CN111860977B - Probability prediction method and probability prediction device for short-term load - Google Patents

Probability prediction method and probability prediction device for short-term load Download PDF

Info

Publication number
CN111860977B
CN111860977B CN202010614204.XA CN202010614204A CN111860977B CN 111860977 B CN111860977 B CN 111860977B CN 202010614204 A CN202010614204 A CN 202010614204A CN 111860977 B CN111860977 B CN 111860977B
Authority
CN
China
Prior art keywords
load
predicted
prediction model
user
probability
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
Application number
CN202010614204.XA
Other languages
Chinese (zh)
Other versions
CN111860977A (en
Inventor
陈启鑫
郑可迪
王毅
顾宇轩
郭鸿业
康重庆
夏清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202010614204.XA priority Critical patent/CN111860977B/en
Publication of CN111860977A publication Critical patent/CN111860977A/en
Application granted granted Critical
Publication of CN111860977B publication Critical patent/CN111860977B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Economics (AREA)
  • Mathematical Optimization (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Software Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Probability & Statistics with Applications (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Operations Research (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Marketing (AREA)
  • Algebra (AREA)
  • Computing Systems (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Quality & Reliability (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)

Abstract

The invention discloses a probability prediction method and a probability prediction device of short-term load, wherein the method comprises the following steps: dividing historical load data recorded by all user intelligent electric meters in an area to be predicted into a training set, an integration set and a set to be predicted; carrying out related calculation aiming at historical load data in a training set to obtain a division result of a user under a plurality of different clustering quantities; performing related training on each user group after clustering according to the division result to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the region to be predicted; obtaining an integrated load probability prediction model of the area to be predicted according to the integrated concentrated historical load data, each prediction model and the real overall load value; and carrying out probability prediction on the load to be predicted in the concentrated load according to the load probability prediction model of the area to be predicted. The probability prediction method can finely utilize the historical load data of the user intelligent electric meter and improve the accuracy of the load probability prediction of the area to be predicted.

