CN114707772A - Power load prediction method and system based on multi-feature decomposition and fusion - Google Patents

Power load prediction method and system based on multi-feature decomposition and fusion Download PDF

Info

Publication number
CN114707772A
CN114707772A CN202210627288.XA CN202210627288A CN114707772A CN 114707772 A CN114707772 A CN 114707772A CN 202210627288 A CN202210627288 A CN 202210627288A CN 114707772 A CN114707772 A CN 114707772A
Authority
CN
China
Prior art keywords
neural network
power load
feature
data
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210627288.XA
Other languages
Chinese (zh)
Other versions
CN114707772B (en
Inventor
李雪梅
袁晨迅
王梓颖
张彩明
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.)
Shandong University
Original Assignee
Shandong 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 Shandong University filed Critical Shandong University
Priority to CN202210627288.XA priority Critical patent/CN114707772B/en
Publication of CN114707772A publication Critical patent/CN114707772A/en
Application granted granted Critical
Publication of CN114707772B publication Critical patent/CN114707772B/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
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas 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

Abstract

The invention relates to the field of data processing of prediction purposes, and discloses a power load prediction method and a system based on multi-feature decomposition and fusion, wherein the method comprises the following steps: acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed; extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting features of the short-term fluctuation feature matrix to obtain a local feature matrix; performing feature fusion on the global features, the local features and the historical power load time sequence data; and inputting the fusion characteristics into a prediction model to obtain a power load prediction result. The error of the power load data prediction is obviously reduced.

