CN112508275B - Power distribution network line load prediction method and equipment based on clustering and trend indexes - Google Patents
Power distribution network line load prediction method and equipment based on clustering and trend indexes Download PDFInfo
- Publication number
- CN112508275B CN112508275B CN202011414424.4A CN202011414424A CN112508275B CN 112508275 B CN112508275 B CN 112508275B CN 202011414424 A CN202011414424 A CN 202011414424A CN 112508275 B CN112508275 B CN 112508275B
- Authority
- CN
- China
- Prior art keywords
- load
- data
- prediction
- clustering
- historical
- 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
- 238000009826 distribution Methods 0.000 title claims abstract description 128
- 238000000034 method Methods 0.000 title claims abstract description 59
- 230000008859 change Effects 0.000 claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 34
- 230000006403 short-term memory Effects 0.000 claims abstract description 16
- 230000007787 long-term memory Effects 0.000 claims abstract description 15
- 238000004140 cleaning Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000002159 abnormal effect Effects 0.000 claims description 18
- 230000015654 memory Effects 0.000 claims description 15
- 238000007637 random forest analysis Methods 0.000 claims description 12
- 230000002354 daily effect Effects 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 238000013450 outlier detection Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 230000003203 everyday effect Effects 0.000 claims description 3
- 230000002441 reversible effect Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000007774 longterm Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000003064 k means clustering Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 241001522296 Erithacus rubecula Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000003860 storage 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] 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
-
- 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)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Educational Administration (AREA)
- Quality & Reliability (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a method and equipment for predicting load of a power distribution network line based on clustering and trend indexes, wherein the method comprises the following steps: acquiring and cleaning load historical time sequence data of each distribution area in the power distribution network; dividing all the distribution areas into a plurality of clustering clusters through clustering according to the load historical data set, summing and reconstructing the load historical data set of each clustering cluster, and acquiring a plurality of load samples according to the days; then acquiring holiday information corresponding to each load sample, and calculating the load change trend index of the same period in the last year; training a corresponding long-term and short-term memory neural network load prediction model by using the load sample, the holiday information and the load change trend index of each cluster; and predicting corresponding load data by using the trained load prediction model of each type, and finally superposing the prediction results of each type of load to obtain the prediction result of the total load of the power distribution network line. The method can improve the precision of the short-term load prediction of the power distribution network so as to achieve the aim of guiding the dispatching operation of the power distribution network.
Description
Technical Field
The invention relates to the technical field of power distribution network load analysis and prediction, in particular to a long-short term memory neural network power distribution network line load prediction method based on clustering and trend indexes.
Background
The short-term load prediction of the power distribution network is an important technology for economic and efficient operation of the power distribution network, and mainly excavates rules and influence factors of load change of the power distribution network according to historical operation data of the power distribution network, information such as external weather and holidays of time and the like, and infers the change trend of the load in a short period of time in the future. The method has the advantages that the short-term load of the power distribution network is accurately predicted, and the method has important significance for reducing power generation cost and improving the fine operation management level of the power distribution network.
Because the types of the loads contained in the power distribution network are complex and various, the short-term fluctuation change rule of each type of load is different. In addition, each type of load fluctuation has uncertainty due to the influence of characteristic factors such as season change, meteorological change and holidays. With the increase of the load types of the power distribution system and the input and grid connection of bidirectional flexible load equipment such as an electric automobile and the like, the nonlinear complexity of the load of the power distribution system is gradually enhanced, and the difficulty of short-term load prediction is also continuously increased. The traditional short-term load prediction method mainly adopts a trend analysis method, a multiple linear regression method, an autoregressive moving average method and the like, and the prediction method is effective to the stably-changing short-term load, but has poor prediction effect under the condition of frequent load fluctuation. Therefore, how to comprehensively consider the characteristics of the rules influencing the load change so as to improve the prediction accuracy and effect becomes the key point and difficulty point of the short-term load prediction research. In recent years, with the popularization and application of smart electric meters, the collection and storage of mass operation data of a power distribution network provide sufficient data support for load prediction. Meanwhile, the rise of the artificial intelligence algorithm and the application of mainstream machine learning methods such as a random forest, a support vector machine and an artificial neural network also provide a new idea and method for the short-term load prediction of the power distribution network.
At present, a large amount of research is carried out by domestic and foreign scholars aiming at the short-term load prediction of a power distribution network, wherein partial research results show that the reasonable classification of the load can improve the prediction effect. The learners adaptively divide the load according to the seasonal temperature, and the optimized outlier robust extreme learning machine algorithm is adopted to predict the load data, so that the short-term load prediction effect is improved. And the students perform multi-stage clustering on the platform load based on the power consumption data, construct a load prediction model based on the impulse neural network, and realize accurate classification prediction of the load.
However, current research on short-term load prediction is still deficient in several areas: 1) because main users in different transformer areas influence the load characteristics of the transformer areas, the load characteristics of the special transformer areas and the public transformer areas have larger difference and have respective change rules, the load forecasting of the distribution lines is directly carried out, load mixed data is directly forecasted, the power utilization characteristic rules of different transformer areas cannot be mined, and the line load forecasting precision is insufficient; 2) the power load has the characteristic of obvious distribution trend in different seasons, and in addition, holidays and production and life planning arrangement can also influence the change rule of the load in different periods to different degrees. Generally, the short-term load prediction mainly considers the influences of temperature and holiday factors, generally adopts feature extraction or only inputs historical data of a recent period to perform prediction, does not fully utilize historical contemporaneous data of the load, lacks deep analysis on the load change trend, and cannot fully mine the change rule of the load in different periods; 3) with continuous deepening of the intelligent power distribution network, the equipment can collect and store mass distribution network data, load data serve as time sequence data, the load time sequence characteristics are not considered in the regression prediction process of a common machine learning algorithm, effective information of the time sequence cannot be transmitted, the load prediction effect and generalization performance of the model are influenced to a certain extent, and the prediction accuracy needs to be further improved by adopting a prediction model suitable for the load time sequence data.
