CN109508835B - Smart power grid short-term power load prediction method integrating environmental feedback - Google Patents
Smart power grid short-term power load prediction method integrating environmental feedback Download PDFInfo
- Publication number
- CN109508835B CN109508835B CN201910000264.XA CN201910000264A CN109508835B CN 109508835 B CN109508835 B CN 109508835B CN 201910000264 A CN201910000264 A CN 201910000264A CN 109508835 B CN109508835 B CN 109508835B
- Authority
- CN
- China
- Prior art keywords
- power
- power load
- block
- day
- sample
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000007613 environmental effect Effects 0.000 title claims abstract description 13
- 238000012549 training Methods 0.000 claims description 66
- 238000012360 testing method Methods 0.000 claims description 48
- 238000013528 artificial neural network Methods 0.000 claims description 41
- 238000001514 detection method Methods 0.000 claims description 41
- 238000004088 simulation Methods 0.000 claims description 36
- 230000003203 everyday effect Effects 0.000 claims description 19
- 238000012706 support-vector machine Methods 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 12
- 230000005611 electricity Effects 0.000 claims description 7
- 238000003064 k means clustering Methods 0.000 claims description 7
- 238000012935 Averaging Methods 0.000 claims description 6
- 230000002354 daily effect Effects 0.000 claims description 5
- 239000012634 fragment Substances 0.000 claims description 3
- 238000013139 quantization Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 abstract description 3
- 230000035945 sensitivity Effects 0.000 abstract description 3
- 238000010276 construction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000004069 differentiation Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000013277 forecasting method Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003442 weekly effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000000638 solvent extraction Methods 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06315—Needs-based resource requirements planning or analysis
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Educational Administration (AREA)
- General Health & Medical Sciences (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a smart grid short-term power load prediction method integrating environmental feedback, which is characterized in that clustering analysis is carried out according to the essential rule of power load data change of each power consumption unit, a target prediction area is divided into blocks, prediction model groups are respectively established aiming at the blocks, and the interference of excessive related models is avoided; analyzing the power load time sequence of each block in a week by taking days as a unit, dividing time intervals according to the fluctuation rule of the power load time sequence, and respectively establishing a prediction model, so that the prediction precision of power load prediction is improved; meanwhile, the influence factor of the sudden change of the power load is considered, the mapping relation between the average temperature, the average humidity and the wind speed and the prediction error of the power load is established, the external environmental factor is intelligently embedded, and the power load prediction model integrating the environmental feedback is obtained, so that the sensitivity and the adaptability of the prediction model to the sudden change event of the power load are greatly improved, the robustness of the prediction method is ensured, and the prediction precision of the short-term power load prediction is improved.
Description
Technical Field
The invention belongs to the technical field of power load prediction, and particularly relates to a short-term power load prediction method of a smart power grid, which integrates environmental feedback.
Background
The power load prediction is connected with power grid energy and user requirements, the power load prediction method is significant to electric energy scheduling and green power utilization, fine management of power grid energy can be realized through the accurate power grid short-term power load prediction method, and the power load prediction method is an important guarantee for stable power utilization of residents and stable economic development.
The power load time sequence is a comprehensive result of various activities related to a power grid, has certain periodicity, meanwhile, the power load time sequence also contains the influence of external environment variables, and the precision of short-term power load prediction is easily interfered by weather environment factors such as air temperature, humidity and the like. According to statistics, in the first half of 2012, major cities such as Shenzhen, Nanchang, Haikou, and Western Ann in China have 11 large-scale power failure accidents, and the power industry considers that the weather changes cause the sudden increase of power loads in a short period, and the untimely scheduling of the power energy is an important reason for the accidents.
Disclosure of Invention
Aiming at the problem of short-term power load sudden change caused by the change of the existing climate environment, the invention provides the smart grid short-term power load forecasting method integrating environmental feedback, so that the sensitivity and adaptability of a forecasting model to a power load sudden change event are improved, the robustness of the forecasting method is ensured, and the forecasting precision of the short-term power load forecasting is improved.
In order to achieve the technical purpose, the invention adopts the technical scheme that:
a smart grid short-term power load prediction method integrating environmental feedback comprises the following steps:
step 1, acquiring all power utilization units N in a target prediction area1Processing the historical voltage and current data of each electricity consumption unit to obtain a corresponding power sequence, wherein the power sequences of all the electricity consumption units form a power training sample of a target prediction area;
obtaining another historical time period N of the target prediction area in the same way as the obtaining method of the power training sample1Power test samples for successive weeks; simultaneously obtaining the average of the target prediction region in the historical time period of the power test sample every dayTemperature, average humidity and wind speed data to form an environment sample of the target prediction area;
step 2, clustering all power consumption units according to the power sequences of all power consumption units in the power training samples, and dividing the target prediction area into N3A plurality of blocks;
step 3, establishing a power load prediction model group based on a PID neural network by using a power training sample;
step 3.1, respectively establishing and N by using all power sequences in the power training samples in each block3N corresponding to each block3A block power load prediction model group; each block power load prediction model group comprises 7 day power load prediction model groups, wherein the 7 day power load prediction model groups respectively correspond to 7 days in a week; each day power load prediction model group comprises N4A power load prediction model based on PID neural network, wherein N4Each power load prediction model is respectively corresponding to one-day N4The time intervals are corresponding;
step 3.2, processing all power sequences in the power training samples in each block, wherein each block obtains a corresponding power load training sample;
step 3.3, sequentially training the nth day of the b th times of any two continuous weeks in the power load of the a-th block2The power load value of the time period is used as input data, and the b th day and the n th day of the next week2Training the nth day of the block a by using the power load value of the time interval as output data2A PID neural network-based power load prediction model for a time period; wherein, a is 1,2, …, N3，b＝1,2,…,7，n2＝1,2,…,N4；
Step 4, establishing an environment feedback model based on a support vector machine by utilizing the power test sample and the environment sample:
step 4.1, processing the power test samples according to the step 3.2, and obtaining corresponding power load test samples for each block;
step 4.2, randomly determining the current week of simulation prediction and the week w of simulation prediction0+m0And simulating the prediction of the number of days b0；
Step 4.3, selecting each block to simulate and predict the number of days b correspondingly0The model group of the power load prediction of every day is used for sequentially testing the corresponding w-th power load in the power load test sample01 week b0Times of day and w0Week b0The power load value of each time interval of each block of the day is used as the corresponding block corresponding to the day b0Corresponding to the input data of the power load prediction model of the PID neural network, and obtaining the w-th data0+1 week b0Simulating and predicting the power load of each time interval of each block every day;