Description

Probability prediction method and probability prediction device for short-term load
Technical Field
The invention relates to the technical field of power load prediction, in particular to a probability prediction method and a probability prediction device for short-term load.
Background
In recent years, the popularity of smart meters has increased on the user side, and power grid companies can obtain high-frequency and high-resolution user load data from smart meter data when performing regional load prediction. Generally, the load of an area is usually aggregated by hundreds of user loads, and the user loads recorded by the smart meter will help to improve the accuracy of the load prediction of the whole area. Load prediction in the traditional sense is mainly to train a single model by recording the change of the overall load of the area and then constructing corresponding input features. Considering that the intelligent electric meter data is used for refining and predicting the whole load of the area, a bottom-up hierarchical load prediction method is generated, namely prediction models are respectively built for the load of a single user, and then the prediction models of all users in the area are summed up to obtain the whole prediction model.
At present, the technology for load prediction based on historical load data is mature and is put into practical use in most power grid companies at home and abroad. Scholars at Polish-Stoholwa industry university propose a model for forecasting Electric Power based on model regression (Dudek G. Pattern-based local linear regression models for short-term load forecasting [ J ]. Electric Power Systems Research,2016,130: 139-) and apply to the actual load data concentration of Polish Electric network. Researchers at the university of new south Wales, Australia, have proposed a short-term resident user load prediction model based on long-term short-term memory artificial neural network (LSTM) (Kong W, Dong Z Y, Jia Y, et al. short-term residual load for estimation based on LSTM recovery neural network [ J ]. IEEE Transactions on Smart Grid,2019,10(1):841 851.) and tested the prediction accuracy on a Smart meter dataset provided by the Intelligent Grid City, SGSC, published by the Australia government. Scholars at the university of Stephen Cleveland, UK, have adopted a bottom-up hierarchical prediction method (Stephen B, Tang X, Harvey P R, et al, incorporated practice the fact of sub-profile model for short period of organized burden load for evaluation [ J ]. IEEE Transactions on Smart Grid,2017,8(4): 1591-. In general, even though a layered prediction method is adopted to improve the prediction accuracy in some researches, most methods are based on the result of a single model, and the prediction accuracy still has a space for improvement. The ensemble learning method can be used to compensate for the shortcomings of a single prediction model. Researchers at southern positive university of singapore use ensemble learning to integrate multiple wavelet transform-based prediction methods (Li S, Goel L, Wang p. an intense approach for short-term load for estimating by iterative learning machine [ J ]. Applied Energy,2016,170:22-29.) to improve prediction accuracy.
In the energy internet era, due to the access of distributed power supplies, electric vehicles and the like, the fluctuation of loads on a user side is more severe than before, the traditional load prediction only focuses on the load size at a certain future time, and in practical application, a power grid company always focuses more on the probability distribution condition of the future load so as to more accurately realize the depiction of random factors, namely realize the probability prediction. In recent years, methods of hierarchical prediction and integrated prediction tend to focus on point prediction, and neglect the importance of probabilistic prediction.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the art described above.
Therefore, one purpose of the invention is to provide a short-term load probability prediction method, which can finely utilize historical load data of a user intelligent electric meter, improve the accuracy of load probability prediction of a region to be predicted, and facilitate the advanced planning of power generation operation scheduling of a power grid.
A second object of the present invention is to provide a device for predicting the probability of a short-term load.
A third object of the invention is to propose an electronic device.
A fourth object of the invention is to propose a non-transitory computer-readable storage medium.
In order to achieve the above object, an embodiment of a first aspect of the present invention provides a method for predicting a probability of a short-term load, including: step S1, acquiring historical load data recorded by all user intelligent electric meters in the area to be predicted, and dividing the historical load data into a training set, an integration set and a set to be predicted according to a preset proportion; step S2, calculating the weekly average load curve of each user according to the historical load data in the training set, calculating the similarity matrix of the weekly average load curve among different users based on cosine similarity, and calculating the division results of the users under a plurality of different clustering quantities according to the spectral clustering method based on k-means clustering and the similarity matrix, wherein any point on the weekly average load curve is the load average value of the same user in the training set at the same time of each week; step S3, respectively training a probability prediction model of a quantile regression neural network for each user group after clustering according to the division result, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain prediction models of a plurality of probability prediction results of the overall load of the region to be predicted; step S4, calculating pinball loss functions of a minimized integration set according to historical load data of the integration set, constructing a linear programming problem and solving optimal integration weights according to each prediction model and a real overall load value, taking the pinball loss functions of the minimized integration set as a target function and the integration weights of each prediction model as optimization variables, and accumulating each prediction model by the optimal integration weights to obtain an integrated load probability prediction model of the area to be predicted; and step S5, carrying out probability prediction on the load in the to-be-predicted set according to the to-be-predicted regional load probability prediction model.
According to the short-term load probability prediction method provided by the embodiment of the invention, historical load data recorded by all user intelligent electric meters in the area to be predicted is obtained, and the historical load data is divided into a training set, an integration set and a set to be predicted according to a preset proportion. Then, aiming at historical load data in a training set, a weekly average load curve of each user is calculated, a similarity matrix of the weekly average load curve among different users is calculated based on cosine similarity, the division results of the users under a plurality of different clustering quantities are calculated according to a spectral clustering method based on k-means clustering and the similarity matrix, probability prediction models of a quantile regression neural network are respectively trained for each user group after clustering according to the division results, and output probability distribution results of the probability prediction models are accumulated through convolution to respectively obtain prediction models of a plurality of probability prediction results of the overall load of the area to be predicted. And then according to historical load data of the integrated set, calculating a pinball loss function of the minimized integrated set, according to each prediction model and a real overall load value, taking the pinball loss function of the minimized integrated set as a target function, taking the integrated weight of each prediction model as an optimization variable, constructing a linear programming problem, solving an optimal integrated weight, and accumulating each prediction model by the optimal integrated weight to obtain an integrated to-be-predicted regional load probability prediction model. And finally, carrying out probability prediction on the concentrated loads to be predicted according to the load probability prediction model of the area to be predicted. Therefore, historical load data of the user intelligent electric meter can be utilized in a refined mode, the accuracy of load probability prediction of the area to be predicted is improved, and the power generation operation scheduling of the power grid can be planned in advance.