Disclosure of Invention
The invention provides a method and equipment for predicting line load of a power distribution network based on clustering and trend indexes, which improve the accuracy of short-term load prediction of the power distribution network so as to achieve the aim of guiding the dispatching operation of the power distribution network.
In order to achieve the technical purpose, the invention adopts the following technical scheme:
a power distribution network line load prediction method comprises the following steps:
s10, acquiring and cleaning historical load data of the power distribution network, and constructing a daily historical load time sequence of each distribution area, wherein the daily historical load time sequences of all the distribution areas form a historical load data set of the distribution area;
s20, dividing N distribution areas into K distribution area cluster clusters through clustering according to the load historical data set of the distribution areas; for each platform area cluster, summing the load historical data sets of all platform areas, and reconstructing to obtain the load historical data set of the platform area cluster;
s30, normalizing the load historical data set of the distribution area cluster to obtain a load historical time sequence of the distribution area cluster, and taking the normalized load historical time sequence as a load sample of the distribution area cluster;
s40, aiming at each load sample, acquiring holiday information of the day and calculating the load change trend index of the same year;
s50, taking load change trend indexes and holiday information of the last load sample in the same period of the last year and the last 1 load sample as input, taking the last 1 load sample as output, performing off-line training on a long-term and short-term memory neural network load prediction model correspondingly constructed by each region cluster, and determining optimal parameters for all the continuous load samples (l + 1) in each region cluster;
s60, predicting load of power distribution network line
S61, acquiring load historical time sequences of all the transformer areas in l days before the target prediction day and holiday information of the target prediction day;
s62, summing the load historical time sequences of all the distribution areas in each distribution area cluster, dividing the load historical time sequences according to the days to obtain l load samples with continuous time of each distribution area cluster, and calculating the load change trend index of the corresponding distribution area cluster in the same period of the last year of the target forecast day;
s63, inputting the load samples, the holiday information and the load change trend indexes of the last year and the same period of the one time of each platform area cluster obtained in the step S60 into the corresponding long-short term memory neural network load prediction model obtained in the step S50, and outputting load prediction data of the corresponding platform area cluster on the target prediction day;
and S64, respectively carrying out reverse normalization processing on the load prediction data of all the K distribution area clusters in the target prediction day, and then adding and calculating to obtain the load prediction data of the power distribution network circuit in the target prediction day.
Further, the step of cleaning the historical load data of the power distribution network comprises the following steps: and eliminating abnormal data by adopting an outlier detection method, and filling missing data of each distribution area by adopting a random forest method improved by interpolation.
Further, a method of rejecting abnormal data by adopting an outlier detection method is adopted, a local abnormal factor of the load data is calculated, and if the local abnormal factor is smaller than a preset threshold value, the load data is rejected;
the calculation formula of the local abnormal factor of the load data is as follows:
in the formula, LOFk(O) local anomaly factor, N, for the load data point O in the kth neighborhoodk(O) is a neighborhood point set of the load data point O in the kth neighborhood, and P is NkLoad data points in (O); rhok(O) represents the local reachable density of the load data point O in the kth neighborhood, representing the average reachable distance of all points to O in the kth neighborhood; ρ is a unit of a gradientk(P) represents the local reachable density of the load data point P in the k-th neighborhood, representing the average reachable distance of all points in the k-th neighborhood to P.
Further, the method for filling missing data of each platform area by adopting a random forest method improved by interpolation comprises the following steps:
(1) taking the load data of all time points of the platform area every day as 1 row of the matrix, and constructing a load data matrix X of the platform area;
(2) counting the missing condition of the distribution network load data matrix X; traversing data of each day, taking 1 day with the least data loss in the matrix X as a data filling day, and taking the other days as non-data filling days;
(3) before filling, processing missing data on a non-data filling day by using linear interpolation to obtain a matrix Xnew;
(4) constructing a data set Train by all columns without data missing in the matrix Xnew, and dividing the data set Train into two parts: the data on the non-data filling date is a training set Xtrain, and the data on the data filling date is a label set Ytrain;
(5) constructing a data set Test for all columns in which data missing exists in the matrix Xnew, wherein data on a non-current data filling date is a verification set Xtest, and data on a current data filling date is data to be filled;
(6) training a random forest filling model by taking each column in a training set Xtrain as a training sample, wherein training labels are composed of a label set Ytrain without a missing part;
(7) after the training of the filling model is finished, acquiring data to be filled according to the verification set Xtest, and filling the data to be filled into the load data matrix X;
(8) and (5) repeating the steps (2) to (7) until the matrix X does not lack data.
Further, in step S20, clustering is performed by using a K-means algorithm, where the determination method of the optimal clustering number K is as follows: setting the value range of the clustering number k as [ kmin,Kmax]And calculating the clustering error square sum corresponding to each clustering number K, and determining the optimal clustering number K by adopting an elbow rule according to a curve formed by the clustering number and the clustering error square sum.
Further, the calculation method of the load change trend index comprises the following steps:
wherein x (t) represents the daily load time sequence of each region cluster, x0Represents the arithmetic mean of the load at all time points x (t), and Q (t) represents the load variation trend index of x (t).