Description

Power load prediction method and system based on multi-feature decomposition and fusion
Technical Field
The invention relates to the field of data processing for prediction purposes, in particular to a power load prediction method and system based on multi-feature decomposition and fusion.
Background
The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.
With the continuous development of the science and technology level and the improvement of the economic level, the electric power becomes one of important energy sources for development, the electric power load can be effectively predicted, the production of the electric power can be reasonably planned, a more scientific power utilization mode can be formulated, the prevention of emergencies in the electric power industry can be facilitated, the development of the electric power in regions can be better served, and therefore the economic benefit of the society can be improved.
The prediction of the electrical load experiences a change from a traditional timing model to a machine-learned timing model. Conventional timing models such as differential autoregressive moving average model have been widely used for prediction of power load data, such as application publication No.: CN109615117A patent, patent No.: CN104794549 uses a differential auto-regressive moving average model to predict the power load, but the traditional time sequence model cannot capture the non-linear factors in the data, so the prediction result is poor. Following the development of artificial intelligence related technologies, the timing sequence model of machine learning better solves the shortcomings of the traditional timing sequence model, such as application publication number: CN114298408A, CN114254828A, etc. all utilize a neural network to predict power load data, but some existing power load predictions using a machine learning method still have the following problems:
(1) changes in the power load are affected by factors such as local economic conditions, electricity prices, and weather, and the use of time series data of the power load alone is not sufficient to reflect the effective characteristics affecting the changes in the power load.
(2) The power load data is a sequence with characteristics such as nonlinearity, noise and the like, and the direct use of the original sequence for prediction can lead to the mixed feature in the power load time sequence data, thereby causing the difficulty in extracting the features by a neural network. Although some time series decomposition methods such as ensemble empirical mode decomposition and variational mode decomposition are widely applied to power load prediction at present, most methods perform prediction by using a neural network after performing one-time decomposition on power load data, so that the decomposition can use all data sets, cause leakage of future data, and are not feasible in practical application work.
(3) The problem of gradient disappearance exists when the traditional cyclic neural networks such as a long-short term memory neural network, a gated cyclic unit neural network, a time convolution neural network and the like are directly used, long-time sequence prediction is poor, and the influence degree of changes of different characteristics on power load is difficult to extract.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides a power load prediction method and a system based on multi-feature decomposition and fusion; the error of the power load data prediction is obviously reduced.
In a first aspect, the invention provides a power load prediction method based on multi-feature decomposition and fusion;
the power load prediction method based on multi-feature decomposition and fusion comprises the following steps:
acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed;
extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting the features of the short-term fluctuation feature matrix to obtain a local feature matrix; performing feature fusion on the global features, the local features and the historical power load time sequence data;
and inputting the fusion characteristics into a prediction model to obtain a power load prediction result.
In a second aspect, the invention provides a power load prediction system based on multi-feature decomposition and fusion;
the power load prediction system based on multi-feature decomposition and fusion comprises:
an acquisition module configured to: acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed;
a feature extraction module configured to: extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting features of the short-term fluctuation feature matrix to obtain a local feature matrix;
a feature fusion module configured to: performing feature fusion on the global features, the local features and the historical power load time sequence data;
a prediction module configured to: and inputting the fusion characteristics into a prediction model to obtain a power load prediction result.
Compared with the prior art, the invention has the beneficial effects that:
the method adds economic data and weather data which influence the power change as important features, utilizes the improved time convolution neural network to extract the long-term influence of the economic change on the power change and the long-term trend features in the power load data, utilizes the convolution-bidirectional long-short term memory neural network with time attention to extract the influence of different states of weather on the power load and the short-term fluctuation features in the power load data, and reduces the fluctuation of weather and economic development on a power load prediction model; by utilizing the variational modal window decomposition method, the leakage of future data of a power load sequence caused by one-time decomposition is prevented, the internal characteristics of the original data of the power load are separated, and the power economic index characteristic influencing the long-term change of the power and the weather index characteristic influencing the short-term change of the power, which are constructed by the method, are respectively combined, so that the characteristics can be conveniently and efficiently extracted by a neural network, and the error of power load data prediction is remarkably reduced.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
Fig. 1 is a flowchart of a power load prediction method based on multi-feature decomposition and fusion according to an embodiment of the present disclosure.
Fig. 2 is a flowchart of a global feature extraction module according to an embodiment of the present application.
Fig. 3 is a flowchart of a local feature extraction module according to an embodiment of the present application.
Fig. 4 is a structural diagram of a time convolution neural network of a multi-scale attention mechanism proposed by a global feature extraction module according to an embodiment of the present application.
FIG. 5 is a block diagram of a bidirectional long-short term memory recurrent neural network with a time attention mechanism for a local feature extraction module according to an embodiment of the present invention.
Detailed Description
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, and it should be understood that the terms "comprises" and "comprising", and any variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
All data are obtained according to the embodiment and are legally applied on the data on the basis of compliance with laws and regulations and user consent.
Example one
The embodiment provides a power load prediction method based on multi-feature decomposition and fusion;
as shown in fig. 1, the power load prediction method based on multi-feature decomposition and fusion includes:
s101: acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed;