In addition, the method for predicting the probability of the short-term load according to the embodiment of the present invention may further have the following additional technical features:
in an embodiment of the present invention, let M be the number of users in the area to be predicted, T be the number of time periods to acquire the historical load, and T be0Recording the frequency of the load for the smart meter every day, wherein M, T and T0Are positive integers, and the step S1 specifically includes: step S11, converting the historical load data into a historical load matrix L according to a preset conversion algorithmM×TAnd is provided with LtFor the total load of the area to be predicted at time t, and set Lm,tThe load of the mth user at the time T is shown, wherein T is a positive integer less than or equal to T, and M is a positive integer less than or equal to M; step S12, the historical load matrix LM×TAccording to the sequence of time, dividing the training set into training sets LtrIntegration set LenAnd a to-be-predicted set are combined, and a training set L is settrHas a time length of TtrIntegration set LenHas a time length of Ten。
In an embodiment of the present invention, the step S2 specifically includes: step S21, training set LtrThe load in the system is averaged according to the week to obtain the weekend of each userLoad sign
Figure BDA0002563201800000031
Wherein T' is not more than 7T0And form a cycle characteristic load matrix
Figure BDA0002563201800000032
The vector of the m-th row of the week characteristic load matrix
Figure BDA0002563201800000033
The weekly average power load of the mth user; step S22, according to the cycle characteristic load matrix, calculating a similarity matrix S between different users based on cosine similarityM×MThe similarity matrix SM×MRow m and column n of element Sm,nFor the electricity utilization similarity between the mth user and the nth user:
Figure BDA00025632018000000410
step S23, according to the spectral clustering method based on k-means clustering and the similarity matrix SM×MDividing M users into k groups to obtain a clustered user group division result;
in step S24, k is N ═ log2M]+1 value [ ·]To round down a function:
kj＝min{2j-1,M}
wherein N is a positive integer, and j is a positive integer less than or equal to N;
step S25, repeating step S23 to obtain N groups of different division results
Figure BDA00025632018000000411
In an embodiment of the present invention, the step S3 specifically includes: step S31, for k clustersjDivision result M of timejWill be subordinate to 1 st to kjThe user loads of the user groups are respectively added to obtain kjHistorical electricity load vectors of the group user group; step S32, aiming at the kjHistorical electricity load vectors of the group user groups are defined, wherein the total electricity load of the a-th group user group is
Figure BDA0002563201800000041
a＝1,2,…,kjThe total load of electricity consumption of each user group is respectively in the time period of t
Figure BDA0002563201800000042
For output, input characteristics are formed by the unique heat code of the week, the hour number, the load of the same time period of yesterday, the load of the last time period of yesterday, the load of the same time period of the previous day, the load of the same time period of three days before and the load of the same time period of four and three days before corresponding to the t time period
Figure BDA0002563201800000043
Using training set LtrCorresponding TtrFeatures in time periods and corresponding outputs
Figure BDA0002563201800000044
t＝1,2,…,TtrTraining kjThe quantile regression neural network is respectively recorded as
Figure BDA0002563201800000045
Wherein the input vector of the neural network corresponding to the a-th group
Figure BDA0002563201800000046
Length of and
Figure BDA0002563201800000047
are the same in length; step S33, integrating the collection LenCorresponding TenIn a time period, for time T (T ═ T)tr+1,Ttr+2,…,Ttr+Ten) To input features
Figure BDA0002563201800000048
Regression neural networks as quantiles respectively
Figure BDA0002563201800000049
Is input to obtain kjLoad probability distribution of individual user groups; step S34, calculating k according to preset discrete convolutionjAccumulating the probability distribution of the loads to obtain a prediction model of a plurality of probability prediction results of the whole load of the area to be predicted, and setting fj,q(Xj,t) As the number of clusters is kjThe resulting prediction of the q quantile of the total load, fj,qFor the prediction model after convolution accumulation of each population, Xj,tFor the input features of the model at time t, q may be (0, 1)]And satisfies the following formula:
Pr(Lt<fj,q(Xj,t))＝q。
in an embodiment of the present invention, step S4 specifically includes: step S41, for kjIntegrating the probability distribution results of N total loads obtained under N values, and setting the integration coefficient as wjPinball loss function at q quantiles
Figure BDA0002563201800000051
The following were used:
Figure BDA0002563201800000052
wherein
Figure BDA0002563201800000053
And (3) providing a q quantile prediction result for the probability prediction model, wherein y is a real overall load value, and an optimization problem is constructed through the following formula:
Figure BDA0002563201800000054
Figure BDA0002563201800000055
Figure BDA0002563201800000056
wherein the content of the first and second substances,
Figure BDA0002563201800000057
is a set of quantiles for the probability prediction,
Figure BDA0002563201800000058
as an integration set LenThe set of time periods of (a), wherein,
Figure BDA0002563201800000059
the problem objective function is a convex function, constraint conditions are linear, the optimal integration weight is obtained after solving based on a convex optimization problem solving technology
Figure BDA00025632018000000510
Step S42, according to the optimal integration weight obtained in the step S41, the prediction models under different cluster quantities
Figure BDA00025632018000000511
Accumulating to obtain an integrated load probability prediction model of the area to be predicted:
Figure BDA00025632018000000512
wherein f isen,qRepresenting integrated probabilistic predictive models, Xt＝[X1,t,X2,t,…,XN,t]And representing the input characteristics of the integrated model at the time t.
In order to achieve the above object, a second embodiment of the present invention provides a short-term load probability prediction apparatus, including: the system comprises a dividing module, a prediction module and a prediction module, wherein the dividing module is used for acquiring historical load data recorded by all user intelligent electric meters in an area to be predicted and dividing the historical load data into a training set, an integration set and a set to be predicted according to a preset proportion; the first calculation module is used for calculating a weekly average load curve of each user according to historical load data in the training set, calculating a similarity matrix of the weekly average load curves among different users based on cosine similarity, and calculating division results of the users under different clustering quantities according to a spectral clustering method based on k-means clustering and the similarity matrix, wherein any point on the weekly average load curve is the load average value of the same user in the training set at the same moment in each week; the second calculation module is used for respectively training a probability prediction model of a quantile regression neural network for each user group after clustering according to the division result, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the area to be predicted; the third calculation module is used for calculating a pinball loss function of the minimized integration set according to historical load data of the integration set, constructing a linear programming problem and solving an optimal integration weight by taking the pinball loss function of the minimized integration set as a target function and the integration weight of each prediction model as an optimization variable according to each prediction model and a real overall load value, and accumulating each prediction model by the optimal integration weight to obtain an integrated regional load probability prediction model to be predicted; and the prediction module is used for carrying out probability prediction on the load in the to-be-predicted set according to the to-be-predicted regional load probability prediction model.