Further, the method for determining the optimal hyper-parameters of the long-term and short-term memory neural network comprises the following steps: the method comprises the steps of firstly defining the number of hidden layers of a long-term and short-term memory neural network, the number of neurons of each layer of the network and the value range of a learning rate, setting different super parameter values to carry out multiple load prediction, comparing load prediction results, and selecting a group of super parameter values with the minimum load average prediction error as the optimal super parameters of a power distribution network load prediction model.
Further, the long-short term memory neural network load prediction model comprises the following parameters which are updated by training iteration: input gate weight, input gate bias, forgetting gate weight, forgetting gate bias, output gate weight, output gate bias.
Further, the time span of the historical load data of the power distribution network in step S10 is at least 2 years or more, and the historical load data of the load in the same period of the last year can be included to establish a load change trend index in the same period of the last year.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the method of any one of the above claims when executing the computer program.
Advantageous effects
The invention has the beneficial effects that:
1. the method comprises the steps of cleaning the load data of the power distribution network, detecting and eliminating abnormal load values, filling load missing values, comprehensively improving the load data of each distribution area, and providing basic data for load prediction of the power distribution network line.
2. The load characteristics of distribution network areas are clustered by adopting a K-means clustering method, and the load data with similar power consumption characteristics are summed and reconstructed, so that the purity of training data of different types of load prediction models can be improved, the prediction precision is ensured, unified and centralized load prediction can be performed on the areas with similar characteristics, and compared with the load prediction of a single area, the load prediction calculation efficiency can be greatly improved.
3. The time span is selected to contain the historical actual load data of more than 2 years, so that the training sample can contain the load data of the same period of the last year, the load change trend index of the same period of the last year is established, the up-and-down floating rule of the short-term load on the long-term scale can be reflected by adding the load change trend index into the input of the prediction model, and the load prediction effect can be further improved.
4. A power distribution network load prediction model is established based on a long-term and short-term memory neural network, effective information in a time sequence is transmitted according to the structure of a neuron gate of the model, the characteristic rule of the short-term load time sequence can be mined through the self-learning capability of the model, and the model is suitable for load time sequence data prediction and high in prediction precision.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention;
FIG. 2 is a random forest distribution network data padding matrix diagram in the embodiment of the invention;
FIG. 3 is a graph of clustered elbows in an embodiment of the present invention;
FIG. 4 is a graph illustrating a variation trend of the load active power in the same period of each year of the year in the embodiment of the present invention;
FIG. 5 is a schematic diagram of a long term short term memory neural network neuron structure according to an embodiment of the present invention;
FIG. 6 is a diagram illustrating the predicted results of each type of load in the embodiment of the present invention;
FIG. 7 is a graph comparing load prediction results of different methods in an embodiment of 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.
As shown in fig. 1, the present invention provides a distribution line load prediction method of a long-short term memory neural network based on value clustering and trend indicators, comprising the following steps:
s10, acquiring and cleaning historical load data of the power distribution network, and constructing a daily load historical time sequence of each distribution area, wherein the daily load historical time sequences form a load historical data set of the distribution area;
s11, acquiring historical measurement data required by power distribution network line load prediction, wherein the historical measurement data specifically comprises the following time series data of characteristic quantities: load of the transformer area and information of festivals and holidays. The data format was 15 minutes for one data point, containing 96 data points a day.
In consideration of the influence of historical synchronous data on the load, the load prediction is performed by selecting actual power distribution network operation data with the time span of 2 years and a half, so that the applicability of the load prediction model in different environments and different periods can be enhanced.
S12, cleaning active power data of the platform area load, detecting and removing abnormal data by adopting an outlier detection method, filling missing data of each platform area by adopting a random forest method, thereby obtaining complete historical load data of each platform area, correspondingly obtaining a daily historical load time sequence of each platform area, and forming a load historical data set of the platform area by the daily historical load time sequences;
considering that the original distribution network load data has some abnormal values and missing values, the present embodiment first detects the abnormal values of the load data by using an outlier detection method, and is implemented by calculating local abnormal factors of data points, where a specific calculation formula is as follows:
in the formula, LOFk(O) local anomaly factor, N, for the load data point O in the kth neighborhoodk(O) is a neighborhood point set of the load data point O in the kth neighborhood, and P is NkLoad data points in (O); rhok(O) represents the local reachable density of the load data point O in the k-th neighborhood, and represents the average reachable distance of all points to O in the k-th neighborhoodSeparating; rhok(P) represents the local reachable density of the load data point P in the k-th neighborhood, representing the average reachable distance of all points in the k-th neighborhood to P.
The LOF value can measure the density between data points, the farther the distance between the data points is, the lower the density is, the closer the distance between the data points is, the higher the density is, the larger the LOF value is, the lower the density is, the abnormal data is more likely to be, and the abnormal data is removed.
In addition, a random forest algorithm improved by interpolation is adopted to fill the missing data of the load of the power distribution network, and the specific process is as follows:
(1) taking the load data of all time points of the transformer area every day as 1 row of the matrix, and constructing a load data matrix X of the transformer area;
(2) counting the missing condition of the distribution network load data matrix X; taking 1 day with least data loss in the matrix X as a data filling day, and taking the other days as non-data filling days;
(3) before filling, processing missing data on a non-data filling day by using linear interpolation to obtain a matrix Xnew;
(4) constructing a data set Train by all columns without data missing in the matrix Xnew, and dividing the data set Train into two parts: the data on the non-data filling date is a training set Xtrain, and the data on the data filling date is a label set Ytrain;
(5) constructing a data set Test for all columns in which data missing exists in the matrix Xnew, wherein data on a non-current data filling date is a verification set Xtest, and data on a current data filling date is data to be filled;
(6) training a random forest filling model by using each column in a training set xtrin as a training sample, wherein training labels consist of a label set Ytrain without a missing part;
(7) after the training of the filling model is finished, acquiring data to be filled according to the verification set Xtest, and filling the data to be filled to the load data matrix X;
(8) and (5) repeating the steps (2) to (7) until the matrix X does not lack data.