s102: extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting features of the short-term fluctuation feature matrix to obtain a local feature matrix; performing feature fusion on the global features, the local features and the historical power load time sequence data;
s103: and inputting the fusion characteristics into a prediction model to obtain a power load prediction result.
Further, the historical macro-economic data comprises: a Gross Regional Product (GRP), a Regional Consumer Price Index (RCPI), a Regional industry total Product (RGIP), and a power selling Price (REP).
Illustratively, some macro-economic data for the area of interest is obtained, including: the regional production total value (GRP), the regional Resident Consumption Price Index (RCPI), the regional industry total value (RGIP) and the regional electricity selling price (REP) are provided, and the economic data only provide monthly data, because macroscopic economic data indexes such as the regional production total value and the like are horizontal values reflecting a certain aspect of the current period of the society, and the daily data have no specific practical significance. However, in order to correspond to the dimension of the daily data of the power load subsequently and facilitate the dimension reduction processing, the monthly data is converted into the daily data by using a linear interpolation method. Therefore, the overall trend of data distribution is guaranteed to be unchanged, and data dimension correspondence is achieved. An electric power economy data characteristic matrix defined as formula (1)
Figure 820296DEST_PATH_IMAGE001
Wherein
Figure 362136DEST_PATH_IMAGE002
Is the length of the power load timing data.
Figure 861251DEST_PATH_IMAGE003
（1）
Further, the historical weather indicator data includes: daily maximum temperature, minimum temperature, average temperature, relative humidity and rainfall for the target area.
Illustratively, the Maximum Temperature (MXT), Minimum Temperature (MNT), Average Temperature (AT), Relative Humidity (RH), and Rainfall (RF) weather indicators of the study area per day are obtained, and defined as the weather feature matrix of equation (2)
Figure 691803DEST_PATH_IMAGE004
。
Figure 419719DEST_PATH_IMAGE005
（2）
Further, according to the obtained data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed; the method specifically comprises the following steps:
s1011: performing dimensionality reduction on the macroscopic economic data to obtain an electric power economic index; performing dimensionality reduction on the weather index data to obtain a weather index;
s1012: decomposing historical power load time sequence data into a low-frequency characteristic sequence and a plurality of high-frequency characteristic sequences;
s1013: combining the power economic index with the low-frequency characteristic sequence to obtain a characteristic matrix reflecting the long-term trend; combining the weather index with the high-frequency characteristic sequence to obtain a characteristic matrix reflecting short-term fluctuation;
s1014: and carrying out normalization processing on the historical power load time sequence data, the feature matrix of the long-term trend and the feature matrix of the short-term fluctuation.
Illustratively, the dimension reduction processing of S1011 is performed by principal component analysis.
Illustratively, the decomposing of the historical power load time sequence data of S1012 is performed by using a variational modal window decomposition algorithm.
Exemplarily, S1011: performing dimensionality reduction on the macroscopic economic data to obtain an electric power economic index; performing dimensionality reduction on the weather index data to obtain a weather index; the method specifically comprises the following steps:
feature matrix to be acquired
Figure 347224DEST_PATH_IMAGE006
Using principal component analysis PCA to perform dimensionality reduction treatment, and compiling the dimensionality reduction treatment into an electric power economic index
Figure 17239DEST_PATH_IMAGE007
And weather index
Figure 131826DEST_PATH_IMAGE008
The method comprises the following steps:
firstly, the original feature matrix is expressed by formula (3)
Figure 115963DEST_PATH_IMAGE009
And (6) normalizing to obtain a normalized matrix Z.
Figure 394846DEST_PATH_IMAGE010
（3）
Wherein the content of the first and second substances,
Figure 235763DEST_PATH_IMAGE011
the number of lines is represented,
Figure 837645DEST_PATH_IMAGE012
the number of columns is shown,
Figure 687790DEST_PATH_IMAGE013
the values of the original feature matrix are represented,
Figure 809461DEST_PATH_IMAGE014
、
Figure 821279DEST_PATH_IMAGE015
representing each component
Figure 644878DEST_PATH_IMAGE016
Mean and standard deviation of.
And then, the initial characteristic value is extracted by using a formula (4).
Figure 236397DEST_PATH_IMAGE017
（4）
Wherein
Figure 461842DEST_PATH_IMAGE018
Representation matrix
Figure 660873DEST_PATH_IMAGE019
And obtaining a correlation coefficient matrix of
Figure 237348DEST_PATH_IMAGE020
Characteristic value
Figure 163716DEST_PATH_IMAGE021
。
Selecting principal components with characteristic values larger than 1, and calculating a score coefficient matrix of the original data for the principal components with characteristic values larger than 1 according to formula (5)
Figure 243667DEST_PATH_IMAGE022
（
Figure 66130DEST_PATH_IMAGE023
The number of principal components whose eigenvalues are greater than 1).
Figure 615054DEST_PATH_IMAGE024
（5）
Wherein the content of the first and second substances,
Figure 345112DEST_PATH_IMAGE025
is a load of the main component, and is,
Figure 545150DEST_PATH_IMAGE026
representing the characteristic value corresponding to each main component. After the score coefficient matrix is calculated, each main component value is calculated by using an equation (6), and then a final power economic index is calculated by using an equation (7)
Figure 335251DEST_PATH_IMAGE027
。
Figure 105892DEST_PATH_IMAGE028
（6）
Figure 842904DEST_PATH_IMAGE029
（7）
Wherein, the first and the second end of the pipe are connected with each other,
Figure 897448DEST_PATH_IMAGE030
is shown as
Figure 592871DEST_PATH_IMAGE031
The main components of the composition are as follows,
Figure 631234DEST_PATH_IMAGE032
denotes the first
Figure 453828DEST_PATH_IMAGE033
The characteristic value corresponding to each principal component.
Calculating weather index
Figure 362878DEST_PATH_IMAGE034
I.e., repeating equations (3) - (7).
Further, the S1012: decomposing historical power load time sequence data into a low-frequency characteristic sequence and a plurality of high-frequency characteristic sequences; the method specifically comprises the following steps:
and decomposing the power load time sequence data by using a variational modal window decomposition method.
Further, decomposing the power load time sequence data by using a variational modal window decomposition method; the method specifically comprises the following steps:
the method comprises the following steps of cutting an input power load sequence into blocks by adopting a sliding window, wherein the cut power load sequence is as follows:
Figure 494782DEST_PATH_IMAGE035
Figure 489283DEST_PATH_IMAGE036
wherein, the first and the second end of the pipe are connected with each other,
Figure 568097DEST_PATH_IMAGE037
to set the length of the decomposition window, equation (8) is reused
Figure 347966DEST_PATH_IMAGE038
Decomposing each group of values;
Figure 385192DEST_PATH_IMAGE039
（8）
wherein the content of the first and second substances,
Figure 398147DEST_PATH_IMAGE040
the time is represented by the time of day,
Figure 811811DEST_PATH_IMAGE041
is the original sequence of the electrical load,
Figure 446186DEST_PATH_IMAGE042
is the number of modes that are to be used,
Figure 654313DEST_PATH_IMAGE043