According to the short-term load probability prediction device, historical load data recorded by all user intelligent electric meters in an area to be predicted are obtained through the dividing module, and the historical load data are divided into a training set, an integration set and a set to be predicted according to a preset proportion. The method comprises the steps of calculating a week average load curve of each user according to historical load data in a training set through a first calculation module, calculating a similarity matrix of the week average load curves among different users based on cosine similarity, and calculating division results of the users under different clustering quantities according to a spectral clustering method based on k-means clustering and the similarity matrix. And respectively training a probability prediction model of the quantile regression neural network for each user group after clustering according to the division result through a second calculation module, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the area to be predicted. And calculating a pinball loss function of the minimized integration set according to historical load data of the integration set through a third calculation module, constructing a linear programming problem and solving an optimal integration weight by taking the pinball loss function of the minimized integration set as a target function and the integration weight of each prediction model as an optimization variable according to each prediction model and a real overall load value, and accumulating each prediction model by the optimal integration weight to obtain an integrated regional load probability prediction model to be predicted. And performing probability prediction on the load in the to-be-predicted concentration through a prediction module according to the load probability prediction model of the to-be-predicted region. Therefore, historical load data of the user intelligent electric meter can be utilized in a refined mode, the accuracy of load probability prediction of the area to be predicted is improved, and the power generation operation scheduling of the power grid can be planned in advance.
In order to achieve the above object, a third embodiment of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the program to implement the method for predicting the probability of short-term load according to the first embodiment of the present invention.
According to the electronic equipment provided by the embodiment of the invention, the processor executes the computer program stored on the memory, so that the historical load data of the user intelligent electric meter can be finely utilized, the accuracy of load probability prediction of the area to be predicted is improved, and the advance plan of power generation operation scheduling of a power grid is facilitated.
To achieve the above object, a fourth aspect of the present invention provides a non-transitory computer-readable storage medium, having a computer program stored thereon, where the computer program is executed by a processor to implement a method for predicting a probability of a short-term load according to an embodiment of the first aspect of the present invention.
According to the non-transitory computer-readable storage medium provided by the embodiment of the invention, through executing the stored computer program, the historical load data of the user intelligent electric meter can be finely utilized, the accuracy of load probability prediction of the area to be predicted is improved, and the advance plan of power generation operation scheduling of a power grid is facilitated.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow diagram of a method of probabilistic prediction of short term load according to one embodiment of the present invention; and
fig. 2 is a block diagram of a short-term load probability prediction device according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
A method for predicting the probability of a short-term load and a device for predicting the probability of a short-term load, an electronic apparatus, and a non-transitory computer-readable storage medium according to embodiments of the present invention are described below with reference to the accompanying drawings.
FIG. 1 is a flow diagram of a method of probabilistic prediction of short term load according to one embodiment of the present invention. In the embodiment of the invention, the probability prediction method of the short-term load is suitable for practical application of power grid companies, power distribution companies and power trading centers.
As shown in fig. 1, the method for predicting the probability of short-term load according to the embodiment of the present invention may include the following steps:
and S1, acquiring historical load data recorded by all user intelligent electric meters in the area to be predicted, and dividing the historical load data into a training set, an integration set and a set to be predicted according to a preset proportion. The historical load data can include metering information of the intelligent electric meter, and the preset proportion can be calibrated according to actual conditions.
To better describe the present invention, M can be set as the number of users in the area to be predicted, T is the number of time periods for acquiring the historical load, T0Recording the frequency of the load for the smart meter every day, wherein M, T and T0Are all positive integers.
Specifically, the historical load data is converted into a historical load matrix L according to a preset conversion algorithmM×TAnd is provided with LtFor the total load of the area to be predicted at time t, and set Lm,tThe load of the mth user at the time T is shown, wherein T is a positive integer less than or equal to T, and M is a positive integer less than or equal to M; step S12, history load matrix LM×TAccording to the sequence of time, dividing the training set into training sets LtrIntegration set LenAnd a to-be-predicted set are combined, and a training set L is settrHas a time length of TtrIntegration set LenHas a time length of Ten. The preset conversion algorithm can be calibrated according to actual conditions. It should be noted that the training set L described in this embodimenttrAnd an integration set LenA certain ratio, such as 4 to 1, may be maintained over the length of time.
S2, aiming at historical load data in a training set, calculating a weekly average load curve of each user, calculating a similarity matrix of the weekly average load curves among different users based on cosine similarity, and calculating division results of the users under a plurality of different clustering quantities according to a spectral clustering method based on k-means clustering and the similarity matrix, wherein any point on the weekly average load curve is the load average value of the same user in the training set at the same time in each week.
It should be noted that the spectral clustering method of k-means clustering described in this embodiment is a clustering method based on a similarity matrix, and it is considered that M subjects are divided into k populations according to the degree of similarity, so that the intra-population similarity is high, and the similarity between the populations is low, where M and k are both positive integers. The division is realized by the following steps:
first, a similarity matrix S of the subject is obtained, which is positive and symmetric, and all elements are non-negative. According to the theory of linear algebra, S can be subjected to the following spectral decomposition (eigenvalue decomposition):
S＝QΛQT
wherein Λ is a diagonal matrix, and diagonal elements are characteristic values of S, which are arranged in a sequence from small to large; q is an orthogonal matrix, and the column vector of the orthogonal matrix is the eigenvector corresponding to the eigenvalue.
Taking the first k columns of Q to form a spectral feature matrix V of the subjectM×kTo V pairM×kAnd (5) clustering by using the k-means to obtain a user group division result after clustering. The theory and application of spectral clustering technique based on k-means clustering are detailed in Ng A Y, Jordan M I, Weiss Y.on spectral clustering: Analysis and an algorithm [ C]//Advances in Neural Information Processing Systems.2002:849-856)。
In an embodiment of the present invention, step S2 may specifically include:
step S21, training set LtrThe load in the system is averaged according to the week to obtain the week characteristic load of each user
Figure BDA0002563201800000081
Wherein T' is less than or equal to 7T0And form a cycle characteristic load matrix
Figure BDA0002563201800000082
Vector of m-th row of cycle characteristic load matrix
Figure BDA0002563201800000083
The weekly average power load of the mth user.
Step S22, according to the week characteristic load matrix, calculating the similarity matrix S between different users based on cosine similarityM×MSimilarity matrix SM×MRow m and column n of element Sm,nFor use between an m-th user and an n-th userElectrical similarity.
Figure BDA0002563201800000091
Step S23, according to the spectral clustering method based on k-means clustering and the similarity matrix SM×MAnd dividing the M users into k groups to obtain the clustered user group division result.
In step S24, k is N ═ log2M]+1 value, [ ·]For a rounding down function:
kj＝min{2j-1,M}
wherein N is a positive integer, and j is a positive integer less than or equal to N.
