CN113537649A - Power grid load prediction method and system based on neural network and dynamic mode decomposition - Google Patents
Power grid load prediction method and system based on neural network and dynamic mode decomposition Download PDFInfo
- Publication number
- CN113537649A CN113537649A CN202111083651.8A CN202111083651A CN113537649A CN 113537649 A CN113537649 A CN 113537649A CN 202111083651 A CN202111083651 A CN 202111083651A CN 113537649 A CN113537649 A CN 113537649A
- Authority
- CN
- China
- Prior art keywords
- power grid
- grid load
- time sequence
- data
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 56
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 40
- 239000011159 matrix material Substances 0.000 claims abstract description 77
- 238000001228 spectrum Methods 0.000 claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000010276 construction Methods 0.000 claims description 26
- 230000000737 periodic effect Effects 0.000 claims description 19
- 238000005070 sampling Methods 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 abstract description 3
- 238000004458 analytical method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000003111 delayed effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005312 nonlinear dynamic Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a power grid load prediction method and a system based on a neural network and dynamic mode decomposition, wherein the method comprises the steps of collecting original power grid load data, processing the original power grid load data and constructing a sliding window matrix, and decomposing the sliding window matrix by utilizing SVD (singular value decomposition) and DMD (digital micromirror device) to construct a power grid load linear model; constructing a time sequence of the sliding window corresponding to the power grid load value, and performing fluctuation decomposition on the time sequence of the sliding window corresponding to the power grid load value to obtain a singular index time sequence and a multi-fractal spectrum time sequence corresponding to three decomposition factors of the time sequence; performing time sequence prediction neural network training; combining the power grid load linear model and the charge deviation data output by the time sequence prediction neural network to obtain a power grid load prediction model, and predicting power grid load data through the power grid load prediction model; compared with the prior art, the method has a more interpretable algorithm structure and can find abnormal events.
Description
Technical Field
The invention relates to the technical field of artificial intelligence prediction, in particular to a power grid load prediction method and system based on a neural network and dynamic mode decomposition.
Background
The total load of the power system is the sum of total power consumed by all the electric equipment in the system; adding power consumed by industrial, agricultural, post and telecommunications, traffic, municipal, commercial and urban and rural residents to obtain comprehensive power load of a power system; the comprehensive power load plus network loss is the power supplied by each power plant in the system, and is called the power supply load of the power system, which is called the power supply amount for short; the power supply load plus the power consumed by each power plant is the plant power consumption, which is the power generated by each generator in the system, and is called the power generation load of the system, i.e. the power generation amount. Power load prediction is an important part of power management, and the load prediction data provided by it is extremely important for the control, operation and planning of power systems. The accurate prediction of the power load data not only plays an important role in determining the operation mode of the power system, but also plays an important role in determining the optimal scheduling, inter-site power transmission scheme and load scheduling scheme of the power system. In addition, the accuracy of power load prediction also directly affects the safety, reliability, economy and power quality of the operation of the power system, and is related to the production planning and scheduling operation of the power system. Also in the past decade, the increasing market competition, the aging of infrastructure, and the integration of renewable energy requirements have made load prediction not only more important, but also more difficult.
The national scholars have conducted extensive and intensive research on the load prediction theory of the power system for a long time, and a plurality of effective methods such as a regression analysis method, a time series method, a neural network method, a wavelet analysis method and the like are provided. For a prediction problem, various prediction methods can be established. Different prediction methods provide different prediction information and different prediction accuracy.
The prediction method does not consider the fluctuation of the power grid load, the fluctuation of the power grid load is generated by many driving factors, such as different weather, seasonal conditions, holidays, work periods, fluctuation of economic properties and the like, and data generated by the factors have a fine time pattern, so that the power grid load data is poor in prediction accuracy and low in robustness.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a power grid load prediction method and system based on a neural network and dynamic mode decomposition, which have high precision and strong robustness.
In order to achieve the purpose, the invention adopts the following technical scheme:
a power grid load prediction method based on a neural network and dynamic mode decomposition specifically comprises the following steps:
collecting original power grid load data;
converting the original power grid load data into a high-dimensional space to obtain high-dimensional power grid load data;
constructing a sliding window matrix according to the high-dimensional power grid load data;
decomposing each sliding window matrix by using singular value decomposition and dynamic mode decomposition to construct a power grid load linear model;
obtaining power grid load deviation data by subtracting the original power grid load data from the power grid load data obtained by the power grid load linear model;
acquiring a trend factor, a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a randomness factor according to the sliding window matrix;
using a singular index time sequence and a multi-fractal spectrum time sequence corresponding to the trend factor, the periodic fluctuation factor and the randomness factor as input data of time sequence prediction neural network training, and using power grid load deviation data as output data of the time sequence prediction neural network training to perform neural network training;
and constructing a power grid load prediction model by using the power grid load linear model and the predicted power grid load deviation data output by the time sequence prediction neural network, and acquiring the predicted power grid load data according to the power grid load prediction model.