is a function of the dirichlet allocation function,
Figure 826669DEST_PATH_IMAGE044
which represents a convolution of the signals of the first and second,
Figure 44023DEST_PATH_IMAGE045
，
Figure 47751DEST_PATH_IMAGE046
is the partial derivative.
Finally, a decomposed low-frequency sequence is obtained
Figure 443092DEST_PATH_IMAGE047
And high frequency sequence
Figure 899481DEST_PATH_IMAGE048
；
Figure 654947DEST_PATH_IMAGE049
Figure 716444DEST_PATH_IMAGE050
Figure 531953DEST_PATH_IMAGE051
Figure 960792DEST_PATH_IMAGE052
Figure 785528DEST_PATH_IMAGE053
Figure 498269DEST_PATH_IMAGE054
Figure 484680DEST_PATH_IMAGE055
Figure 118924DEST_PATH_IMAGE056
Figure 966925DEST_PATH_IMAGE057
Figure 534173DEST_PATH_IMAGE058
Figure 957064DEST_PATH_IMAGE059
Figure 609762DEST_PATH_IMAGE060
Figure 261454DEST_PATH_IMAGE061
Figure 948788DEST_PATH_IMAGE062
Figure 480263DEST_PATH_IMAGE063
Figure 620257DEST_PATH_IMAGE064
It should be understood that the prior art is to sequence the electrical loads
Figure 324908DEST_PATH_IMAGE065
Decomposing with formula (8) to obtain
Figure 617480DEST_PATH_IMAGE066
A discrete pattern, each pattern
Figure 116595DEST_PATH_IMAGE066
Component of
Figure 743885DEST_PATH_IMAGE067
Each of
Figure 189910DEST_PATH_IMAGE067
Centered at the center frequency of each eigenmode function component
Figure 851836DEST_PATH_IMAGE068
Nearby.
The component obtained by decomposition is a group of vector values, and the low-frequency sequence of the decomposition is expressed as
Figure 538163DEST_PATH_IMAGE069
；
Load power sequence
Figure 387170DEST_PATH_IMAGE071
First three values of
Figure 433624DEST_PATH_IMAGE072
Decomposing to obtain low-frequency sequence
Figure 950056DEST_PATH_IMAGE073
It is clear that, in the case of a,
Figure 994235DEST_PATH_IMAGE074
the first three numerical values of
Figure 81271DEST_PATH_IMAGE075
Are not equal to each other, explain
Figure 665836DEST_PATH_IMAGE074
The set of components is based on
Figure 302354DEST_PATH_IMAGE076
The whole sequence is obtained by decomposition,
Figure 314172DEST_PATH_IMAGE074
the local values will still contain some information of the entire sequence. So is directly used
Figure 154083DEST_PATH_IMAGE074
The prediction of the subsequent input neural network has low error effect, but the prediction cannot be decomposed in practical application.
Further, the S1013: combining the power economic index with the low-frequency characteristic sequence to obtain a characteristic matrix reflecting the long-term trend; combining the weather index with the high-frequency characteristic sequence to obtain a characteristic matrix reflecting short-term fluctuation; the method specifically comprises the following steps:
economic index of electric power
Figure 276760DEST_PATH_IMAGE077
And low frequency sequence of power load data
Figure 971047DEST_PATH_IMAGE078
Characteristic matrix spliced in parallel into long-term trend
Figure 153766DEST_PATH_IMAGE079
；
Weather index to be constructed
Figure 464662DEST_PATH_IMAGE080
And power load data high frequency sequence
Figure 130043DEST_PATH_IMAGE081
Respectively connected in parallel to form a feature matrix with short-term fluctuation
Figure 741153DEST_PATH_IMAGE082
。
Further, the S1014: carrying out normalization processing on historical power load time sequence data, a characteristic matrix reflecting long-term trend and a characteristic matrix of short-term fluctuation; the method specifically comprises the following steps:
long term trend feature matrix
Figure 94774DEST_PATH_IMAGE083
And short term fluctuation characteristics
Figure 96228DEST_PATH_IMAGE084
Electric power load sequence
Figure 826287DEST_PATH_IMAGE085
Carrying out normalization processing;
Figure 777056DEST_PATH_IMAGE086
（9）
wherein the content of the first and second substances,
Figure 567158DEST_PATH_IMAGE087
in order to be the normalized data, the data,
Figure 852646DEST_PATH_IMAGE088
is the original data.
Further, the S102: extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; the method comprises the following specific steps:
and (4) extracting the features of the feature matrix of the long-term trend by adopting a global feature extraction module.
As shown in fig. 2, the network structure of the global feature extraction module includes:
the system comprises a time convolution neural network, a decoding layer, a first full-connection layer, a second full-connection layer and an output layer which are connected in sequence.
The time convolution neural network has two advantages, namely, the time convolution neural network effectively relieves the problem of gradient disappearance of a predicted long-time sequence through a causal convolution part, and the strong multilayer convolution kernel can efficiently extract local information of the sequence.
The time convolution neural network adopts a multi-scale attention mechanism, and the specific structure is shown in fig. 4.
The time convolution neural network of the multi-scale attention mechanism comprises a self-attention mechanism layer, a first expansion convolution layer, a second expansion convolution layer and a third expansion convolution layer which are sequentially connected;
the first expansion convolution layer is provided with a first attention mechanism module;
the second expansion coiling layer is provided with a second attention mechanism module;
the third expansion convolution layer is provided with a third attention mechanism module;
the output end of the self-attention mechanism layer is connected with the input end of the first expansion convolution layer in a residual error mode;
the output end of the first expansion convolutional layer is connected with the input end of the second expansion convolutional layer in a residual error mode;
and the output end of the second expansion convolutional layer is connected with the input end of the third expansion convolutional layer through residual errors.
The traditional time convolution neural network carries out convolution operation to the same degree on the whole sequence, and cannot extract influences on different characteristics.
Here, a feature matrix is input
Figure 386395DEST_PATH_IMAGE089
The objective is to extract long-term trend characteristics of power load changes. The time convolution neural network implementation formula of the multi-scale attention mechanism is shown in formulas (10) to (14):
further, the working principle of the self-attention mechanism layer is as follows:
Figure 644201DEST_PATH_IMAGE090
（10）
Figure 90357DEST_PATH_IMAGE091
（11-1）
Figure 863141DEST_PATH_IMAGE092
（11-2）
Figure 200582DEST_PATH_IMAGE093
（11-3）
wherein, the formula (10) is
Figure 109632DEST_PATH_IMAGE094
The self-attention mechanism is used for processing, namely the self-attention mechanism is used for endowing the internal information of the feature matrix with different weights, and subsequent neural networks are used for feature extraction and remote learning, so that the feature extraction efficiency is improved.
Wherein the content of the first and second substances,
Figure 992268DEST_PATH_IMAGE095
are respectively provided with
Figure 986769DEST_PATH_IMAGE096
The same tensor, as shown in equations (11-1), (11-2) and (11-3),
Figure 65583DEST_PATH_IMAGE097
the parameter matrices are represented separately and are,
Figure 829140DEST_PATH_IMAGE098
represent
Figure 131945DEST_PATH_IMAGE099
The feature dimension of (a) is,
Figure 630054DEST_PATH_IMAGE100
representing an activation function.
Then will obtain
Figure 43718DEST_PATH_IMAGE101
Inputting the feature matrix into the attention machineIn the time convolution neural network, an attention mechanism is added to each expansion convolution layer, so that weight assignment can be carried out on a convolution operation result in the characteristic extraction process, the simplification operation of the conventional time convolution neural network on sequence extraction characteristics is improved, and the characteristic extraction capability of the time convolution neural network is remarkably improved.