When the data of the last day is processed, the condition that linear interpolation is needed for processing does not exist actually, and a large amount of effective information is filled in by random work, so that the data of the day with the largest missing load can be filled.
And finally, after all data are traversed, the data load matrix of the power distribution network does not contain missing values any more, and data filling is completed. The random forest distribution network data filling matrix is shown in fig. 2.
According to the embodiment of the invention, through data cleaning, the phenomena of abnormal and missing acquisition values of distribution network data caused by faults, noise interference, data transmission errors or power utilization abnormity of measurement equipment are solved, and high-quality basic data are provided for a distribution network load prediction model.
S20, dividing N distribution areas into K distribution area cluster clusters through clustering according to the load historical data set of the distribution areas; for each platform area cluster, summing the load historical data sets of all platform areas, and reconstructing to obtain the load historical data set of the platform area cluster;
s21, inputting the load historical data sets of the N distribution areas obtained after cleaning in the step S10, clustering by adopting a K-means algorithm, setting the cluster number i to change from 2 to 10, calculating the clustering error square sum under each cluster number i, and determining the optimal cluster number K of the load by adopting an elbow rule according to the cluster number and the clustering error square sum curve;
wherein, the principle of the elbow rule is as follows: in the process that the clustering number K is continuously increased, the sample data is divided more finely, the aggregation degree of each cluster is gradually increased, and therefore the sum of squares of errors is gradually reduced. When the number of clusters K is smaller than the optimal number of clusters, increasing K increases the degree of aggregation of each cluster, and the magnitude of reduction of the sum of squared errors SSE is large. When the clustering number K reaches the optimal clustering number, the descending amplitude of the error square sum SSE tends to be flat when the clustering number K is increased, so that the curve is in the shape of an elbow. Generally, the K value corresponding to the elbow position is taken as the optimal clustering number, and the error is reduced fastest under the clustering number. The K-means clustering error sum of squares calculation formula is as follows:
wherein k is the number of clusters, miIs the ciCluster center of class sample, xqIs a subject ofiSamples in classes.
S22, obtaining the cluster category number K according to the step S21, firstly randomly selecting K from the load historical data sets of N distribution areas as initial cluster center load data, then calculating Euclidean distances between N platform region load data and initial central load data, distributing N platform region load data samples to nearest clustering central load by comparing the calculated distances to form K clusters, calculating the average value of the data samples contained in each cluster, replacing the initial clustering central load data with the obtained K average values to serve as new central load data of the K clusters, and then continuously and iteratively calculating the clustering error sum of squares, when the clustering error sum of squares is smaller than a threshold value Z, indicating that the clustering precision meets the requirement, the clustering algorithm area is stable, the load data of the clustering center is basically not changed, the K-means clustering algorithm is completed, and the load data of the N distribution areas are clustered into K distribution area clustering clusters according to the curve shape and the clustering characteristics.
And S23, summing the load historical data sets of all the distribution areas aiming at each distribution area cluster, and reconstructing to obtain the load historical data set of the distribution area cluster.
S30, normalizing the load historical data set of the distribution area cluster to obtain a load historical time sequence of the distribution area cluster, and taking the normalized load historical time sequence as a load sample of the distribution area cluster; the normalization formula is specifically as follows:
wherein x is the load data point to be processed in the time series of the load history, xmaxAnd xminThe maximum value and the minimum value in the load historical time series are respectively, and x' is the value of the load data point after normalization.
S40, aiming at each load sample, acquiring holiday information of the day and calculating the load change trend index of the same year; the specific formula for calculating the load change trend index is as follows:
wherein x (t) represents the daily load time sequence of each region cluster, x0Represents the arithmetic mean of the load at all time points x (t), and Q (t) represents the load variation trend index of x (t).
When the tendency index q (t) is larger than 1, the load is in a growing tendency, and a larger index indicates a larger load growing tendency. When the tendency index q (t) is larger than 0 and smaller than 1, the load is in a decreasing tendency, and a smaller index indicates a larger load decreasing tendency.
S50, taking load change trend indexes and holiday information of the last load sample in the same period of the last year and the last 1 load sample as input, taking the last 1 load sample as output, performing off-line training on a long-term and short-term memory neural network load prediction model correspondingly constructed by each region cluster, and determining optimal parameters for all the continuous load samples (l + 1) in each region cluster;
assuming that the last load sample is represented as x' (t), and the load samples are represented as x (t) before normalization, the last load sample has a load change trend index of the same year in the past as Q (t-m), and m is the number of time points of the same year and the same year in the past.