step 4.4, selecting each block to simulate and predict the number of days b correspondingly0The model group of the power load prediction of every day is used for sequentially testing the corresponding w-th power load in the power load test sample0Week b0The power load value and w-th power load value of each block in each time period of day0+1 week b0The power load simulation predicted value of each time interval of each block of the day is used as the corresponding block corresponding day b0Corresponding to the input data of the power load prediction model of the PID neural network, and obtaining the w-th data0+2 weeks b0Simulating and predicting the power load of each time interval of each block every day;
step 4.5, and so on until the w-th is obtained0+m0Week b0The power load of each time interval of each block in every day is simulated and predicted value, and the w th time is calculated0+m0Week b0Summing the predicted values of the power load in each time interval of each block every day to obtain the w-th0+m0Week b0The power load simulation predicted value of the times of the day;
step 4.6, comparing, simulating and predicting the frequency of week to be w0+m0And b is the simulated prediction of the number of days0The power load simulation predicted value and the power load test sample period are w0+m0B is the number of days0The sum of the power load values of each block in each time period to obtain the w-th0+m0The number of days of the week is b0Simulating a predicted error value for the electrical load;
step 4.7, repeating the steps 4.2-4.6 to obtain N5Then, obtain N5Individual power load simulation predicted value and N5Simulating a predicted error value for each power load;
step 4.8, taking the simulation prediction value of each power load and the average temperature, average humidity and wind speed of the simulation prediction days in the environment sample as input data, taking the simulation prediction error value of each power load as output data, and training a support vector machine to obtain an environment feedback model based on the support vector machine;
step 5, optimally predicting the power load with the target prediction frequency of w + m and the target prediction frequency of d by utilizing the power load prediction models based on the PID neural networks obtained in the step 3 and the environment feedback model based on the support vector machine obtained in the step 4; wherein the most recent week, including day d, is w;
step 5.1, acquiring voltage and current data of the w-1 th and w-th cycles of all power consumption units, and processing according to the step 1 to obtain a power prediction sample of a target prediction area;
step 5.2, processing the power prediction samples according to the step 3.2, and obtaining corresponding power load prediction samples for each block;
step 5.3, selecting a daily power load prediction model group with the number d of days of each block; sequentially taking the power load value of each time interval of each block of d days of the w-1 week and d days of the w week in the power load prediction sample as input data of each PID neural network-based power load prediction model of each time interval of d corresponding to the day corresponding to the block, and obtaining the power load prediction value of each time interval of each block of d days of the w +1 week;
step 5.4, selecting a day power load prediction model group with d as the day number of each block, and sequentially using the power load value of each time interval of each block of the d day number of the w week number and the power load prediction value of each time interval of each block of the d day number of the w +1 week number in the power load prediction sample as input data of the PID neural network-based power load prediction model of the d time interval of the d day number of the w +2 week number to obtain the power load prediction value of each time interval of each block of the d day number of the w +2 week number;
step 5.5, repeating the steps until a power load predicted value of each time interval of each block at d days of the w + m week is obtained, and summing the power load predicted values of each time interval of each block at d days of the w + m week to obtain a power load predicted value at d days of the w + m week;
and 5.6, according to the weather forecast and the information of the meteorological bureau, obtaining the average temperature, the average humidity and the wind speed data of the target forecast week w + m and the target forecast day d of the target forecast area, and taking the average temperature, the average humidity, the wind speed data and the power load forecast value of the target forecast week w + m and the target forecast day d as the input data of the environment feedback model based on the support vector machine to obtain the optimal power load forecast value of the target forecast week w + m and the target forecast day d of the target forecast area.
Further, voltage and current historical data of each electricity consumption unit are acquired by the power detection equipment arranged by each electricity consumption unit at the same frequency, and the specific processes of processing the voltage and current historical data to obtain a power training sample and a power testing sample of the target prediction area are as follows:
at sample time interval T1Dividing the voltage data and the current data, calculating a time interval T at each sample time1Average value of voltage and average value of current in each sample time interval T1Takes the intermediate time of (1) as a sample time, and takes each sample time interval T1The voltage average value and the current average value in the time range are used as voltage values and current values of sample time, and the product of the voltage value and the current value of each sample time is used as a power value of the sample time;
each power detection device corresponds to 2 power sequences formed by power values of all sample moments in different historical time periods, and the power sequences of all the power detection devices in 2 different historical time periods respectively form a power training sample and a power testing sample.
Further, the specific process of step 2 is as follows:
the time interval T from zero to twenty-four points in the same 1 day is taken as the times of day2Sequentially and randomly obtaining each electric power detection device N in the power training sample2Each length is T2Power sequence segment of (10) or less2≤15；
Sequentially testing N of each power detection device in the power training sample2Averaging the power sequence segments to obtain power average sequences of all the power detection devices;
using each power detection device as an element p to be clusteredi,i＝1,2,3...,n1，n1The number of the power detection devices is determined, and the coordinates of the elements to be clustered are power average sequences of the corresponding power detection devices;
selecting the number of clustering centers as N3Performing k-means clustering, and dividing the target prediction region into N according to the clustering result3Block, 4 ≤ N3And 8, each block comprises a plurality of power detection devices.
Further, the process of performing k-means clustering is as follows:
step A1, taking the power average sequence of each power detection device as the coordinate of the element to be clustered, and randomly selecting N3Each element is taken as a clustering center, and each clustering center represents 1 clustering cluster;
step A2, sequentially calculating the distances between all elements and each cluster center, and sequentially assigning each element to the cluster represented by the cluster center closest to the element;
the distance between the Euclidean distance quantization element and each cluster center is adopted, and the formula is as follows:
wherein distance (p)i,kj) Represents the element piHomomeric center kjThe distance of (a) to (b),
an nth coordinate representing an ith element,
n is the nth coordinate of the jth cluster center, N is the sample time interval included in 1 dayT1The number of (2);
step A3, calculating the average coordinate of all elements in each cluster by Euclidean distance, and calculating the distance between the average coordinate in each cluster and each cluster center, if the distance between the average coordinate of each cluster and each cluster center is less than or equal to the threshold value
Finishing clustering to obtain a current clustering result; if the distance between the average coordinate of any cluster and the cluster center is larger than the threshold value
And taking the average coordinate of each cluster as a new cluster center of each cluster, and turning to the step A2.