Figure 927360DEST_PATH_IMAGE102
（12）
Equation (12) represents the feature extraction process of the time convolution neural network, wherein
Figure 604329DEST_PATH_IMAGE103
Which represents the kernel of the convolution,
Figure 573422DEST_PATH_IMAGE104
for the number of convolution kernels to be,
Figure 541509DEST_PATH_IMAGE105
in order to be the coefficient of expansion,
Figure 545237DEST_PATH_IMAGE106
show that
Figure 189845DEST_PATH_IMAGE107
The feature matrix before the time instant.
Further, the residual concatenation refers to:
Figure 646235DEST_PATH_IMAGE108
（13）
further, the working principle of the first attention mechanism module, the second attention mechanism module and the third attention mechanism module is consistent; wherein the working principle of the first attention mechanism module is as follows:
Figure 152433DEST_PATH_IMAGE109
（14）
equation (13) represents the residual block portions, each performed twice
Figure 213930DEST_PATH_IMAGE110
Transforming, activating functions
Figure 763860DEST_PATH_IMAGE111
The ReLU activation function is used.
Equation (14) represents the final feature result obtained using the time attention mechanism. Wherein
Figure 707546DEST_PATH_IMAGE112
、
Figure 266703DEST_PATH_IMAGE113
In the form of a matrix of parameters,
Figure 995756DEST_PATH_IMAGE114
a vector of the offset is represented, and,
Figure 982166DEST_PATH_IMAGE115
in order to activate the function(s),
Figure 350831DEST_PATH_IMAGE116
is the training data length.
A decoding layer: and carrying out head-to-tail splicing on the multidimensional output of the time convolution neural network to form a one-dimensional vector, thereby achieving the purpose of reducing the dimension.
Extracting features by a time convolution neural network of a multi-scale attention mechanism, then changing the features into one dimension by a decoding layer, and finally outputting the extracted global features by two full-connection layers
Figure 448100DEST_PATH_IMAGE117
。
Further, performing feature extraction on the feature matrix of the short-term fluctuation; the method specifically comprises the following steps:
and (3) extracting features of the feature matrix of the short-term fluctuation by adopting a convolution-bidirectional long-short term memory neural network with a time attention mechanism so as to mine the influence of different weather states on power change.
As shown in fig. 3, the convolutional-bidirectional long-short term memory neural network with time attention mechanism comprises:
the system comprises a first parallel-connected one-dimensional convolutional neural network with four different receptive fields, a second parallel-connected one-dimensional convolutional neural network with four different receptive fields, an aggregation layer, a maximum pooling layer, a bidirectional long-short term memory neural network, a time attention mechanism layer, a full connection layer and an output layer which are connected in sequence.
Wherein, the one-dimensional convolution neural network of four different receptive fields of first parallelly connected includes: the first convolutional neural network, the second convolutional neural network, the third convolutional neural network and the fourth convolutional neural network are connected in parallel;
wherein, the one-dimensional convolution neural network of four different receptive fields of second parallel connection includes: the fifth convolutional neural network, the sixth convolutional neural network, the seventh convolutional neural network and the eighth convolutional neural network are connected in parallel;
the output end of the first convolutional neural network is connected with the input end of the fifth convolutional neural network; the output end of the fifth convolutional neural network is connected with the input end of the aggregation layer;
the output end of the second convolutional neural network is connected with the input end of the sixth convolutional neural network; the output end of the sixth convolutional neural network is connected with the input end of the aggregation layer;
the output end of the third convolutional neural network is connected with the input end of the seventh convolutional neural network; the output end of the seventh convolutional neural network is connected with the input end of the aggregation layer;
the output end of the fourth convolutional neural network is connected with the input end of the eighth convolutional neural network; the output end of the eighth convolutional neural network is connected with the input end of the aggregation layer;
the polymerization layer has the working principle that input features are connected in parallel.
It should be understood that the feature matrix is
Figure 280926DEST_PATH_IMAGE118
The convolution neural networks with different receptive fields can extract more complete local information than a single one-dimensional convolution neural network, then the local information is input into the bidirectional long and short term memory neural network after passing through the maximum pooling layer, the interaction between future information and past information is increased by the bidirectional long and short term memory neural network, and the characteristics before data are more effectively memorized; aiming at different influences of the continuous change of weather on the future of the power load, the method adds a time attention mechanism after the bidirectional long-short term memory neural network extracts the characteristics, extracts the influences of the weather state on the power load at different time periods, and extracts a structural diagram of the bidirectional long-short term memory recurrent neural network with time attention, as shown in fig. 5. In the bidirectional long and short term memory neural network, updating is carried out according to the formulas (15) - (21). Wherein, the formulas (15) to (20) are steps of the long-short term memory cyclic neural network method, and the formula (21) is a step of the bidirectional long-short term memory neural network.
Figure 188971DEST_PATH_IMAGE119
（15）
Figure 107248DEST_PATH_IMAGE120
（16）
Figure 8208DEST_PATH_IMAGE121
（17）
Figure 695541DEST_PATH_IMAGE122
（18）
Figure 227017DEST_PATH_IMAGE123
（19）
Figure 117744DEST_PATH_IMAGE124
（20）
Figure 556815DEST_PATH_IMAGE125
（21）
Wherein
Figure 98655DEST_PATH_IMAGE126
The result of the processing at the input gate is shown,
Figure 863349DEST_PATH_IMAGE127
a result of the forgetting-to-gate process is indicated,
Figure 241371DEST_PATH_IMAGE128
the result of the output gate is shown,
Figure 687396DEST_PATH_IMAGE129
it is shown that the activation function is,
Figure 83743DEST_PATH_IMAGE130
、
Figure 753758DEST_PATH_IMAGE131
、
Figure 868345DEST_PATH_IMAGE132
、
Figure 665531DEST_PATH_IMAGE133
a matrix of the parameters is represented and,
Figure 916383DEST_PATH_IMAGE134
、
Figure 757300DEST_PATH_IMAGE135
、
Figure 359183DEST_PATH_IMAGE136
、
Figure 694481DEST_PATH_IMAGE137
a vector of the offset is represented, and,
Figure 330998DEST_PATH_IMAGE138
indicating the value of the hidden state at the last moment,
Figure 342817DEST_PATH_IMAGE139
which is indicative of the current state of the device,
Figure 635258DEST_PATH_IMAGE140
an input representing the current time of day is presented,
Figure 757935DEST_PATH_IMAGE141
a temporary hidden variable representing the current time of day,
Figure 734112DEST_PATH_IMAGE142
indicating the forward structural state of the cell,
Figure 182411DEST_PATH_IMAGE143
indicating the state of the backward structure of the cell,
Figure 758886DEST_PATH_IMAGE144
is the output of the bidirectional long-short term memory neural network.