S50, constructing a long-term and short-term memory neural network load prediction model, training the model offline and determining the optimal parameters:
the off-line training model comprises the following specific steps:
the input gate inputs the last time state, the last hidden layer state unit and the current state. After the input of the input gate is converted by a nonlinear function, the state information is screened by the forgetting gate, so that the LSTM neural network clears the state information which is useless at present in the previous step, and the state information of which part needs to be forgotten by the neural network is determined according to the three variables input by the input gate, and the useful information is determined to enter a new current state. And finally, the output gate utilizes the new current state to calculate and determine how much information is output to the current hidden layer state unit, and the current hidden layer state unit enters the next LSTM neuron to calculate, so that the relation between the previous time sequence and the next time sequence is established. Wherein, the specific calculation formula among all variables is as follows:
i(t)＝σ(Wih(t-1)+Uix(t)+bi)
a(t)＝tanh(Wah(t-1)+Uax(t)+ba)
f(t)＝σ(Wfh(t-1)+Ufx(t)+bf)
c(t)＝i(t)⊙a(t)+ft⊙c(t-1)
o(t)＝σ(Uox(t)+Woh(t-1)+bo)
h(t)＝o(t)·tanh(c(t))
wherein the input gate output comprises two parts i(t)And a(t)，Wi，WaConnection weight for hiding layer state at last moment, Ui，UaAs a connection weight of the input gate, bi，baRespectively, are offset for the respective input gates. The output of the forgetting gate is f(t)，Wf，Uf，bfRespectively, the forgetting gate's round-robin weight, input weight, and forgetting gate bias. New current state c(t)The output of the input gate and the forgetting gate and the last time state c are respectively(t-1)The result of the output gate output is determined to be o(t)，Uo，Wo，boOutput gate weight, round robin weight, and output gate offset, respectively. Hidden layer state h(t)The output gate output and the current state jointly determine, and moreover, sigma in the formula is an activation function, generally tan h or sigmoid function.
In order to compare the prediction effect of the power distribution network line load prediction model, the prediction result evaluation index refers to a common index of the prediction model, and the deviation between a predicted value and an actual value is measured by two indexes, namely Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE). The specific error formula is shown as follows:
wherein n is the number of samples, y'iIndicates the predicted value of the load, yiRepresenting the actual value of the load. In order to reflect the advantages of the prediction model in the aspect of line load prediction accuracy, a hysteresis vector machine prediction method which is widely applied in the prediction field is added to carry out error comparison with the text method.
S51, building corresponding K long-short term memory neural network load prediction models according to the load clustering number K, wherein the hyper-parameters of the models comprise the number of hidden layers of the neural network, the number of neurons in each layer of the network and the learning rate;
s52, taking load change trend indexes and holiday information of the last load sample in the same period of the last year and the last 1 load sample as input, taking the last 1 load sample as output, performing off-line training on a long-term and short-term memory neural network load prediction model correspondingly constructed by each region cluster, and determining optimal parameters for all the continuous load samples (l + 1) in each region cluster; and finally obtaining a load prediction model of each platform area cluster through iterative training of long-term and short-term memory neural network parameters.
The method for determining the optimal hyper-parameter of the long-term and short-term memory neural network comprises the following steps: the method comprises the steps of firstly defining the number of hidden layers of a long-term and short-term memory neural network, the number of neurons of each layer of the network and the value range of a learning rate, setting different super parameter values to carry out multiple load prediction, comparing load prediction results, and selecting a group of super parameter values with the minimum load average prediction error as the optimal super parameters of a power distribution network load prediction model.
S60, predicting load of power distribution network line
S61, acquiring load historical time sequences of all the transformer areas in l days before the target prediction day and holiday information of the target prediction day;
s62, summing the load historical time sequences of all the distribution areas in each distribution area cluster, dividing the load historical time sequences according to the days to obtain l load samples with continuous time of each distribution area cluster, and calculating the load change trend index of the corresponding distribution area cluster in the same period of the last year of the target forecast day;
s63, inputting the load samples, the holiday information and the load change trend indexes of the last year and the same period of the one time of each platform area cluster obtained in the step S60 into the corresponding long-short term memory neural network load prediction model obtained in the step S50, and outputting load prediction data of the corresponding platform area cluster on the target prediction day;
and S64, respectively carrying out reverse normalization processing on the load prediction data of all the K distribution area clusters in the target prediction day, and then adding and calculating to obtain the load prediction data of the power distribution network circuit in the target prediction day.
The specific formula of the inverse normalization is as follows:
in the formula, xmaxAnd xminThe maximum value and the minimum value of each type of load normalization variable are respectively, and x' is load prediction data output by the load prediction model.
The following specific examples are provided to further explain the technical scheme of the invention:
according to actual distribution network model data of 46 transformer areas 2016-2018 of a certain area on a 10kV line in 7 months, prediction analysis is carried out on the load data of the distribution line 1 day ahead (96 data points), and the specific prediction flow is shown in fig. 1.
S201, clustering load data of 46 distribution areas in a certain line of a power distribution network in the embodiment of the invention, and enabling a clustering number i to change from 2 to 10 to obtain a load clustering error square sum curve chart in the embodiment of the invention shown in FIG. 3 under the change of the clustering number, wherein according to an elbow rule, an elbow position is selected as 4 for the optimal clustering number, and the load data of the distribution areas are clustered into 4 types;
s301, calculating historical synchronous trend indexes of each type of load, selecting load data affecting a short-term load in a historical week and holiday information as training input, and adding trend indexes at the same time of the last year as load prediction model input for training. Fig. 4 shows that load power data of the prediction verification set in 2018 and 2017 year contemporaneous load trend indexes corresponding to the load power data are obtained, every two sub-graphs from top to bottom are similar loads, and it can be seen from fig. 4 that the last year contemporaneous trend index of each type of load can basically reflect the up-and-down fluctuation trend of the load in this year, and therefore, the last year contemporaneous trend index can be used as the input of the prediction model to reflect the up-and-down fluctuation trend of the load in a long-term scale.