Step S25, repeating the above step S23 to obtain N different groups of division results
Figure BDA0002563201800000092
And obtaining the division result of the user under a plurality of different clustering quantities.
And S3, respectively training a probability prediction model of the quantile regression neural network for each user group after clustering according to the division result, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the region to be predicted.
It should be noted that the quantile regression neural network described in this embodiment is a supervised neural network model, and can provide a probability distribution of output under a certain input characteristic. The input feature vector is set as X, the output random variable is set as y, and the method is based on the traditional fully-connected neural network and is used for estimating the condition distribution of the random variable y | X by training parameters such as weight, bias and the like of neurons in the network through training methods such as gradient descent and the like. The trained quantile regression neural network model may be represented by the symbol fq(X) represents wherein q ∈ (0, 1)]Indicating the size of the quantile. The meaning of quantile is that event y ≦ fqThe probability of (X) | X occurring is exactly q, i.e.:
Pr(y≤fq(X)|X)＝q
the theory of quantile regression neural networks is detailed in (Taylor J W.A qualitative regression neural network approach to influencing the conditional severity of porous return [ J ]. Journal of formation, 2000,19(4): 299-311).
In an embodiment of the present invention, step S3 may specifically include:
step S31, for k cluster numberjDivision result of time
Figure BDA0002563201800000093
Will be subordinate to 1 st to kjThe user loads of the user groups are respectively added to obtain kjHistorical electrical load vectors for the group of users.
Step S32, for kjHistorical electricity load vectors of the group user groups are defined, wherein the total electricity load of the a-th group user group is
Figure BDA0002563201800000101
a＝1,2,…,kjThe total load of electricity consumption of each user group is respectively in the time period of t
Figure BDA0002563201800000102
For output, input characteristics are formed by week unique hot codes, hour numbers, loads of the same period of yesterday, loads of the same period of the previous day, loads of the same period of the previous three days and loads of the same period of the previous four three days which correspond to the t period
Figure BDA0002563201800000103
Using training set LtrCorresponding TtrFeatures in individual time periods and corresponding outputs
Figure BDA0002563201800000104
t＝1,2,…,TtrTraining kjThe quantile regression neural network is respectively recorded as
Figure BDA0002563201800000105
Wherein the input vector of the neural network corresponding to the a-th group
Figure BDA0002563201800000106
Length of and
Figure BDA0002563201800000107
are the same length.
It should be noted that the week-only encoding described in this embodiment applies a single-hot encoding technique, which is an encoding technique for discrete features, also called bit-efficient encoding, by using N-bit boolean variables to encode N states. For example, assuming that there are 3 values for a discrete feature, i.e., 1, 2, and 3, the one-hot codes for those three states are 001, 010, and 100, respectively, and occupy 3-bit boolean variables.
Step S33, integrating the collection LenCorresponding TenIn a time period, for time T (T ═ T)tr+1,Ttr+2,…,Ttr+Ten) To input features
Figure BDA0002563201800000108
Respectively as quantile regression neural network
Figure BDA0002563201800000109
Is input to obtain kjLoad probability distribution of individual user groups; .
Step S34, calculating k according to preset discrete convolutionjAccumulating the probability distribution of the individual loads to obtain a prediction model of a plurality of probability prediction results of the overall load of the area to be predicted, and setting 1fj,q(Xj,t) As the number of clusters is kjThe resulting prediction of the q quantile of the total load, fj,qFor the prediction model after convolution accumulation of each population, Xj,tFor the input features of the model at time t, q may be (0, 1)]And satisfies the following formula:
Pr(Lt<fj,q(Xj,t))＝q。
s4, calculating pinball loss functions of the minimized integration set according to historical load data of the integration set, constructing a linear programming problem and solving optimal integration weights according to each prediction model and a real overall load value by taking the pinball loss functions of the minimized integration set as a target function and the integration weights of each prediction model as optimization variables, and accumulating each prediction model by the optimal integration weights to obtain an integrated regional load probability prediction model to be predicted.
Specifically, for kjIntegrating the probability distribution results of N total loads obtained under N values, and setting the integration coefficient as wjPinball loss function at q quantiles
Figure BDA00025632018000001010
The following were used:
Figure BDA0002563201800000111
wherein
Figure BDA0002563201800000112
And (3) providing a q quantile prediction result for the probability prediction model, wherein y is a real overall load value, and an optimization problem is constructed through the following formula:
Figure BDA0002563201800000113
Figure BDA0002563201800000114
Figure BDA0002563201800000115
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002563201800000116
is a set of quantiles for the probability prediction,
Figure BDA0002563201800000117
for integration set LenThe set of time periods of (a), wherein,
Figure BDA0002563201800000118
the problem objective function is a convex function, constraint conditions are linear, the optimal integration weight is obtained after solving based on a convex optimization problem solving technology
Figure BDA0002563201800000119
The prediction models under different clustering numbers are subjected to the optimal integration weight obtained in the step S41
Figure BDA00025632018000001110
Accumulating to obtain an integrated load probability prediction model of the area to be predicted:
Figure BDA00025632018000001111
wherein, fen,qRepresenting integrated probabilistic predictive models, Xt＝[X1,t,X2,t,…,XN,t]And representing the input characteristics of the integrated model at the time t.
And S5, carrying out probability prediction on the load concentrated to be predicted according to the load probability prediction model of the area to be predicted.
In the embodiment of the invention, the short-term load probability prediction method realizes the utilization of the fine-granularity user load recorded by the intelligent electric meter, establishes the short-term coincidence probability prediction method based on user clustering and ensemble learning, fully utilizes the information provided by the intelligent electric meter, considers the cluster characteristics of the user load, divides the users by using clustering, realizes the probability load prediction from bottom to top and can predict the probability distribution of the regional load with high accuracy compared with the prior art. By the method, the load condition at the future moment can be better estimated, so that the safety and economy of system operation scheduling are improved, and the method is beneficial to improving the operation organization benefits of power grid companies, power distribution companies and power markets, so that the method has important practical significance and good application prospect.
In summary, according to the short-term load probability prediction method provided by the embodiment of the invention, historical load data recorded by all user intelligent electric meters in the area to be predicted is obtained, and the historical load data is divided into a training set, an integration set and a set to be predicted according to a preset proportion. Then, aiming at historical load data in a training set, a weekly average load curve of each user is calculated, a similarity matrix of the weekly average load curve among different users is calculated based on cosine similarity, the division results of the users under a plurality of different clustering quantities are calculated according to a spectral clustering method based on k-means clustering and the similarity matrix, probability prediction models of a quantile regression neural network are respectively trained for each user group after clustering according to the division results, and output probability distribution results of the probability prediction models are accumulated through convolution to respectively obtain prediction models of a plurality of probability prediction results of the overall load of the area to be predicted. And then according to historical load data of the integrated set, calculating a pinball loss function of the minimized integrated set, according to each prediction model and a real overall load value, taking the pinball loss function of the minimized integrated set as a target function, taking the integrated weight of each prediction model as an optimization variable, constructing a linear programming problem, solving an optimal integrated weight, and accumulating each prediction model by the optimal integrated weight to obtain an integrated to-be-predicted regional load probability prediction model. And finally, carrying out probability prediction on the concentrated loads to be predicted according to the load probability prediction model of the area to be predicted. Therefore, historical load data of the user intelligent electric meter can be utilized in a refined mode, the accuracy of load probability prediction of the area to be predicted is improved, and the power generation operation scheduling of the power grid can be planned in advance.