Further, the threshold value
Is 0.05.
Further, the specific process of step 3.2 is as follows:
step B1, sequentially summing the power sequences of all the power detection devices in the same block in the power training sample to obtain the sum N3N corresponding to each block3A block power sequence;
step B2, sequentially selecting N of B in the block power sequence of the a-th block according to the number of the days of the week1The power sequence segments of the blocks form a power training subsample of which the day time of the a-th block is b; wherein a is 1,2, …, N3，b＝1,2,…,7；
Step B3, randomly selecting 20 block power sequence segments from the power training subsample with the a-th block B, and averaging to obtain the average segment of the block power sequence with the a-th block B; each sample time in the block power sequence average segment corresponds to a power average value;
step B4, using the average segment of the block power sequence, dividing the day with the a-th block B into N4The time interval is divided into the following steps: block power sequence average segmentIf the absolute difference between the power average value at the later sample moment and the power average value at the current sample moment is more than 30% of the minimum value of the power average values at the two sample moments, the current sample moment is a critical sample moment, the critical sample moment is taken as the ending moment of the current time period, and the later sample moment of the critical sample moment is taken as the starting moment of the next time period; the block power sequence average fragment end sample time does not meet the critical sample time condition;
step B5, according to the time interval division result, utilizing the power training subsample with the a-th block as B in the day time, and sequentially calculating the power load training subsample with the a-th block as B in the day time;
the power value of each sample time in the c-th time period is separated from the sample time interval T1The sum of the products as the power load value in the c-th period; wherein, c is 1,2, …, N41 day N, then4The power load values of each time interval form 1 power load sequence with the a-th block of b days and the time span of N1Correspondingly obtaining N by the power training subsamples of continuous cycles1A sequence of electrical loads, N1The power load sequence forms a power load training subsample with b as the day of the a-th block, and b is taken as 1,2, … and 7 form a power load training sample of the a-th block in turn according to 7 days of the week.
Further, the time span N1The values of (A) are as follows: n is a radical of1≥50。
Further, the number N of the electric load simulation predicted values5The values of (A) are as follows: n is a radical of5≥200。
Further, the sample time interval T1Is 20 minutes.
Further, the voltage and current historical data is data after being subjected to insensitive Kalman filtering.
Advantageous effects
According to the scheme, historical voltage and current data of each power consumption unit are collected at the same time to form a power time sequence, the essential rule of power load data change is fully reflected, classification is carried out based on the characteristics of the power load data, the power consumption peak rule of each unit in the same time period is researched, units with similar power sequence fluctuation rules are clustered through a clustering algorithm, a stable block power load time sequence is finally formed, the complexity and the heterogeneity of the power load data are reduced, power load prediction model groups are respectively established and predicted for blocks, and the accuracy and the stability of power load prediction are improved.
Secondly, on the basis of dividing the power consumption units in each day of the week, the scheme analyzes the time period of the concentrated power consumption of the power consumption units in the blocks in one day, divides the block power time sequence with the length of the day into a plurality of steady-state time periods, and establishes a power load prediction model in each time period, so that the instability of power load number prediction is further reduced, and the power load prediction accuracy and robustness are greatly improved.
Thirdly, according to the scheme, an environment feedback model based on a support vector machine is established according to influences of temperature, humidity and wind speed weather information on crowd traveling and work and rest, the predicted value of the power load is corrected, a power load prediction system is perfected, and the scientificity and accuracy of power load prediction are further improved.
Drawings
Fig. 1 is a flowchart of the short-term power load prediction of the smart grid with environmental feedback integrated according to the present invention.
Detailed Description
The following describes embodiments of the present invention in detail, which are developed based on the technical solutions of the present invention, and give detailed implementation manners and specific operation procedures to further explain the technical solutions of the present invention.
Aiming at the problem of short-term power load sudden change, the invention provides a smart grid short-term power load prediction method integrating environmental feedback, through cluster analysis, a target prediction region is divided into blocks by utilizing an average power sequence in each region within a certain time span, prediction models are respectively established aiming at the blocks, the interference of excessive related models is avoided, and the calculation cost pair is reduced while the power load prediction precision is ensured; analyzing the power load time sequence of each block in a week by taking days as a unit, dividing time intervals according to fluctuation rules of the time intervals, establishing a mapping relation between the time intervals and the power load steady state, reducing negative effects of input data fluctuation on a prediction model, and improving prediction accuracy of power load prediction; meanwhile, the influence factor of the sudden change of the power load is considered, the mapping relation between the average temperature, the average humidity and the wind speed and the prediction error of the power load is established, the external environmental factor is intelligently embedded, and the power load prediction model integrating the environmental feedback is obtained, so that the sensitivity and the adaptability of the prediction model to the sudden change event of the power load are greatly improved, the robustness of the prediction method is ensured, and the prediction precision of the short-term power load prediction is improved.
This embodiment will further explain the scheme of the present invention with reference to fig. 1, which specifically includes the following steps:
step 1: and installing power detection equipment and acquiring historical voltage and current data.