S302, the data of 943 days of long-term load data of a certain line of the power distribution network is divided into two parts, wherein the data of 731 days of 2016 and 2017 are used as training sample sets, and the data of 212 days of 1-7 months in 2018 are used as verification sample sets.
S401, the implementation case of the invention is written by Python codes, an LSTM neural network is built based on Google learning framework TensorFlow, an activation function of an LSTM layer is set to be tanh, a loss function is set to be Mean Square Error (MSE), BATCH _ SIZE is 96, and a neuron structure diagram of the long-short term memory neural network is shown in figure 5.
S50, forecasting each type of load through the long and short term memory neural network after online application training, wherein the result is shown in fig. 6, the four types of load forecasting results are added to be used as the total line load forecasting result, in order to improve the forecasting precision and robustness of the method, a Support Vector Machine (SVM) neural network algorithm with better forecasting performance is selected to forecast the load based on the forecasting capability of each model, the long and short term memory neural network forecasting without clustering is also set for forecasting, the forecasting result is shown in fig. 7, and the forecasting error is shown in table 1. Through result comparison and analysis, the method for predicting the line load provided by the invention has the advantages of minimum error and optimal accuracy of a prediction model.
TABLE 1
The distribution line load prediction method of the long-short term memory neural network based on the clustering and trend indexes can improve the short-term load prediction precision of the distribution network so as to achieve the purpose of guiding the dispatching operation of the distribution network.
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 (8)
1. A power distribution network line load prediction method based on clustering and trend indexes is characterized by comprising the following steps:
s10, acquiring and cleaning historical load data of the power distribution network, and constructing a daily load historical time sequence of each distribution area, wherein the daily load historical time sequences form a load historical data set of the distribution area;
s20, dividing N distribution areas into K distribution area cluster clusters through clustering according to the load historical data set of the distribution areas; summing the load historical data sets of all the distribution areas aiming at each distribution area cluster, and reconstructing to obtain the load historical data set of the distribution area cluster;
s30, normalizing the load historical data set of the distribution area cluster to obtain a load historical time sequence of the distribution area cluster, and taking the normalized load historical time sequence as a load sample of the distribution area cluster;
s40, aiming at each load sample, acquiring holiday information of the day and calculating the load change trend index of the same year;
the calculation method of the load change trend index comprises the following steps:
wherein x (t) represents the daily load time sequence of each region cluster, x0Represents x (t) the arithmetic mean of the load at all time points, Q (t) represents the load variation trend index of x (t);
s50, taking the load change trend index of the last load sample in the same period of the last year and the holiday information of the last 1 load sample as input, taking the last 1 load sample as output, performing off-line training on a long-term and short-term memory neural network load prediction model correspondingly constructed by each region cluster to determine optimal parameters for all the continuous-time load samples (l + 1) in each region cluster;
s60, predicting load of power distribution network line
S61, acquiring load historical time sequences of all the transformer areas in l days before the target prediction day and holiday information of the target prediction day;
s62, summing the load historical time sequences of all the distribution areas in each distribution area cluster, dividing the load historical time sequences according to the days to obtain l load samples with continuous time of each distribution area cluster, and calculating the load change trend index of the corresponding distribution area cluster in the same period of the last year of the target forecast day;
s63, inputting the load samples, the holiday information and the load change trend indexes of the last year and the same period of the one time of each platform area cluster obtained in the step S60 into the corresponding long-short term memory neural network load prediction model obtained in the step S50, and outputting load prediction data of the corresponding platform area cluster on the target prediction day;
and S64, respectively carrying out reverse normalization processing on the load prediction data of all the K distribution area clusters in the target prediction day, and then adding and calculating to obtain the load prediction data of the power distribution network circuit in the target prediction day.
2. The method of claim 1, wherein cleaning power distribution grid historical load data comprises: abnormal data are removed by adopting an outlier detection method, and missing data of each distribution area are filled by adopting a random forest method improved by interpolation;
the method for filling missing data of each transformer area by adopting a random forest method improved by interpolation comprises the following steps:
(1) taking the load data of all time points of the platform area every day as 1 row of the matrix, and constructing a load data matrix X of the platform area;
(2) counting the missing condition of the distribution network load data matrix X; taking 1 day with least data loss in the matrix X as a data filling day, and taking the other days as non-data filling days;
(3) before filling, processing missing data on a non-data filling day by using linear interpolation to obtain a matrix Xnew;
(4) constructing a data set Train by all columns without data missing in the matrix Xnew, and dividing the data set Train into two parts: the data on the non-data filling date is a training set Xtrain, and the data on the data filling date is a label set Ytrain;
(5) constructing a data set Test for all columns in which data missing exists in the matrix Xnew, wherein data on a non-current data filling date is a verification set Xtest, and data on a current data filling date is data to be filled;
(6) training a random forest filling model by taking each column in a training set Xtrain as a training sample, wherein training labels are composed of a label set Ytrain without a missing part;
(7) after the training of the filling model is finished, acquiring data to be filled according to the verification set Xtest, and filling the data to be filled into the load data matrix X;
(8) and (5) repeating the steps (2) to (7) until the matrix X does not lack data.