Further, the working principle of the time attention mechanism layer comprises:
Figure 685253DEST_PATH_IMAGE145
（22）
Figure 781516DEST_PATH_IMAGE146
（23）
Figure 338400DEST_PATH_IMAGE147
（24）
equations (22) to (24) are the implementation equations of the time attention mechanism,
Figure 136591DEST_PATH_IMAGE148
and
Figure 866650DEST_PATH_IMAGE149
a matrix of the parameters is represented and,
Figure 66687DEST_PATH_IMAGE150
a vector of the offset is represented, and,
Figure 607521DEST_PATH_IMAGE151
to pair
Figure 893009DEST_PATH_IMAGE152
The attention weight is calculated, and finally, the probability is calculated using the attention weight as equation (23), and then weighted summation is performed using equation (24) to calculate the output.
Finally, the product is output after passing through two full connecting layers to obtain
Figure 364442DEST_PATH_IMAGE153
A local feature
Figure 418985DEST_PATH_IMAGE154
，
Figure 379988DEST_PATH_IMAGE155
Is the training data length.
Further, performing feature fusion on the global features, the local features and the historical power load time sequence data; the method specifically comprises the following steps:
load power sequence
Figure 915223DEST_PATH_IMAGE156
Global features
Figure 987084DEST_PATH_IMAGE157
And local features
Figure 161714DEST_PATH_IMAGE158
Continuously connected in parallel and spliced into a feature matrix
Figure 28039DEST_PATH_IMAGE159
；
Further, the step S103: inputting the fusion characteristics into a prediction model to obtain a power load prediction result; the method specifically comprises the following steps:
the power load prediction problem is described by equation (25):
Figure 491381DEST_PATH_IMAGE160
（25）
wherein
Figure 852086DEST_PATH_IMAGE161
In order to predict the result of the event,
Figure 881222DEST_PATH_IMAGE162
is a multi-layer perceptron model of the setting.
To better demonstrate the effectiveness of the method of the invention, the invention uses several indicators for evaluating the regression problem for the presentation of results, Mean Absolute Error (MAE), Mean Absolute Percent Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Scale Error (MASE). Corresponding to the respective formulae (26-29).
Figure 184028DEST_PATH_IMAGE163
（26）
Figure 931404DEST_PATH_IMAGE164
（27）
Figure 95800DEST_PATH_IMAGE165
（28）
Figure 979442DEST_PATH_IMAGE166
（29）
Wherein, the first and the second end of the pipe are connected with each other,
Figure 656411DEST_PATH_IMAGE167
representing a time sequence in time
Figure 625504DEST_PATH_IMAGE168
The actual value of (a) is,
Figure 577280DEST_PATH_IMAGE169
to represent
Figure 331740DEST_PATH_IMAGE170
The predicted value of (a) is determined,
Figure 241928DEST_PATH_IMAGE171
is the test set length.
This method was developed based on python 3.7. In using the variational modal window decomposition method, the invention sets the number of decompositions
Figure 698317DEST_PATH_IMAGE172
The time window of the decomposition is set to 20, wherein the first term is a low frequency trend term, and the second, third, fourth and fifth terms are high frequency trend terms. In the parameter setting of the global feature extraction module, the invention designs a time convolution neural network comprising three hidden layers, and the expansion rates are respectively set to be 1, 2 and 4; the number of convolution kernels is set to 128, 32 respectively; the window of the convolution kernel is 3. In the parameter setting of the local feature extraction module, the number of the convolution kernels of the one-dimensional convolution part is set to be 256, and the windows of the convolution kernels are respectively 1, 3, 5 and 10. The multi-layer perceptron is set to have three fully connected layers in the result prediction module.
The method comprises the following steps of: a difference Autoregressive Integrated Moving Average model (ARIMA), a time convolution Neural Network (TCN), a Bidirectional Long Short-Term Memory Neural Network (Bi-LSTM) and a Dual-Stage Attention mechanism cyclic Neural Network model (Dual-Stage Attention-Based Current Neural Network, DARNN) are compared, four indexes of the method are the lowest, and the first ranking in the comparison method proves the effectiveness of the method.
Example two
The embodiment provides a power load prediction system based on multi-feature decomposition and fusion;
the power load prediction system based on multi-feature decomposition and fusion comprises:
an acquisition module configured to: acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed;
a feature extraction module configured to: extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting features of the short-term fluctuation feature matrix to obtain a local feature matrix;
a feature fusion module configured to: performing feature fusion on the global features, the local features and the historical power load time sequence data;
a prediction module configured to: and inputting the fusion characteristics into a prediction model to obtain a power load prediction result.
It should be noted here that the above-mentioned obtaining module, the feature extracting module, the feature fusing module and the predicting module correspond to steps S101 to S103 in the first embodiment, and the above-mentioned modules are the same as the examples and application scenarios realized by the corresponding steps, but are not limited to the contents disclosed in the first embodiment. It should be noted that the modules described above as part of a system may be implemented in a computer system such as a set of computer-executable instructions.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. The power load prediction method based on multi-feature decomposition and fusion is characterized by comprising the following steps:
acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed;
extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting features of the short-term fluctuation feature matrix to obtain a local feature matrix; performing feature fusion on the global features, the local features and the historical power load time sequence data;
and inputting the fusion characteristics into a prediction model to obtain a power load prediction result.
2. The multi-feature decomposition and fusion based power load forecasting method according to claim 1, wherein the feature matrix reflecting the long-term tendency and the feature matrix of the short-term fluctuation are constructed based on the acquired data; the method specifically comprises the following steps:
performing dimensionality reduction on the macroscopic economic data to obtain an electric power economic index; performing dimensionality reduction on the weather index data to obtain a weather index;
decomposing historical power load time sequence data into a low-frequency characteristic sequence and a plurality of high-frequency characteristic sequences;
combining the power economic index with the low-frequency characteristic sequence to obtain a characteristic matrix of the long-term trend; combining the weather index with the high-frequency characteristic sequence to obtain a short-term fluctuation characteristic matrix;
and carrying out normalization processing on the historical power load time sequence data, the feature matrix of the long-term trend and the feature matrix of the short-term fluctuation.
3. The multi-feature decomposition and fusion based power load prediction method according to claim 2, wherein historical power load time series data is decomposed into a low frequency feature sequence and a plurality of high frequency feature sequences; the method specifically comprises the following steps:
and decomposing the power load time sequence data by using a variational modal window decomposition method.
