 | 2019-05-07 | 新奥数能科技有限公司 | Heat load prediction method and device based on dynamic time warping |
| CN110443417A (en) * | 2019-07-05 | 2019-11-12 | 上海电力大学 | Multiple-model integration load forecasting method based on wavelet transformation |
| CN110472777A (en) * | 2019-07-29 | 2019-11-19 | 上海电力大学 | A kind of micro-capacitance sensor short-term load forecasting method based on empirical mode decomposition |
| CN111860979A (en) * | 2020-07-01 | 2020-10-30 | 广西大学 | Short-term load prediction method based on TCN and IPSO-LSSVM combined model |
| CN111932402A (en) * | 2020-08-13 | 2020-11-13 | 河南理工大学 | Short-term power load bidirectional combined prediction method based on similar day and LSTM |
| CN111985719A (en) * | 2020-08-27 | 2020-11-24 | 华中科技大学 | Power load prediction method based on improved long-term and short-term memory network |
| CN112465251A (en) * | 2020-12-08 | 2021-03-09 | 上海电力大学 | Short-term photovoltaic output probability prediction method based on simplest gated neural network |
| CN112669173A (en) * | 2021-01-07 | 2021-04-16 | 云南电网有限责任公司电力科学研究院 | Short-term load prediction method based on multi-granularity features and XGboost model |
Non-Patent Citations (10)
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114257521A (en) * | 2021-12-17 | 2022-03-29 | 北京沃东天骏信息技术有限公司 | Flow prediction method, device, electronic equipment and storage medium |
| CN114257521B (en) * | 2021-12-17 | 2024-05-17 | 北京沃东天骏信息技术有限公司 | Traffic prediction method, traffic prediction device, electronic equipment and storage medium |
| CN114819318A (en) * | 2022-04-18 | 2022-07-29 | 广西电网有限责任公司南宁供电局 | Load processing method based on ATT-CNN-BiGRU |
| CN115375002A (en) * | 2022-07-12 | 2022-11-22 | 国网江苏省电力有限公司电力科学研究院 | Short-term power load prediction method, system, storage medium and computing device |
| CN116632841A (en) * | 2023-07-26 | 2023-08-22 | 国网山东省电力公司信息通信公司 | Power distribution area short-term electricity load prediction method and system integrating multiple time sequence characteristics |
| CN117374973A (en) * | 2023-12-04 | 2024-01-09 | 国网西藏电力有限公司 | Meteorological data-based power grid load prediction method and system |
| CN117374973B (en) * | 2023-12-04 | 2024-03-29 | 国网西藏电力有限公司 | Meteorological data-based power grid load prediction method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113505923B (en) | 2022-06-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113505923B (en) | Regional power grid short-term load prediction method and system | |
| CN112949945B (en) | Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network | |
| CN110909919A (en) | Photovoltaic power prediction method of depth neural network model with attention mechanism fused | |
| CN111753893A (en) | Wind turbine generator power cluster prediction method based on clustering and deep learning | |
| CN112036658B (en) | Wind power real-time prediction method based on forgetting long-term and short-term memory with difference | |
| CN111275571B (en) | Resident load probability prediction deep learning method considering microclimate and user mode | |
| CN114676822B (en) | Multi-attribute fusion air quality forecasting method based on deep learning | |
| CN112418346B (en) | Numerical weather forecast total radiation system error classification calculation method | |
| CN112633604A (en) | Short-term power consumption prediction method based on I-LSTM | |
| CN113469427A (en) | Convolution LSTM-based photovoltaic power station day-ahead power generation amount prediction method and system | |
| CN115423163A (en) | Method and device for predicting short-term flood events of drainage basin and terminal equipment | |
| Aliberti et al. | Forecasting Short-term Solar Radiation for Photovoltaic Energy Predictions. | |
| CN115759389A (en) | Day-ahead photovoltaic power prediction method based on weather type similar day combination strategy | |
| CN115186923A (en) | Photovoltaic power generation power prediction method and device and electronic equipment | |
| CN114692981A (en) | Medium-and-long-term runoff forecasting method and system based on Seq2Seq model | |
| CN117370937A (en) | Overhead line current-carrying capacity probability prediction method and system considering space-time correlation | |
| CN116845875A (en) | WOA-BP-based short-term photovoltaic output prediction method and device | |
| CN115907204A (en) | Forest transpiration water consumption prediction method for optimizing BP neural network by sparrow search algorithm | |
| CN116402194A (en) | Multi-time scale load prediction method based on hybrid neural network | |
| CN115238967A (en) | Photovoltaic power prediction method and device combining cloud picture and adjacent power station cluster | |
| CN116187540A (en) | Wind power station ultra-short-term power prediction method based on space-time deviation correction | |
| CN114611788A (en) | Cold weather objective judging method based on GRU network | |
| CN115860165A (en) | Neural network basin rainfall runoff forecasting method and system considering initial loss | |
| CN114912716A (en) | Short-term power load prediction method and system based on bidirectional long-term and short-term memory network | |
| CN112348700B (en) | Line capacity prediction method combining SOM clustering and IFOU equation |
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 |