3. The method according to claim 2, characterized in that, a method of rejecting abnormal data by an outlier detection method is adopted to calculate a local abnormal factor of the load data, and if the local abnormal factor is smaller than a preset threshold, the load data is rejected;
the calculation formula of the local abnormal factor of the load data is as follows:
in the formula, LOFk(O) local anomaly factor, N, for the load data point O in the kth neighborhoodk(O) is a neighborhood point set of the load data point O in the kth neighborhood, and P is NkLoad data points in (O); rhok(O) represents the local reachable density of the load data point O in the kth neighborhood, representing the average reachable distance of all points in the kth neighborhood to O; rhok(P) represents the local reachable density of the load data point P in the k-th neighborhood, representing the average reachable distance of all points in the k-th neighborhood to P.
4. The method of claim 1, wherein the clustering is performed by using a K-means algorithm in step S20, wherein the determination method of the optimal clustering number K is: setting the value range of the clustering number k as [ kmin，kmax]And calculating the clustering error square sum corresponding to each clustering number K, and determining the optimal clustering number K by adopting an elbow rule according to a curve formed by the clustering number and the clustering error square sum.
5. The method of claim 1, wherein the optimal hyper-parameters of the long-short term memory neural network are determined by the following method: the method comprises the steps of firstly defining the number of hidden layers of a long-term and short-term memory neural network, the number of neurons of each layer of the network and the value range of a learning rate, setting different super parameter values to carry out multiple load prediction, comparing load prediction results, and selecting a group of super parameter values with the minimum load average prediction error as the optimal super parameters of a power distribution network load prediction model.
6. The method of claim 1, wherein the long-short term memory neural network load prediction model, iteratively updated parameters by training, comprises: input gate weight, input gate bias, forgetting gate weight, forgetting gate bias, output gate weight, output gate bias.
7. The method according to claim 1, wherein the time span of the historical load data of the distribution network in step S10 is at least 2 years or more, and can include historical data of loads in the same period of the last year, which is used for establishing a load change trend index in the same period of the last year.
8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 7 when executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011414424.4A CN112508275B (en) | 2020-12-07 | 2020-12-07 | Power distribution network line load prediction method and equipment based on clustering and trend indexes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011414424.4A CN112508275B (en) | 2020-12-07 | 2020-12-07 | Power distribution network line load prediction method and equipment based on clustering and trend indexes |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112508275A CN112508275A (en) | 2021-03-16 |
| CN112508275B true CN112508275B (en) | 2022-05-17 |
Family
ID=74971872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011414424.4A Active CN112508275B (en) | 2020-12-07 | 2020-12-07 | Power distribution network line load prediction method and equipment based on clustering and trend indexes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112508275B (en) |
Families Citing this family (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113222216B (en) * | 2021-04-14 | 2024-07-23 | 国网江苏省电力有限公司营销服务中心 | Cold and hot electric load prediction method, device and system |
| CN113128781B (en) * | 2021-04-30 | 2021-12-10 | 大连理工大学 | Distributed industrial energy operation optimization platform for automatically constructing intelligent model and algorithm |
| CN113205368B (en) * | 2021-05-25 | 2022-11-29 | 合肥供水集团有限公司 | Industrial and commercial customer clustering method based on time sequence water consumption data |
| CN113447879A (en) * | 2021-05-27 | 2021-09-28 | 国网山东省电力公司营销服务中心(计量中心) | Electric energy meter metering error prediction method, system and device based on multi-stress influence |
| CN113377881B (en) * | 2021-06-04 | 2024-06-07 | 国网辽宁省电力有限公司 | Longitudinal and transverse mixed clustering multidimensional analysis method for power distribution network load |
| CN113361776A (en) * | 2021-06-08 | 2021-09-07 | 国网上海市电力公司 | Power load probability prediction method based on user power consumption behavior clustering |
| CN113592528B (en) * | 2021-06-22 | 2024-09-13 | 国网河北省电力有限公司营销服务中心 | Baseline load estimation method and device and terminal equipment |
| CN113592185B (en) * | 2021-08-05 | 2022-10-14 | 四川大学 | Power load prediction method based on Transformer |
| CN113657490B (en) * | 2021-08-16 | 2022-05-31 | 沭阳县源美装饰材料有限公司 | Door and window silence detection method based on artificial intelligence |
| CN113436733B (en) * | 2021-08-26 | 2021-11-30 | 肾泰网健康科技(南京)有限公司 | Characteristic construction method of hemodialysis scheme generation model based on fusion experience |
| CN113515468B (en) * | 2021-09-14 | 2021-11-30 | 成都时识科技有限公司 | Data updating method, storage space setting method, device, chip and equipment |
| CN114169568A (en) * | 2021-11-03 | 2022-03-11 | 国网浙江省电力有限公司瑞安市供电公司 | Prophet model-based power distribution line current prediction and heavy overload early warning and system |
| CN114117852B (en) * | 2021-11-18 | 2022-10-21 | 华北电力大学 | Regional heat load rolling prediction method based on finite difference working domain division |
| CN114118588B (en) * | 2021-11-30 | 2024-08-02 | 广东电网有限责任公司广州供电局 | Method for predicting peak-to-peak power failure in summer based on game feature extraction under clustering undersampling |
| CN114170588B (en) * | 2021-12-13 | 2023-09-12 | 西南交通大学 | Eye feature-based bad state identification method for railway dispatcher |