Sequentially and respectively installing power detection equipment for all power utilization units in the target prediction area;
collecting historical data of voltage and current in each power detection device at the same sampling frequency, sequentially carrying out insensitive Kalman filtering on the historical data of voltage and current in each power detection device, and sampling time interval T1Dividing the filtered voltage and current historical data, and calculating the time interval T of each sample time1Average value of internal voltage and current, time interval T of each sample1Takes the intermediate time of (1) as a sample time, and takes each sample time interval T1The average value of the internal voltage and the current is used as the voltage and the current value of the sample moment, the product of the voltage value and the current value of each sample moment is used as the power value of the sample moment, the power sequence of each electric power detection device is obtained, and the power sequences of all the electric power detection devices form a power training sample;
acquiring voltage and current historical data of each power detection device in another historical time period at the same sampling frequency, and obtaining a power test sample by adopting the same data acquisition method as the power training sample; meanwhile, the average temperature, average humidity and wind speed data of each day in the historical time period are obtained through weather forecast and weather bureau information, and an environment sample is obtained;
the collected historical data of the voltage and the current all take Monday zero as a collection starting point and twenty-four Monday as a collection ending point, and the power training sample and the power testing sample both contain N1Historical data of voltage and current, N, for successive cycles1A value of at least 50;
sample time interval T1Is 20 minutes;
each power detection device corresponds to 2 power sequences containing all sample moments in different historical times;
in the power training sample and the power testing sample, each power sequence contains 3 power data per hour, namely power values of 10 th minute, 30 th minute and 50 th minute;
step 2: partitioning the target prediction area into blocks by using power training samples;
the time interval T from zero to twenty-four points in the same 1 day is taken as the times of day2Sequentially and randomly obtaining N of each power detection device in the power training sample2Each length is T2Power sequence segment of (10) or less2≤15；
Sequentially testing N of each power detection device in the power training sample2Averaging the power sequence segments to obtain power average sequences of all the power detection devices in the power training sample;
taking each power detection device as an element to be clustered, and marking the element as pi,i＝1,2,3...,n1，n1The number of the electric power detection devices is determined, and the coordinates of the elements to be clustered are power average sequences of the corresponding electric power detection devices in the power training samples;
selecting the number of clustering centers as N3Performing k-means clustering, and dividing the target prediction region into N according to the clustering result3Block, 4 ≤ N3≦ 8, each block containing multiple power detection devices, the same block may not be geographically contiguous;
the clustering process is as follows:
step A1: detecting each electric power in the power training sampleTaking the power average sequence of the equipment to be tested as the coordinate of the element to be clustered, and randomly selecting N3Each element is taken as a clustering center, and each clustering center represents 1 clustering cluster;
step A2, sequentially calculating the distances between all elements and each cluster center, and sequentially assigning each element to the cluster represented by the cluster center closest to the element;
the distance between the Euclidean distance quantization element and each cluster center is adopted, and the formula is as follows:
wherein distance (p)i,kj) Represents the element piHomomeric center kjThe distance of (a) to (b),
an nth coordinate representing an ith element,
an nth coordinate representing a jth cluster center;
step A3: calculating the average coordinate of all elements in each cluster by Euclidean distance, and calculating the distance between the average coordinate in each cluster and the cluster center according to the average coordinate, wherein if the distance between the average coordinate of each cluster and the cluster center is less than or equal to a threshold value
Finishing clustering to obtain a current clustering result; if the distance between the average coordinate of any cluster and the cluster center is larger than the threshold value
Taking the average coordinate of each cluster as a new cluster center of each cluster, and turning to the step A2;
example, threshold value
And is selected to be 0.05.
And step 3: establishing a power load prediction model group based on a PID neural network by using a power training sample;
sequentially summing power sequences of all the power detection devices in the same block in the power training sample to obtain N3A block power sequence;
by using N3A block power sequence of N3Each block power load prediction model group comprises 7 daily power load prediction model groups, and the 7 daily power load prediction model groups respectively correspond to seven days in a week; total N37 power load prediction model groups which respectively correspond to different blocks in different days and are based on the PID neural network;
in the following, taking the construction process of the PID neural network-based power load prediction model set for the day of 7 (assumed to be monday) in the block 2 as an example, as shown in fig. 1, the construction process of the PID neural network-based power load prediction model set for the day of the week in each block will be described:
selecting N contained in current block power sequence1All the power sequence segments of the blocks corresponding to Monday in the continuous week form the power training subsample of the current block corresponding to Monday, wherein the power training subsample of the current block corresponding to Monday comprises N1Each block power sequence segment with the time span of 1 day and corresponding Monday;
randomly selecting 20 block power sequence segments from power training subsamples of a current block corresponding to Monday, averaging the 20 block power sequence segments to obtain 1 block power sequence average segment, wherein the block power sequence average segment comprises 72(24 x 3) sample moments, each sample moment corresponds to one power average value, and dividing the Monday of the current block into N by using the block power sequence average segment4The time interval is divided as follows:
in the block power sequence average segment, recording a sample time which satisfies that the absolute difference value of the power average value at the later sample time and the power average value at the current sample time is more than 30% of the minimum value of the power average values at the two sample times as a critical sample time, taking the critical sample time as the ending time of the current time period, and taking the later sample time of the critical sample time as the starting time of the next time period; the block power sequence average fragment end sample time does not meet the critical sample time condition;
according to the time interval division result, sequentially calculating the power load training subsamples of the current block corresponding to the Monday by using the power training subsamples of the current block corresponding to the Monday, wherein the power load training subsamples of the current block corresponding to the Monday comprise N1A power load sequence, each power load sequence corresponding to Monday of a week, each power load sequence comprising N4Electrical load value, N4The power load value is equal to N4The time intervals are corresponding;
the power load value of a certain time interval is the sum of the product of the power value of each sample time in the time interval and T;
taking the power load training subsample corresponding to Monday of the current block as training data of a power load prediction model group based on the PID neural network corresponding to Monday of the current block;
the power load prediction model group based on the PID neural network corresponding to Monday of the current block comprises N4A power load prediction model based on PID neural network, as N4The time intervals are corresponding;
training the nth of any two consecutive weekly weekends in the subsample sequentially by the power load2The power load value of a time interval is used as input data of a power load prediction model based on a PID neural network corresponding to the time interval, n2＝1,2,3,...,N4The nth of the next weekly Monday2The power load value of the time interval is used as the output data of the power load prediction model based on the PID neural network corresponding to the time interval, the PID neural network is trained, and the corresponding nth power load value is obtained2A PID neural network-based power load prediction model for a time period;
sequentially adding n2From 1 to N4In total obtain N4A power load prediction model based on PID neural network corresponding to the time interval respectively, thereby obtaining the Monday corresponding to the current blockA power load prediction model set based on a PID neural network;
sequentially selecting all blocks, and sequentially selecting all times of the day for each block, and obtaining N by analogy with the construction process of a power load prediction model group based on a PID neural network corresponding to a Monday of a certain block37 power load prediction model sets based on the PID neural network.