4. The multi-feature decomposition and fusion based power load prediction method according to claim 2, wherein a power economy index is combined with the low frequency feature sequence to obtain a feature matrix reflecting a long-term trend; combining the weather index with the high-frequency characteristic sequence to obtain a characteristic matrix reflecting short-term fluctuation; the method specifically comprises the following steps:
economic index of electric power
Figure 475925DEST_PATH_IMAGE001
And low frequency sequence of power load data
Figure 769503DEST_PATH_IMAGE002
Characteristic matrix spliced in parallel into long-term trend
Figure 747823DEST_PATH_IMAGE003
；
Weather index to be constructed
Figure 484966DEST_PATH_IMAGE004
And power load data high frequency sequence
Figure 915947DEST_PATH_IMAGE005
Respectively connected in parallel to form a feature matrix with short-term fluctuation
Figure 13216DEST_PATH_IMAGE006
。
5. The multi-feature decomposition and fusion based power load prediction method according to claim 1, wherein feature extraction is performed on a feature matrix of a long-term trend to obtain a global feature matrix; the method specifically comprises the following steps:
a global feature extraction module is adopted to extract features of the feature matrix of the long-term trend;
the global feature extraction module is characterized in that the network structure comprises:
the system comprises a time convolution neural network, a decoding layer, a first full-connection layer, a second full-connection layer and an output layer which are connected in sequence;
the time convolution neural network of the multi-scale attention mechanism comprises a self-attention mechanism layer, a first expansion convolution layer, a second expansion convolution layer and a third expansion convolution layer which are sequentially connected;
the first expansion coiling layer is provided with a first attention mechanism module;
the second expansion convolution layer is provided with a second attention mechanism module;
the third expansion convolution layer is provided with a third attention mechanism module;
the output end of the self-attention mechanism layer is connected with the input end of the first expansion convolution layer in a residual error mode;
the output end of the first expansion convolutional layer is connected with the input end of the second expansion convolutional layer in a residual error mode;
and the output end of the second expansion convolutional layer is connected with the input end of the third expansion convolutional layer through residual errors.
6. The multi-feature decomposition and fusion based power load prediction method according to claim 1, wherein the feature extraction is performed on the feature matrix of the short-term fluctuation to obtain a local feature matrix; the method comprises the following specific steps:
a convolution-bidirectional long and short term memory neural network with a time attention mechanism is adopted to extract the characteristics of the short term fluctuation characteristic matrix so as to mine the influence of different weather states on the power change;
a convolutional-bi-directional long-short term memory neural network with a temporal attention mechanism, the network structure comprising:
the system comprises a first parallel-connected one-dimensional convolutional neural network with four different receptive fields, a second parallel-connected one-dimensional convolutional neural network with four different receptive fields, a polymerization layer, a maximum pooling layer, a bidirectional long-short term memory neural network, a time attention mechanism layer, a full connection layer and an output layer which are connected in sequence.
7. The multi-feature decomposition and fusion based power load prediction method of claim 6, wherein the first parallel one-dimensional convolutional neural network of four different receptive fields comprises: the first convolutional neural network, the second convolutional neural network, the third convolutional neural network and the fourth convolutional neural network are connected in parallel;
a second parallel one-dimensional convolutional neural network of four different receptive fields, comprising: a fifth convolutional neural network, a sixth convolutional neural network, a seventh convolutional neural network, and an eighth convolutional neural network connected in parallel.
8. The multi-feature decomposition and fusion based power load prediction method of claim 7, wherein an output terminal of the first convolutional neural network is connected to an input terminal of the fifth convolutional neural network; the output end of the fifth convolutional neural network is connected with the input end of the aggregation layer;
the output end of the second convolutional neural network is connected with the input end of the sixth convolutional neural network; the output end of the sixth convolutional neural network is connected with the input end of the aggregation layer;
the output end of the third convolutional neural network is connected with the input end of the seventh convolutional neural network; the output end of the seventh convolutional neural network is connected with the input end of the aggregation layer;
the output end of the fourth convolutional neural network is connected with the input end of the eighth convolutional neural network; and the output end of the eighth convolutional neural network is connected with the input end of the aggregation layer.
9. The multi-feature decomposition and fusion based power load prediction method according to claim 1, wherein the global features, the local features and the historical power load time series data are feature fused; the method specifically comprises the following steps:
load power sequence
Figure 783726DEST_PATH_IMAGE007
Global features
Figure 941038DEST_PATH_IMAGE008
And local features
Figure 406786DEST_PATH_IMAGE009
Continuously connected in parallel and spliced into a feature matrix
Figure 42166DEST_PATH_IMAGE010
；
Inputting the fusion characteristics into a prediction model to obtain a power load prediction result; the method specifically comprises the following steps:
the power load prediction problem is described as:
Figure 729500DEST_PATH_IMAGE011
（22）
wherein
Figure 808445DEST_PATH_IMAGE012
In order to predict the result of the event,
Figure 948440DEST_PATH_IMAGE013
is a multi-layer perceptron model of the setting.
10. The power load prediction system based on multi-feature decomposition and fusion is characterized by comprising the following components:
an acquisition module configured to: acquiring historical macroscopic economic data, historical weather index data and historical power load time sequence data of a target area; according to the acquired data, a feature matrix reflecting long-term trend and a feature matrix reflecting short-term fluctuation are constructed;
a feature extraction module configured to: extracting features of the feature matrix of the long-term trend to obtain a global feature matrix; extracting features of the short-term fluctuation feature matrix to obtain a local feature matrix;
a feature fusion module configured to: performing feature fusion on the global features, the local features and the historical power load time sequence data;
a prediction module configured to: and inputting the fusion characteristics into a prediction model to obtain a power load prediction result.
CN202210627288.XA 2022-06-06 2022-06-06 Power load prediction method and system based on multi-feature decomposition and fusion Active CN114707772B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210627288.XA CN114707772B (en) 2022-06-06 2022-06-06 Power load prediction method and system based on multi-feature decomposition and fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210627288.XA CN114707772B (en) 2022-06-06 2022-06-06 Power load prediction method and system based on multi-feature decomposition and fusion