| CN114282444A (en) * | 2021-12-28 | 2022-04-05 | 宁夏隆基宁光仪表股份有限公司 | Distributed load prediction method and system based on intelligent fusion terminal |
| CN114358185B (en) * | 2022-01-04 | 2024-07-16 | 辽宁工程技术大学 | Multi-dimensional short-term power load prediction method based on improved K-means clustering CCA-BiLSTM |
| CN114548509B (en) * | 2022-01-18 | 2024-10-15 | 湖南大学 | Multi-type load joint prediction method and system for multi-energy system |
| CN114638440B (en) * | 2022-05-17 | 2022-08-09 | 国网湖北省电力有限公司经济技术研究院 | Charging load ultra-short-term prediction method based on charging pile utilization degree |
| CN115047271B (en) * | 2022-06-10 | 2024-07-09 | 福州大学 | Diagnosis method for health condition of reactive compensation equipment |
| CN116703190B (en) * | 2022-12-02 | 2024-06-28 | 北京国电通网络技术有限公司 | Power consumption information processing method, device, electronic equipment and computer readable medium |
| CN116632842B (en) * | 2023-07-26 | 2023-11-10 | 国网山东省电力公司信息通信公司 | Clustering characteristic-based method and system for predicting distribution type photovoltaic load probability of platform |
| CN116663871B (en) * | 2023-08-02 | 2023-10-13 | 苏州安极能新能源发展有限公司 | Method and system for predicting electricity demand |
| CN116701973B (en) * | 2023-08-09 | 2023-10-03 | 无锡市明通动力工业有限公司 | Motor abnormal vibration detection early warning method based on big data |
| CN117436712B (en) * | 2023-12-21 | 2024-04-12 | 山东铁鹰建设工程有限公司 | Real-time monitoring method and system for operation risk of construction hanging basket |
| CN117878927B (en) * | 2024-03-11 | 2024-05-28 | 国网黑龙江省电力有限公司绥化供电公司 | Power system load trend analysis method based on time sequence analysis |
| CN117951555B (en) * | 2024-03-27 | 2024-06-14 | 山东德源电力科技股份有限公司 | Strategy formulation method for load control installation testing device |
| CN117977576B (en) * | 2024-03-28 | 2024-07-16 | 国网四川省电力公司电力科学研究院 | Platform load prediction method based on multi-scale historical load data |
| CN118536009B (en) * | 2024-07-24 | 2024-10-11 | 湖北华中电力科技开发有限责任公司 | Power data model construction method and system based on generation type artificial intelligence |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018161722A1 (en) * | 2017-03-08 | 2018-09-13 | 深圳市景程信息科技有限公司 | Power load forecasting method based on long short-term memory neural network |
| CN109376896A (en) * | 2018-08-29 | 2019-02-22 | 国网重庆市电力公司南岸供电分公司 | A kind of term load forecasting for distribution based on multimodality fusion |
| CN109919370A (en) * | 2019-02-26 | 2019-06-21 | 国网冀北电力有限公司运营监测（控）中心 | A kind of Methods of electric load forecasting and prediction meanss |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109462231B (en) * | 2018-11-15 | 2020-09-01 | 合肥工业大学 | Load optimization scheduling method, system and storage medium for residential micro-grid |
-
2020
- 2020-12-07 CN CN202011414424.4A patent/CN112508275B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018161722A1 (en) * | 2017-03-08 | 2018-09-13 | 深圳市景程信息科技有限公司 | Power load forecasting method based on long short-term memory neural network |
| CN109376896A (en) * | 2018-08-29 | 2019-02-22 | 国网重庆市电力公司南岸供电分公司 | A kind of term load forecasting for distribution based on multimodality fusion |
| CN109919370A (en) * | 2019-02-26 | 2019-06-21 | 国网冀北电力有限公司运营监测（控）中心 | A kind of Methods of electric load forecasting and prediction meanss |
Non-Patent Citations (2)
| Title |
|---|
| 基于大数据和多因素组合分析的单元制配电网精细化负荷预测;李富鹏等;《智慧电力》;20200120(第01期);全文 * |
| 基于负荷特性聚类的样本自适应神经网络台区短期负荷预测;方芳等;《科技导报》;20171228(第24期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112508275A (en) | 2021-03-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112508275B (en) | Power distribution network line load prediction method and equipment based on clustering and trend indexes | |
| Wang et al. | A seasonal GM (1, 1) model for forecasting the electricity consumption of the primary economic sectors | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Cao et al. | Hybrid ensemble deep learning for deterministic and probabilistic low-voltage load forecasting | |
| CN112488395A (en) | Power distribution network line loss prediction method and system | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN114548509B (en) | Multi-type load joint prediction method and system for multi-energy system | |
| CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
| CN105354646A (en) | Power load forecasting method for hybrid particle swarm optimization and extreme learning machine | |
| CN111241755A (en) | Power load prediction method | |
| CN108596242A (en) | Power grid meteorology load forecasting method based on wavelet neural network and support vector machines | |
| CN114219216B (en) | Operation scheduling method, system, device and medium of power system | |
| CN114862032B (en) | XGBoost-LSTM-based power grid load prediction method and device | |
| CN113869616A (en) | Self-adaptive residential user load prediction method | |
| CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
| CN110969312A (en) | Short-term runoff prediction coupling method based on variational modal decomposition and extreme learning machine | |
| CN112288140A (en) | Keras-based short-term power load prediction method, storage medium and equipment | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN114548494B (en) | Visual cost data prediction intelligent analysis system | |
| CN115456312A (en) | Short-term power load prediction method and system based on octyl geometric modal decomposition | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| Cao et al. | Short-term forecasting and uncertainty analysis of photovoltaic power based on the FCM-WOA-BILSTM model | |
| CN109615142A (en) | A kind of wind farm wind velocity combination forecasting method based on wavelet analysis | |
| Ding et al. | A statistical upscaling approach of region wind power forecasting based on combination model | |
| CN115907228A (en) | Short-term power load prediction analysis method based on PSO-LSSVM |
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 |