Step 4, establishing an environment feedback model based on a support vector machine by using the power test sample and the environment sample;
obtaining N by using the k-means clustering result and the power test sample obtained in the step 23A sequence of block power predictions, each block power test sequence having a time span of N1The time interval of the sample time of a continuous week is T1A continuous power value of (a);
sequentially selecting sequence segments with the same number of days in the power test sequence of each block to respectively obtain the total N of the power test sequences of each block corresponding to each day37 power test subsamples, each power test subsample comprising N1A block power test sequence segment with a time span of one day;
using the time division result obtained in step 3 and N37 power test sub-samples, sequentially calculating the power load value in each time interval in the power test sub-samples to obtain N37 power load test sub-samples, namely power load test samples;
n is carried out by utilizing each PID neural network-based power load prediction model and power load test sample obtained in the step 35Sub-power load simulation prediction, N5The value is at least 200, and the power load simulation prediction process is as follows:
arbitrarily determine the current week of simulation prediction, and record as w0Knowing the current and previous power load conditions, arbitrarily determining the simulation prediction week and the simulation prediction day, and recording the simulation prediction week as w0+m0Recording the simulation predicted number of days as b0；
Selecting each block to simulate and predict day b0Based on the number of days of the PID neural networkThe power load prediction model group sequentially corresponds to the w-th power load test sample01 week b0Times of day and w0Week b0The power load value of each time interval of each block of the day is used as the corresponding block corresponding to the day b0Obtaining the w-th data based on the input data of the power load prediction model of the PID neural network0+1 week b0Simulating and predicting the power load of each time interval of each block every day;
selecting each block to simulate and predict day b0The model group for predicting the power load every day based on the PID neural network sequentially corresponds to the w-th power load test sample0Week b0The power load value and w-th power load value of each block in each time period of day0+1 week b0The power load simulation predicted value of each time interval of each block of the day is used as the corresponding block corresponding day b0Obtaining the w-th data based on the input data of the power load prediction model of the PID neural network0+2 weeks b0Simulating and predicting the power load of each time interval of each block every day;
and so on until the w-th0+m0Week b0The power load of each time interval of each block in every day is simulated and predicted value, and the w th time is calculated0+m0Week b0Summing the predicted values of the power load in each time interval of each block every day to obtain the w-th0+m0Week b0The power load simulation predicted value of the times of the day;
comparative w0+m0Week b0Predicted value of power load of day and w-th in power load test sample0+m0Week b0The sum of the power load values of each time interval of each block in every day is obtained to obtain the w0+m0Week b0Simulating a prediction error value of the daily power load;
repeating the above-mentioned power load simulation prediction process N5Then, in total, obtain N5Individual power load simulation predicted value and N5Simulating a predicted error value for each power load;
taking the power load simulation predicted value of each simulation prediction and the average temperature, average humidity and wind speed of the simulation prediction target prediction time in the environment sample as input data of an environment feedback model based on a support vector machine, taking the power load simulation prediction error value of each simulation prediction as output data of the environment feedback model based on the support vector machine, training the support vector machine, and obtaining the environment feedback model based on the support vector machine;
and 5: performing optimal prediction on the power load by using the power load prediction models based on the PID neural networks obtained in the step (3) and the environment feedback model based on the support vector machine obtained in the step (4);
acquiring voltage and current data of each power detection device including the latest 3 weeks, and acquiring a power prediction sample and a power load prediction sample according to the method for acquiring the power test sample in the step 1 and the method for acquiring the power load test sample in the step 4; the voltage and current data of 3 weeks are selected here in order to ensure that data at least including the day as a target forecast day are included, and power load forecast models based on PID neural networks are input and driven to finally obtain a power load forecast value of the target forecast day of the target forecast week.
Recording a target prediction week as w + m, a target prediction day as d, selecting a day power load prediction model group based on a PID (proportion integration differentiation) neural network and corresponding to the day d of each block in a power load prediction sample, and sequentially using power load values of each block in the power load prediction sample corresponding to the day d of the w-1 week and the day d of the w week as input data of the PID neural network-based power load prediction model of the corresponding block in the time period corresponding to the day d of the w +1 week to obtain a power load prediction value of each block in each time period of the day d of the d week of the w +1 week;
selecting a power load prediction model group based on a PID (proportion integration differentiation) neural network corresponding to d days of each block, and sequentially taking a power load value corresponding to each time interval of each block of d days of the w week and a power load prediction value corresponding to each time interval of d days of the w +1 week in a power load prediction sample as input data of the power load prediction model based on the PID neural network corresponding to the time interval of d days corresponding to the corresponding block, so as to obtain a power load prediction value of each time interval of d days of the w +2 week;
and repeating the steps until the predicted value of the power load of each time interval of each block at d days of the w + m week is obtained, and summing the predicted values of the power load of each time interval of each block at d days of the w + m week to obtain the predicted value of the power load of d days of the w + m week.
According to weather forecast and meteorological bureau information, obtaining data of average temperature, average humidity and wind speed of d days of the w + m week, and taking the predicted values of the average temperature, the average humidity, the wind speed and the power load of the d days of the w + m week as input data of an environment feedback model based on a support vector machine to obtain the optimal predicted value of the power load of the d days of the w + m week.
The above embodiments are preferred embodiments of the present application, and those skilled in the art can make various changes or modifications without departing from the general concept of the present application, and such changes or modifications should fall within the scope of the claims of the present application.