Fig. 2 is a block schematic diagram of a short-term load probability prediction device according to one embodiment of the present invention.
As shown in fig. 2, a short-term load probability prediction apparatus 1000 according to an embodiment of the present invention includes: a partitioning module 100, a first calculation module 200, a second calculation module 300, a third calculation module 400 and a prediction module 500.
The dividing module 100 is configured to obtain historical load data recorded by all user smart electric meters in an area to be predicted, and divide the historical load data into a training set, an integration set, and a set to be predicted according to a preset proportion.
The first calculating module 200 is configured to calculate a weekly average load curve of each user according to historical load data in a training set, calculate a similarity matrix of the weekly average load curve between different users based on cosine similarity, and calculate a partitioning result of the users under a plurality of different clustering numbers according to a spectral clustering method based on k-means clustering and the similarity matrix, where any point on the weekly average load curve is a load average value of the same user in the training set at the same time in each week.
The second calculation module 300 is configured to train a probability prediction model of the quantile regression neural network for each user group after clustering according to the partition result, and add up the output probability distribution results of the probability prediction model through convolution, so as to obtain a prediction model of multiple probability prediction results of the overall load of the region to be predicted.
The third calculation module 400 is configured to calculate a pinball loss function of the minimized integration set according to the historical load data of the integration set, construct a linear programming problem and solve an optimal integration weight according to each prediction model and the real overall load value, with the pinball loss function of the minimized integration set as a target function and the integration weight of each prediction model as an optimization variable, and accumulate each prediction model with the optimal integration weight to obtain an integrated load probability prediction model of the area to be predicted.
The prediction module 500 is configured to perform probability prediction on the load in the to-be-predicted set according to the load probability prediction model of the to-be-predicted region.
It should be noted that, details that are not disclosed in the short-term load probability prediction apparatus according to the embodiment of the present invention are referred to details that are disclosed in the short-term load probability prediction method according to the embodiment of the present invention, and detailed description thereof is omitted here.
In summary, the device for predicting the short-term load probability of the embodiment of the invention obtains the historical load data recorded by all the user intelligent electric meters in the area to be predicted through the dividing module, and divides the historical load data into a training set, an integration set and a set to be predicted according to the preset proportion. The method comprises the steps of calculating a week average load curve of each user according to historical load data in a training set through a first calculation module, calculating a similarity matrix of the week average load curves among different users based on cosine similarity, and calculating division results of the users under different clustering quantities according to a spectral clustering method based on k-means clustering and the similarity matrix. And respectively training a probability prediction model of a quantile regression neural network for each clustered user group according to the division result through a second calculation module, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the area to be predicted. And calculating a pinball loss function of the minimized integration set according to historical load data of the integration set through a third calculation module, constructing a linear programming problem and solving an optimal integration weight by taking the pinball loss function of the minimized integration set as a target function and the integration weight of each prediction model as an optimization variable according to each prediction model and a real overall load value, and accumulating each prediction model by the optimal integration weight to obtain an integrated load probability prediction model of the area to be predicted. And performing probability prediction on the load in the to-be-predicted concentration through a prediction module according to the load probability prediction model of the to-be-predicted region. Therefore, historical load data of the user intelligent electric meter can be utilized in a refined mode, the accuracy of load probability prediction of the area to be predicted is improved, and the power generation operation scheduling of the power grid can be planned in advance.
In order to implement the foregoing embodiments, the present invention further provides an electronic device, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for predicting the probability of short-term load of the foregoing embodiments.
According to the electronic equipment provided by the embodiment of the invention, the processor executes the computer program stored on the memory, so that the historical load data of the user intelligent electric meter can be finely utilized, the accuracy of load probability prediction of the area to be predicted is improved, and the advance plan of power generation operation scheduling of a power grid is facilitated.
In order to implement the foregoing embodiments, the present invention also proposes a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the short-term load probability prediction method of the foregoing embodiments.
According to the non-transitory computer-readable storage medium provided by the embodiment of the invention, through executing the stored computer program, the historical load data of the user intelligent electric meter can be finely utilized, the accuracy of load probability prediction of the area to be predicted is improved, and the advance plan of power generation operation scheduling of a power grid is facilitated.
In the description of the present invention, it is to be understood that the terms "central," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," "circumferential," and the like are used in the orientations and positional relationships indicated in the drawings for convenience in describing the invention and to simplify the description, and are not intended to indicate or imply that the referenced device or element must have a particular orientation, be constructed and operated in a particular orientation, and are not to be considered limiting of the invention.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "connected," "secured," and the like are to be construed broadly and can, for example, be fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; either directly or indirectly through intervening media, either internally or in any other relationship. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
In the present invention, unless otherwise expressly stated or limited, the first feature "on" or "under" the second feature may be directly contacting the first and second features or indirectly contacting the first and second features through an intermediate. Also, a first feature "on," "above," and "over" a second feature may be directly on or obliquely above the second feature, or simply mean that the first feature is at a higher level than the second feature. A first feature being "under," "below," and "beneath" a second feature may be directly under or obliquely under the first feature, or may simply mean that the first feature is at a lesser elevation than the second feature.
In the description of the specification, reference to the description of "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.