Publications (2)

Publication Number Publication Date
CN114707772A true CN114707772A (en) 2022-07-05
CN114707772B CN114707772B (en) 2022-08-23

Family

ID=82178036

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210627288.XA Active CN114707772B (en) 2022-06-06 2022-06-06 Power load prediction method and system based on multi-feature decomposition and fusion

Country Status (1)

Country Link
CN (1) CN114707772B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115440390A (en) * 2022-11-09 2022-12-06 山东大学 Method, system, equipment and storage medium for predicting number of cases of infectious diseases
CN116362503A (en) * 2023-03-30 2023-06-30 国网河南省电力公司安阳供电公司 Electric power regulating method and system based on artificial intelligence
CN116960991A (en) * 2023-09-21 2023-10-27 杭州半云科技有限公司 Probability-oriented power load prediction method based on graph convolution network model

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104200277A (en) * 2014-08-12 2014-12-10 南方电网科学研究院有限责任公司 Modeling method for medium and long term power load forecasting
CN111738512A (en) * 2020-06-22 2020-10-02 昆明理工大学 Short-term power load prediction method based on CNN-IPSO-GRU hybrid model
WO2020252784A1 (en) * 2019-06-21 2020-12-24 西门子股份公司 Power load data prediction method and device, and storage medium
AU2020104000A4 (en) * 2020-12-10 2021-02-18 Guangxi University Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model
CN112613641A (en) * 2020-12-07 2021-04-06 河北工业大学 Short-term electric load combination prediction method based on feature decomposition
CN113011630A (en) * 2021-01-25 2021-06-22 国网浙江省电力有限公司杭州供电公司 Method for short-term prediction of space load in zone time of big data power distribution network
CN113592185A (en) * 2021-08-05 2021-11-02 四川大学 Power load prediction method based on Transformer
CN114529051A (en) * 2022-01-17 2022-05-24 杭州电子科技大学 Long-term power load prediction method based on hierarchical residual self-attention neural network

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104200277A (en) * 2014-08-12 2014-12-10 南方电网科学研究院有限责任公司 Modeling method for medium and long term power load forecasting
WO2020252784A1 (en) * 2019-06-21 2020-12-24 西门子股份公司 Power load data prediction method and device, and storage medium
CN111738512A (en) * 2020-06-22 2020-10-02 昆明理工大学 Short-term power load prediction method based on CNN-IPSO-GRU hybrid model
CN112613641A (en) * 2020-12-07 2021-04-06 河北工业大学 Short-term electric load combination prediction method based on feature decomposition
AU2020104000A4 (en) * 2020-12-10 2021-02-18 Guangxi University Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model
CN113011630A (en) * 2021-01-25 2021-06-22 国网浙江省电力有限公司杭州供电公司 Method for short-term prediction of space load in zone time of big data power distribution network
CN113592185A (en) * 2021-08-05 2021-11-02 四川大学 Power load prediction method based on Transformer
CN114529051A (en) * 2022-01-17 2022-05-24 杭州电子科技大学 Long-term power load prediction method based on hierarchical residual self-attention neural network

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
LJUBISA SEHOVAC: "Deep Learning for Load Forecasting: Sequence to Sequence Recurrent Neural Networks With Attention", 《IEEE ACCESS》 *
徐晓: "经济序列的趋势分析――基于时间序列的混合模型", 《科技风》 *
梁寿愚等: "基于双图正则非负低秩分解的电力负荷短期预测", 《计算机与现代化》 *
陈耀红等: "基于季节调整与神经网络融合算法的售电量预测", 《大众用电》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115440390A (en) * 2022-11-09 2022-12-06 山东大学 Method, system, equipment and storage medium for predicting number of cases of infectious diseases
CN116362503A (en) * 2023-03-30 2023-06-30 国网河南省电力公司安阳供电公司 Electric power regulating method and system based on artificial intelligence
CN116362503B (en) * 2023-03-30 2023-11-07 国网河南省电力公司安阳供电公司 Electric power regulating method and system based on artificial intelligence
CN116960991A (en) * 2023-09-21 2023-10-27 杭州半云科技有限公司 Probability-oriented power load prediction method based on graph convolution network model
CN116960991B (en) * 2023-09-21 2023-12-29 杭州半云科技有限公司 Probability-oriented power load prediction method based on graph convolution network model

Also Published As

Publication number Publication date
CN114707772B (en) 2022-08-23

Similar Documents

Publication Publication Date Title
CN114707772B (en) Power load prediction method and system based on multi-feature decomposition and fusion
Ebrahimian et al. The price prediction for the energy market based on a new method
Kankal et al. Modeling and forecasting of Turkey’s energy consumption using socio-economic and demographic variables
Zhu et al. Carbon price analysis using empirical mode decomposition
Bahrami et al. Short term electric load forecasting by wavelet transform and grey model improved by PSO (particle swarm optimization) algorithm
López et al. Application of SOM neural networks to short-term load forecasting: The Spanish electricity market case study
Ouyang et al. Agricultural commodity futures prices prediction via long-and short-term time series network
CN111680841B (en) Short-term load prediction method, system and terminal equipment based on principal component analysis
Neeraj et al. Long short-term memory-singular spectrum analysis-based model for electric load forecasting
Yao et al. Crude oil price prediction based on LSTM network and GM (1, 1) model
Wang et al. Long time series deep forecasting with multiscale feature extraction and Seq2seq attention mechanism
Ouyang et al. Discovery and prediction of stock index pattern via three-stage architecture of TICC, TPA-LSTM and multivariate LSTM-FCNs
Rathnayaka et al. An unbiased GM (1, 1)-based new hybrid approach for time series forecasting
Gao et al. An adaptive deep-learning load forecasting framework by integrating transformer and domain knowledge
Li et al. What can be learned from the historical trend of crude oil prices? An ensemble approach for crude oil price forecasting
Li et al. Parking demand forecasting based on improved complete ensemble empirical mode decomposition and GRU model
Gao et al. A two-layer SSA-XGBoost-MLR continuous multi-day peak load forecasting method based on hybrid aggregated two-phase decomposition
Precioso et al. Thresholding methods in non-intrusive load monitoring
Zhang et al. A novel decomposition and combination technique for forecasting monthly electricity consumption
Wang et al. A Transformer-based multi-entity load forecasting method for integrated energy systems
Ling et al. Reliability Prediction of the Distribution Network Based on Wavelet Neural Network with Quantum Particle Swarm Optimization Algorithm
CN117151276A (en) Intelligent management system of electricity selling platform
Li et al. A new grey prediction model for forecasting the automobiles ownership in China
Saini et al. Modelling particulate matter using multivariate and multistep recurrent neural networks
Tong et al. A prediction model for complex equipment remaining useful life using gated recurrent unit complex networks

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