Claims (10)
1. A method for forecasting short-term power load of a smart grid by fusing environmental feedback is characterized by comprising the following steps:
step 1, acquiring all power utilization units N in a target prediction area1Processing the historical voltage and current data of each electricity consumption unit to obtain a corresponding power sequence, wherein the power sequences of all the electricity consumption units form a power training sample of a target prediction area;
obtaining another historical time period N of the target prediction area in the same way as the obtaining method of the power training sample1Power test samples for successive weeks; meanwhile, acquiring average temperature, average humidity and wind speed data of the target prediction area in the historical time period of the power test sample every day to form an environment sample of the target prediction area;
step 2, clustering each power consumption unit by adopting a k-means clustering method according to the power sequence of each power consumption unit in the power training sample, and dividing the target prediction area into N3A plurality of blocks;
step 3, establishing a power load prediction model group based on a PID neural network by using a power training sample;
step 3.1, respectively establishing and N by using all power sequences in the power training samples in each block3N corresponding to each block3A block power load prediction model group; each block power load prediction model group comprises 7 day power load prediction model groups, wherein the 7 day power load prediction model groups respectively correspond to 7 days in a week; each day power load prediction model group comprises N4A power load prediction model based on PID neural network, wherein N4Each power load prediction model is respectively corresponding to one-day N4The time intervals are corresponding;
step 3.2, processing all power sequences in the power training samples in each block, wherein each block obtains a corresponding power load training sample;
step 3.3, sequentially training the nth day of the b th times of any two continuous weeks in the power load of the a-th block2The power load value of the time period is used as input data, and the b th day and the n th day of the next week2Training the nth day of the block a by using the power load value of the time interval as output data2A PID neural network-based power load prediction model for a time period; wherein, a is 1,2, …, N3，b＝1,2,…,7，n2＝1,2,…,N4；
Step 4, establishing an environment feedback model based on a support vector machine by utilizing the power test sample and the environment sample:
step 4.1, processing the power test samples according to the step 3.2, and obtaining corresponding power load test samples for each block;
step 4.2, randomly determining the current week of simulation prediction and the week w of simulation prediction0+m0And simulating the prediction of the number of days b0；
Step 4.3, selecting each block to simulate and predict the number of days b correspondingly0The model group of the power load prediction of every day is used for sequentially testing the corresponding w-th power load in the power load test sample01 week b0Times of day and w0Week b0Every day zoneThe power load value of each time interval of the block is used as the corresponding day time b of the corresponding block0Corresponding to the input data of the power load prediction model of the PID neural network, and obtaining the w-th data0+1 week b0Simulating and predicting the power load of each time interval of each block every day;
step 4.4, selecting each block to simulate and predict the number of days b correspondingly0The model group of the power load prediction of every day is used for sequentially testing the corresponding w-th power load in the power load test sample0Week b0The power load value and w-th power load value of each block in each time period of day0+1 week b0The power load simulation predicted value of each time interval of each block of the day is used as the corresponding block corresponding day b0Corresponding to the input data of the power load prediction model of the PID neural network, and obtaining the w-th data0+2 weeks b0Simulating and predicting the power load of each time interval of each block every day;
step 4.5, and so on until the w-th is obtained0+m0Week b0The power load of each time interval of each block in every day is simulated and predicted value, and the w th time is calculated0+m0Week b0Summing the predicted values of the power load in each time interval of each block every day to obtain the w-th0+m0Week b0The power load simulation predicted value of the times of the day;
step 4.6, comparing, simulating and predicting the frequency of week to be w0+m0And b is the simulated prediction of the number of days0The power load simulation predicted value and the power load test sample period are w0+m0B is the number of days0The sum of the power load values of each block in each time period to obtain the w-th0+m0The number of days of the week is b0Simulating a predicted error value for the electrical load;
step 4.7, repeating the steps 4.2-4.6 to obtain N5Then, obtain N5Individual power load simulation predicted value and N5Simulating a predicted error value for each power load;
step 4.8, taking the simulation prediction value of each power load and the average temperature, average humidity and wind speed of the simulation prediction days in the environment sample as input data, taking the simulation prediction error value of each power load as output data, and training a support vector machine to obtain an environment feedback model based on the support vector machine;
step 5, optimally predicting the power load with the target prediction frequency of w + m and the target prediction frequency of d by utilizing the power load prediction models based on the PID neural networks obtained in the step 3 and the environment feedback model based on the support vector machine obtained in the step 4; wherein the most recent week, including day d, is w;
step 5.1, acquiring voltage and current data of the w-1 th and w-th cycles of all power consumption units, and processing according to the step 1 to obtain a power prediction sample of a target prediction area;
step 5.2, processing the power prediction samples according to the step 3.2, and obtaining corresponding power load prediction samples for each block;
step 5.3, selecting a daily power load prediction model group with the number d of days of each block; sequentially taking the power load value of each time interval of each block of d days of the w-1 week and d days of the w week in the power load prediction sample as input data of each PID neural network-based power load prediction model of each time interval of d corresponding to the day corresponding to the block, and obtaining the power load prediction value of each time interval of each block of d days of the w +1 week;
step 5.4, selecting a day power load prediction model group with d as the day number of each block, and sequentially using the power load value of each time interval of each block of the d day number of the w week number and the power load prediction value of each time interval of each block of the d day number of the w +1 week number in the power load prediction sample as input data of the PID neural network-based power load prediction model of the d time interval of the d day number of the w +2 week number to obtain the power load prediction value of each time interval of each block of the d day number of the w +2 week number;
step 5.5, repeating the steps until a power load predicted value of each time interval of each block at d days of the w + m week is obtained, and summing the power load predicted values of each time interval of each block at d days of the w + m week to obtain a power load predicted value at d days of the w + m week;
and 5.6, according to the weather forecast and the information of the meteorological bureau, obtaining the average temperature, the average humidity and the wind speed data of the target forecast week w + m and the target forecast day d of the target forecast area, and taking the average temperature, the average humidity, the wind speed data and the power load forecast value of the target forecast week w + m and the target forecast day d as the input data of the environment feedback model based on the support vector machine to obtain the optimal power load forecast value of the target forecast week w + m and the target forecast day d of the target forecast area.
2. The method according to claim 1, wherein voltage and current historical data of each electricity consumption unit are acquired by the power detection equipment arranged respectively at the same frequency, and the specific process of processing the voltage and current historical data to obtain the power training sample and the power testing sample of the target prediction area is as follows:
at sample time interval T1Dividing the voltage data and the current data, calculating a time interval T at each sample time1Average value of voltage and average value of current in each sample time interval T1Takes the intermediate time of (1) as a sample time, and takes each sample time interval T1The voltage average value and the current average value in the time range are used as voltage values and current values of sample time, and the product of the voltage value and the current value of each sample time is used as a power value of the sample time;
each power detection device corresponds to 2 sections of power sequences formed by power values of all sample moments in different historical time periods, and the power sequences of all the power detection devices in 2 different historical time periods respectively form a power training sample and a power testing sample.