Claims (7)

1. A method for probabilistic prediction of short-term load, comprising:
step S1, acquiring historical load data recorded by all user intelligent electric meters in the area to be predicted, and dividing the historical load data into a training set, an integration set and a set to be predicted according to a preset proportion;
step S2, calculating the weekly average load curve of each user according to the historical load data in the training set, calculating the similarity matrix of the weekly average load curve among different users based on cosine similarity, and calculating the division results of the users under a plurality of different clustering quantities according to the spectral clustering method based on k-means clustering and the similarity matrix, wherein any point on the weekly average load curve is the load average value of the same user in the training set at the same time of each week;
step S3, respectively training a probability prediction model of a quantile regression neural network for each clustered user group according to the division result, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the region to be predicted; wherein the content of the first and second substances,
the step S3 specifically includes:
step S31, for k clustersjDivision result of time
Figure FDA00036389804100000110
Will be subordinate to 1 st to kjThe user loads of the user groups are respectively added to obtain kjHistorical electricity load vectors of the group user group;
step S32, aiming at the kjHistorical electricity load vectors of the group user groups are defined, wherein the total electricity load of the a-th group user group is
Figure FDA0003638980410000011
The total load of electricity consumption of each user group in t time period
Figure FDA0003638980410000012
For output, input characteristics are formed by a unique-hot-code of the week, a number of hours, a load of the same time period of yesterday, a load of the previous time period of yesterday, a load of the same time period of the previous day, a load of the same time period of the previous three days and a load of the same time period of the fourth day corresponding to the t time period
Figure FDA0003638980410000013
Using training set LtrCorresponding TtrFeatures in individual time periods and corresponding outputs
Figure FDA0003638980410000014
Training kjThe quantile regression neural network is respectively recorded as
Figure FDA0003638980410000015
Wherein the input vector of the neural network corresponding to the a-th group
Figure FDA0003638980410000016
Length of and
Figure FDA0003638980410000017
are the same in length;
step S33, integrating the collection LenCorresponding TenIn a time period, for time T (T ═ T)tr+1,Ttr+2,…,Ttr+Ten) To input features
Figure FDA0003638980410000018
Respectively as quantile regression neural network
Figure FDA0003638980410000019
Is input to obtain kjLoad probability distribution of individual user groups;
step S34, calculating k according to preset discrete convolutionjAccumulating the probability distribution of the loads to obtain a prediction model of a plurality of probability prediction results of the whole load of the area to be predicted, and setting fj,q(Xj,t) For clustering number kjThe resulting prediction of the q quantile of the total load, fj,qFor the prediction model after convolution accumulation for each population,
Figure FDA0003638980410000021
for the input features of the model at time t, q may be (0, 1)]And satisfies the following formula:
Pr(Lt<fj,q(Xj,t))＝q，Ltthe total load of the area to be predicted at the moment t;
step S4, calculating pinball loss functions of a minimized integration set according to historical load data of the integration set, constructing a linear programming problem and solving optimal integration weights according to each prediction model and a real overall load value, taking the pinball loss functions of the minimized integration set as objective functions and the integration weights of each prediction model as optimization variables, accumulating each prediction model by the optimal integration weights to obtain an integrated to-be-predicted regional load probability prediction model, wherein the objective functions comprise convex functions, constraint conditions of the convex functions are linear, the objective functions are solved based on a convex optimization problem solving technology, the optimal integration weights are obtained after solving, the prediction models under different clustering numbers are accumulated according to the optimal integration weights to obtain the integrated to-be-predicted regional load probability prediction model, the integrated load probability prediction model of the region to be predicted has the expression:
Figure FDA0003638980410000022
wherein f isen,qRepresenting integrated probabilistic predictive models, Xt＝[X1,t,X2,t,…,XN,t]Representing the input characteristics of the integrated model at the time t, wherein N is a positive integer, and j is a positive integer less than or equal to N;
and step S5, carrying out probability prediction on the load in the to-be-predicted set according to the to-be-predicted regional load probability prediction model.
2. The method of claim 1, wherein M is the number of users in the area to be predicted, T is the number of time periods to obtain historical load, and T is the number of historical load acquisition periods0Recording the frequency of the load for the smart meter every day, wherein M, T and T0Are positive integers, and the step S1 specifically includes:
step S11, converting the historical load data into a historical load matrix L according to a preset conversion algorithmM×TAnd is provided with LtFor the total load of the area to be predicted at time t, and set Lm,tThe load of the mth user at the time T is shown, wherein T is a positive integer less than or equal to T, and M is a positive integer less than or equal to M;
step S12, the historical load matrix LM×TAccording to the sequence of time, dividing the training set into training sets LtrIntegration set LenAnd a to-be-predicted set, and a training set LtrHas a time length of TtrIntegration set LenHas a time length of Ten。
3. The method for predicting the probability of the short-term load according to claim 2, wherein the step S2 specifically comprises:
step S21, training set LtrThe load in the system is averaged according to the week to obtain the week characteristic load of each user
Figure FDA0003638980410000031
Wherein T' is less than or equal to 7T0And form a cycle characteristic load matrix
Figure FDA0003638980410000032
The week characteristic loadVector of m-th row of matrix
Figure FDA0003638980410000038
The weekly average power load of the mth user;
step S22, according to the cycle characteristic load matrix, calculating a similarity matrix S between different users based on cosine similarityM×MThe similarity matrix SM×MRow m and column n of element Sm,nFor the electricity utilization similarity between the mth user and the nth user:
Figure FDA0003638980410000033
step S23, according to the spectral clustering method based on k-means clustering and the similarity matrix SM×MDividing M users into k groups to obtain a clustered user group division result;
in step S24, k is N ═ log2M]+1 value [ ·]To round down a function:
kj＝min{2j-1,M}
wherein N is a positive integer, and j is a positive integer less than or equal to N;
step S25, repeating step S23 to obtain N groups of different division results
Figure FDA0003638980410000037
4. A method for probabilistic prediction of short term load according to claim 3, characterized in that the pinball loss function is constructed by:
to k is pairedjIntegrating the probability distribution results of N total loads obtained under N values, and setting the integration coefficient as wjPinball loss function at q quantiles
Figure FDA0003638980410000034
The following were used:
Figure FDA0003638980410000035
wherein
Figure FDA0003638980410000036
And (3) providing a q quantile prediction result for the probability prediction model, wherein y is a real overall load value, and an optimization problem is constructed through the following formula:
Figure FDA0003638980410000041
Figure FDA0003638980410000042
Figure FDA0003638980410000043
wherein the content of the first and second substances,
Figure FDA0003638980410000044
is a set of quantiles for the probability prediction,
Figure FDA0003638980410000045
as an integration set LenThe set of time periods of (a), wherein,
Figure FDA0003638980410000046
the problem objective function is a convex function, constraint conditions are linear, the optimal integration weight is obtained after solving based on a convex optimization problem solving technology
Figure FDA0003638980410000047
5. A device for predicting the probability of a short-term load, comprising:
the system comprises a dividing module, a prediction module and a prediction module, wherein the dividing module is used for acquiring historical load data recorded by all user intelligent electric meters in an area to be predicted and dividing the historical load data into a training set, an integration set and a set to be predicted according to a preset proportion;
the first calculation module is used for calculating a weekly average load curve of each user according to historical load data in the training set, calculating a similarity matrix of the weekly average load curves among different users based on cosine similarity, and calculating division results of the users under different clustering quantities according to a spectral clustering method based on k-means clustering and the similarity matrix, wherein any point on the weekly average load curve is the load average value of the same user in the training set at the same moment in each week;
the second calculation module is used for respectively training a probability prediction model of a quantile regression neural network for each user group after clustering according to the division result, and performing convolution accumulation on output probability distribution results of the probability prediction model to respectively obtain a prediction model of a plurality of probability prediction results of the overall load of the area to be predicted; wherein the content of the first and second substances,
the second calculation module is specifically applied to:
for a number of clusters of kjDivision result of time
Figure FDA0003638980410000048
Will be subordinate to 1 st to kjThe user loads of the user groups are respectively added to obtain kjHistorical electricity load vectors of the group user group;
for the kjHistorical electricity load vectors of the group user groups are defined, wherein the total electricity load of the a-th group user group is
Figure FDA0003638980410000049
The total load of electricity consumption of each user group in t time period
Figure FDA00036389804100000410
For output, input characteristics are formed by a unique-hot-code of the week, a number of hours, a load of the same time period of yesterday, a load of the previous time period of yesterday, a load of the same time period of the previous day, a load of the same time period of the previous three days and a load of the same time period of the fourth day corresponding to the t time period
Figure FDA00036389804100000411
Using training set LtrCorresponding TtrFeatures in individual time periods and corresponding outputs
Figure FDA0003638980410000051
Training kjThe quantile regression neural network is respectively recorded as
Figure FDA0003638980410000052
Wherein the input vector of the neural network corresponding to the a-th group
Figure FDA0003638980410000053
Length of and
Figure FDA0003638980410000054
are the same in length;
in the integration set LenCorresponding TenIn a time period, for time T (T ═ T)tr+1,Ttr+2,…,Ttr+Ten) To input features
Figure FDA0003638980410000055
Regression neural networks as quantiles respectively
Figure FDA0003638980410000056
Is input to obtain kjLoad probability distribution of individual user groups;
according to preset discrete convolution operation pair kjAccumulating the probability distribution of the individual loads to obtain a plurality of overall loads of the area to be predictedPrediction model of probability prediction result, let fj,q(Xj,t) For clustering number kjThe resulting prediction of the q quantile of the total load, fj,qFor the prediction model after convolution accumulation for each population,
Figure FDA0003638980410000057
for the input features of the model at time t, q may be (0, 1)]And satisfies the following formula:
Pr(Lt<fj,q(Xj,t))＝q，Ltthe total load of the area to be predicted at the moment t;
a third calculation module, configured to calculate a pinball loss function of a minimized integration set according to historical load data of the integration set, construct a linear programming problem and solve an optimal integration weight according to each prediction model and a real overall load value, with the pinball loss function of the minimized integration set as an objective function, with an integration weight of each prediction model as an optimization variable, accumulate each prediction model with the optimal integration weight to obtain an integrated to-be-predicted regional load probability prediction model, where the objective function includes a convex function, constraint conditions of the convex function are linear, solve the objective function based on a convex optimization problem solution technique, obtain the optimal integration weight after solution, accumulate prediction models in different cluster numbers according to the optimal integration weight to obtain the integrated to-be-predicted regional load probability prediction model, the integrated load probability prediction model of the area to be predicted has the expression:
Figure FDA0003638980410000058
wherein f isen,qRepresenting integrated probabilistic predictive models, Xt＝[X1,t,X2,t,…,XN,t]Representing the input characteristics of the integrated model at the time t, wherein N is a positive integer, and j is a positive integer less than or equal to N;
and the prediction module is used for carrying out probability prediction on the load in the to-be-predicted set according to the to-be-predicted regional load probability prediction model.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the method of probabilistic prediction of short term load as claimed in any of claims 1 to 4.
7. A non-transitory computer readable storage medium having stored thereon a computer program, the program being executable by a processor to implement a method of probabilistic prediction of short term load as claimed in any one of claims 1 to 4.
CN202010614204.XA 2020-06-30 2020-06-30 Probability prediction method and probability prediction device for short-term load Active CN111860977B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010614204.XA CN111860977B (en) 2020-06-30 2020-06-30 Probability prediction method and probability prediction device for short-term load