3. The method according to claim 2, wherein the specific process of step 2 is as follows:
the time interval T from zero to twenty-four points in the same 1 day is taken as the times of day2Sequentially and randomly obtaining each electric power detection device N in the power training sample2Each length is T2Power sequence segment of (10) or less2≤15；
Sequentially testing N of each power detection device in the power training sample2Averaging the power sequence segments to obtain the power of each power detection deviceA rate-averaged sequence;
using each power detection device as an element p to be clusteredi,i＝1,2,3...,n1，n1The number of the power detection devices is determined, and the coordinates of the elements to be clustered are power average sequences of the corresponding power detection devices;
selecting the number of clustering centers as N3Performing k-means clustering, and dividing the target prediction region into N according to the clustering result3Block, 4 ≤ N3And 8, each block comprises a plurality of power detection devices.
4. The method of claim 3, wherein the k-means clustering is performed by:
step A1, taking the power average sequence of each power detection device as the coordinate of the element to be clustered, and randomly selecting N3Each element is taken as a clustering center, and each clustering center represents 1 clustering cluster;
step A2, sequentially calculating the distances between all elements and each cluster center, and sequentially assigning each element to the cluster represented by the cluster center closest to the element;
the distance between the Euclidean distance quantization element and each cluster center is adopted, and the formula is as follows:
wherein distance (p)i,kj) Represents the element piHomomeric center kjThe distance of (a) to (b),
an nth coordinate representing an ith element,
denotes the nth coordinate of the jth cluster center, N denotes the sample time interval T included within 1 day1The number of (2);
step (ii) ofA3, calculating the average coordinate of all elements in each cluster by Euclidean distance, and calculating the distance between the average coordinate in each cluster and each cluster center according to the average coordinate, wherein if the distance between the average coordinate of each cluster and each cluster center is less than or equal to a threshold value
Finishing clustering to obtain a current clustering result; if the distance between the average coordinate of any cluster and the cluster center is larger than the threshold value
And taking the average coordinate of each cluster as a new cluster center of each cluster, and turning to the step A2.
5. The method of claim 4, wherein the threshold value is set
Is 0.05.
6. The method according to claim 3, wherein the specific process of step 3.2 is as follows:
step B1, sequentially summing the power sequences of all the power detection devices in the same block in the power training sample to obtain the sum N3N corresponding to each block3A block power sequence;
step B2, sequentially selecting N of B in the block power sequence of the a-th block according to the number of the days of the week1The power sequence segments of the blocks form a power training subsample of which the day time of the a-th block is b; wherein a is 1,2, …, N3，b＝1,2,…,7；
Step B3, randomly selecting 20 block power sequence segments from the power training subsample with the a-th block B, and averaging to obtain the average segment of the block power sequence with the a-th block B; each sample time in the block power sequence average segment corresponds to a power average value;
step B4, utilizing areaBlock power sequence average segment, dividing the day with b as the a-th block into N4The time interval is divided into the following steps: in the block power sequence average segment, if the absolute difference between the power average value at the later sample time and the power average value at the current sample time is more than 30% of the minimum value of the power average values at the two sample times, the current sample time is a critical sample time, the critical sample time is taken as the ending time of the current time interval, and the later sample time of the critical sample time is taken as the starting time of the next time interval; the block power sequence average fragment end sample time does not meet the critical sample time condition;
step B5, according to the time interval division result, utilizing the power training subsample with the a-th block as B in the day time, and sequentially calculating the power load training subsample with the a-th block as B in the day time;
the power value of each sample time in the c-th time period is separated from the sample time interval T1The sum of the products as the power load value in the c-th period; wherein, c is 1,2, …, N41 day N, then4The power load values of each time interval form 1 power load sequence with the a-th block of b days and the time span of N1Correspondingly obtaining N by the power training subsamples of continuous cycles1A sequence of electrical loads, N1The power load sequence forms a power load training subsample with b as the day of the a-th block, and b is taken as 1,2, … and 7 form a power load training sample of the a-th block in turn according to 7 days of the week.
7. The method of claim 1, wherein the time span N1The values of (A) are as follows: n is a radical of1≥50。
8. The method of claim 1, wherein the number N of electrical load simulation predicted values5The values of (A) are as follows: n is a radical of5≥200。