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010614204.XA CN111860977B (en) 2020-06-30 2020-06-30 Probability prediction method and probability prediction device for short-term load

Publications (2)

Publication Number Publication Date
CN111860977A CN111860977A (en) 2020-10-30
CN111860977B true CN111860977B (en) 2022-07-01

Family

ID=72989113

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010614204.XA Active CN111860977B (en) 2020-06-30 2020-06-30 Probability prediction method and probability prediction device for short-term load

Country Status (1)

Country Link
CN (1) CN111860977B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111768023B (en) * 2020-05-11 2024-04-09 国网冀北电力有限公司电力科学研究院 Probability peak load estimation method based on smart city electric energy meter data
CN112270454B (en) * 2020-11-19 2022-09-02 国网北京市电力公司 Method and device for predicting short-term load of power system under influence of extreme factors
CN112308337A (en) * 2020-11-19 2021-02-02 国网北京市电力公司 Prediction method, prediction device and processor for probabilistic short-term load of power system
CN112766537B (en) * 2020-12-24 2023-06-06 沈阳工程学院 Short-term electric load prediction method
CN112926801B (en) * 2021-03-31 2022-11-01 云南电网有限责任公司 Load curve combined prediction method and device based on quantile regression
CN112801428B (en) * 2021-04-08 2021-07-13 国网江苏省电力有限公司苏州供电分公司 Probability early warning-based lightning loss prevention control method
CN113592528A (en) * 2021-06-22 2021-11-02 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
CN114511058B (en) * 2022-01-27 2023-06-02 国网江苏省电力有限公司泰州供电分公司 Load element construction method and device for electric power user portrait
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

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108846517A (en) * 2018-06-12 2018-11-20 清华大学 A kind of probability short-term electric load prediction integrated approach of quantile
CN109978201A (en) * 2017-12-27 2019-07-05 深圳市景程信息科技有限公司 Probability load prediction system and method based on Gaussian process quantile estimate model

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200111174A1 (en) * 2018-10-04 2020-04-09 Yishen Wang Probabilistic Load Forecasting via Point Forecast Feature Integration

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109978201A (en) * 2017-12-27 2019-07-05 深圳市景程信息科技有限公司 Probability load prediction system and method based on Gaussian process quantile estimate model
CN108846517A (en) * 2018-06-12 2018-11-20 清华大学 A kind of probability short-term electric load prediction integrated approach of quantile

Also Published As

Publication number Publication date
CN111860977A (en) 2020-10-30

Similar Documents

Publication Publication Date Title
CN111860977B (en) Probability prediction method and probability prediction device for short-term load
CN111428816B (en) Non-invasive load decomposition method
CN110619420B (en) Attention-GRU-based short-term residential load prediction method
CN113126019B (en) Remote estimation method, system, terminal and storage medium for error of intelligent ammeter
CN110895773A (en) DBN power grid load prediction method and device based on generalized demand side resources
CN109492748A (en) A kind of Mid-long term load forecasting method for establishing model of the electric system based on convolutional neural networks
CN112834927A (en) Lithium battery residual life prediction method, system, device and medium
CN112288328A (en) Energy internet risk assessment method based on gray chromatography
Kong et al. Refined peak shaving potential assessment and differentiated decision-making method for user load in virtual power plants
CN115759336A (en) Prediction method and storage medium for short-term power load prediction
CN111291782B (en) Accumulated load prediction method based on information accumulation k-Shape clustering algorithm
CN115759393A (en) Cumulative load baseline prediction method based on ensemble learning
CN115545333A (en) Method for predicting load curve of multi-load daily-type power distribution network
CN108898273A (en) A kind of user side load characteristic clustering evaluation method based on morphological analysis
CN108879656B (en) Short-term power load prediction method based on sub-sampling SVR integration
CN111144447A (en) Power grid peak-valley time interval division method for preventing peak regulation risk caused by new energy output
Bi et al. Accurate water quality prediction with attention-based bidirectional LSTM and encoder–decoder
CN114154684A (en) Short-term photovoltaic power prediction method based on data mining and multi-core support vector machine
CN111553434A (en) Power system load classification method and system
CN116826710A (en) Peak clipping strategy recommendation method and device based on load prediction and storage medium
CN114839586B (en) Low-voltage station metering device misalignment calculation method based on EM algorithm
CN115577996A (en) Risk assessment method, system, equipment and medium for power grid power failure plan
CN110717779A (en) Electric power transaction system, method and application thereof
CN113962440A (en) DPC and GRU fused photovoltaic prediction method and system
CN113469420A (en) Electric power energy structure evaluation optimization method for multi-element power supply 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