9. Method according to claim 1, characterized in that the sample instants are spaced by a time interval T1Is 20 minutes.
10. The method of claim 1, wherein the voltage and current history data is Kalman filtered data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910000264.XA CN109508835B (en) | 2019-01-01 | 2019-01-01 | Smart power grid short-term power load prediction method integrating environmental feedback |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910000264.XA CN109508835B (en) | 2019-01-01 | 2019-01-01 | Smart power grid short-term power load prediction method integrating environmental feedback |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109508835A CN109508835A (en) | 2019-03-22 |
| CN109508835B true CN109508835B (en) | 2020-11-24 |
Family
ID=65756093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910000264.XA Active CN109508835B (en) | 2019-01-01 | 2019-01-01 | Smart power grid short-term power load prediction method integrating environmental feedback |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109508835B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109961187A (en) * | 2019-03-27 | 2019-07-02 | 廊坊新奥泛能网络科技服务有限公司 | ENERGY PLANNING method and ENERGY PLANNING system |
| CN111985701B (en) * | 2020-07-31 | 2024-03-01 | 国网上海市电力公司 | Power consumption prediction method based on power supply enterprise big data model base |
| CN112734096A (en) * | 2020-12-31 | 2021-04-30 | 国网山东省电力公司青岛供电公司 | Urban saturation load prediction method and system |
| CN113255973A (en) * | 2021-05-10 | 2021-08-13 | 曙光信息产业(北京)有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
| CN115237180B (en) * | 2022-09-23 | 2022-12-06 | 济南明鑫制药股份有限公司 | Intelligent control method for crude product preparation in methyl ester production |
| CN115601198B (en) * | 2022-11-30 | 2023-04-07 | 北京志翔科技股份有限公司 | Power data simulation method, device, equipment and storage medium |
| CN115660233A (en) * | 2022-12-26 | 2023-01-31 | 尚特杰电力科技有限公司 | Photovoltaic power prediction method and device, electronic equipment and storage medium |
| CN116805785B (en) * | 2023-08-17 | 2023-11-28 | 国网浙江省电力有限公司金华供电公司 | Power load hierarchy time sequence prediction method based on random clustering |
| CN117036104B (en) * | 2023-10-08 | 2024-01-09 | 北京前景无忧电子科技股份有限公司 | Intelligent electricity utilization method and system based on electric power Internet of things |
| CN117040029B (en) * | 2023-10-08 | 2024-03-26 | 南方电网数字电网研究院有限公司 | Power distribution network power dispatching method, device, computer equipment and storage medium |
| 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 |
| CN117559450B (en) * | 2024-01-12 | 2024-03-29 | 吉林大学 | Method for improving non-intrusive load monitoring precision based on feedback model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218675A (en) * | 2013-05-06 | 2013-07-24 | 国家电网公司 | Short-term load prediction method based on clustering and sliding window |
| CN103617564A (en) * | 2013-12-04 | 2014-03-05 | 国家电网公司 | Whole-network load prediction method based on local load predicted value comprehensive evaluation |
| CN105608512A (en) * | 2016-03-24 | 2016-05-25 | 东南大学 | Short-term load forecasting method |
| CN106022954A (en) * | 2016-05-16 | 2016-10-12 | 四川大学 | Multiple BP neural network load prediction method based on grey correlation degree |
| CN108830487A (en) * | 2018-06-21 | 2018-11-16 | 王芊霖 | Methods of electric load forecasting based on long neural network in short-term |
| CN109034453A (en) * | 2018-06-21 | 2018-12-18 | 南京邮电大学 | A kind of Short-Term Load Forecasting Method based on multiple labeling neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180114283A1 (en) * | 2016-10-21 | 2018-04-26 | Open Access Technology International, Inc. | Systems and Methods for the Optimization of User Rate Charges |
-
2019
- 2019-01-01 CN CN201910000264.XA patent/CN109508835B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218675A (en) * | 2013-05-06 | 2013-07-24 | 国家电网公司 | Short-term load prediction method based on clustering and sliding window |
| CN103617564A (en) * | 2013-12-04 | 2014-03-05 | 国家电网公司 | Whole-network load prediction method based on local load predicted value comprehensive evaluation |
| CN105608512A (en) * | 2016-03-24 | 2016-05-25 | 东南大学 | Short-term load forecasting method |
| CN106022954A (en) * | 2016-05-16 | 2016-10-12 | 四川大学 | Multiple BP neural network load prediction method based on grey correlation degree |
| CN108830487A (en) * | 2018-06-21 | 2018-11-16 | 王芊霖 | Methods of electric load forecasting based on long neural network in short-term |
| CN109034453A (en) * | 2018-06-21 | 2018-12-18 | 南京邮电大学 | A kind of Short-Term Load Forecasting Method based on multiple labeling neural network |
Non-Patent Citations (3)
| Title |
|---|
| 《Short-Term Load Forecasting in Smart Grid: A Combined CNN and K-Means Clustering Approach》;Xishuang Dong等;《2017 IEEE International Conference on Big Data and Smart Computing (BigComp)》;20171231;第119-125页 * |
| 《基于小波分析法与滚动式时间序列法的》;刘辉等;《中南大学学报（自然科学版）》;20100226;第41卷(第1期);第370-375页 * |
| 《日电力负荷的分时段多模型组合预测》;蔡小华等;《计算机系统应用》;20111231;第20卷(第3期);第98-101页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109508835A (en) | 2019-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109508835B (en) | Smart power grid short-term power load prediction method integrating environmental feedback | |
| CN109376972B (en) | Smart grid short-term power load prediction method based on block clustering | |
| CN110471024B (en) | Intelligent electric meter online remote calibration method based on measurement data analysis | |
| WO2018082523A1 (en) | Load cycle mode identification method | |
| CN109376906B (en) | Travel time prediction method and system based on multi-dimensional trajectory and electronic equipment | |
| CN117036104B (en) | Intelligent electricity utilization method and system based on electric power Internet of things | |
| CN104200288A (en) | Equipment fault prediction method based on factor-event correlation recognition | |
| CN109753591A (en) | Operation flow predictability monitoring method | |
| CN111985701A (en) | Power utilization prediction method based on power supply enterprise big data model base | |
| CN108415884B (en) | Real-time tracking method for structural modal parameters | |
| CN114912720A (en) | Memory network-based power load prediction method, device, terminal and storage medium | |
| CN115481918A (en) | Active sensing and predictive analysis system for unit state based on source network load storage | |
| CN106779147B (en) | Power load prediction method based on self-adaptive hierarchical time sequence clustering | |
| CN114186733A (en) | Short-term load prediction method and device | |
| CN117878979A (en) | Power balance and dynamic compensation system based on electric energy storage | |
| CN117200223A (en) | Day-ahead power load prediction method and device | |
| CN116720983A (en) | Power supply equipment abnormality detection method and system based on big data analysis | |
| CN110489893B (en) | Variable weight-based bus load prediction method and system | |
| CN112418662A (en) | Power distribution network operation reliability analysis method using artificial neural network | |
| CN112766535A (en) | Building load prediction method and system considering load curve characteristics | |
| CN112444772A (en) | Intelligent electric energy meter reliability prediction method and device | |
| CN115907228A (en) | Short-term power load prediction analysis method based on PSO-LSSVM | |
| CN111368257B (en) | Analysis and prediction method and device for coal-to-electricity load characteristics | |
| CN110659681B (en) | Time sequence data prediction system and method based on pattern recognition | |
| CN111625525A (en) | Environmental data repairing/filling method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240126 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Country or region after: China Address before: Yuelu District City, Hunan province 410083 Changsha Lushan Road No. 932 Patentee before: CENTRAL SOUTH University Country or region before: China |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240530 Address after: Northeast of the intersection of Sidingshan Road and Manquan Road, Hefei Circular Economy Demonstration Park, Feidong County, Hefei City, Anhui Province, 230000 Patentee after: Anhui Rongzhao Intelligent Co.,Ltd. Country or region after: China Address before: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee before: Dragon totem Technology (Hefei) Co.,Ltd. Country or region before: China |
|
| TR01 | Transfer of patent right |