Further, the construction method of the sliding window matrix comprises the following steps:
the collected original charge data are promoted to a delay coordinate space by constructing a Hankel matrix H; wherein the expression of the matrix H is:
in the formula: the shape of H is [ m-d, d ], m is the number of load points, d is the dimension of delay embedding;
sampling data in the matrix H by a sliding window to construct a sliding window matrix, wherein the expression of the sliding window matrix is as follows:
in the formula:
representing the nth column load data of the H matrix.
Further, the expression of the power grid load linear model is as follows:
in the formula: a is a linear operator matrix.
Further, the power grid load linear model is optimized, specifically:
obtaining a linear operator matrix according to the power grid load linear model, wherein the expression of the linear operator matrix is as follows:
extracting truncated rank r matrixes
Data and truncating the rank r matrices
Projecting the data to eigen-orthogonal decomposition mode arranged according to the sequence of the eigenvectors to obtain an approximate matrix
；
From approximation matrices
Obtaining an optimized power grid load linear model, wherein the expression of the optimized power grid load linear model is as follows:
。
further, the method for acquiring the singular index time sequence and the fractal spectrum time sequence corresponding to the trend factor, the periodic fluctuation factor and the stochastic factor decomposition factor comprises the following steps:
extracting a power grid load value sequence of each moment of each sliding window matrix, arranging the power grid load value sequences in time, acquiring a time sequence of the power grid load values corresponding to the sliding windows, performing fluctuation decomposition on the time sequence of the power grid load values corresponding to the sliding windows, acquiring a trend component, a periodic component and a remainder component, and analyzing and acquiring a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a trend factor, a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a random factor in each component by utilizing multi-fractal trend fluctuation.
Further, the expression of the power grid load prediction model is as follows:
in the formula:
is the drive signal.
A power grid load prediction system based on a neural network and dynamic mode decomposition comprises an original data acquisition module, an original data processing module, a sliding window matrix construction module, a power grid load linear model construction module, a singular index time sequence and multi-fractal spectrum time sequence construction module, a time sequence prediction neural network training module, a power grid load prediction model construction module and a power grid load data prediction module;
the original data acquisition module is used for acquiring original power grid load data;
the original data processing module is used for converting original power grid load data into a high-dimensional space to obtain high-dimensional power grid load data;
the sliding window matrix construction module is used for sampling the sliding window of the high-dimensional power grid load data and constructing a sliding window matrix;
the power grid load linear model building module is used for decomposing the sliding window matrix by using singular value decomposition and dynamic mode decomposition to build a power grid load linear model;
the singular index time sequence and multi-fractal spectrum time sequence construction module is used for acquiring a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a trend factor, a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a randomness factor in the sliding window according to the sliding window matrix;
the time sequence prediction neural network training module is used for taking a singular index time sequence and a multi-fractal spectrum time sequence corresponding to the tendency factor, the periodic fluctuation factor and the randomness factor as input data of time sequence prediction neural network training, taking power grid load deviation data as output data of the time sequence prediction neural network training, carrying out the neural network training and obtaining a time sequence prediction neural network;
the power grid load prediction model construction module is used for constructing a power grid load prediction model by utilizing a power grid load linear model and prediction power grid load deviation data output by a time sequence prediction neural network;
and the power grid load data prediction module is used for predicting the power grid load data through the power grid load prediction model.
The invention has the beneficial effects that:
1. according to the method, the original power grid load data is decomposed in a high-dimensional space through a dynamic mode to obtain a power grid load linear model, so that abnormal events can be found more accurately;
2. the method is pure data-driven prediction, and has low cost and strong implementation;
3. the method learns the forcing signal of the designated training window in an unsupervised mode through the time sequence prediction neural network, and finally effectively reflects the fluctuation condition and the fluctuation irregularity degree of the original sequence through the singular index and the multi-fractal spectrum of the time sequence decomposition factor of the power grid load deviation matrix, so that the neural network learns better characteristics, the network can construct the power grid load deviation more accurately, and the power grid load data prediction is more accurate and stable.
Drawings
FIG. 1 is a block flow diagram of the method of the present invention;
fig. 2 is a schematic diagram of the system of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and examples.
In the description of the present invention, it is to be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplicity of description, and do not indicate or imply that the referenced devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and thus, are not to be construed as limiting the present invention.
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 otherwise specified.
Example 1
The application environment of the embodiment of the invention is as follows: the data generated by different weather, seasonal conditions, holidays, work periods, economic property fluctuation and the like have fine time patterns, and all the factors can cause the load prediction of the power grid to deviate, so that the accuracy of the prediction result is low, and the stability is low.
As shown in fig. 1, the present embodiment provides a power grid load prediction method based on a neural network and dynamic pattern decomposition, which specifically includes the following steps:
the method comprises the following steps: and acquiring original power grid load data through observation data.
Step two: according to the Cover theorem, the original power grid load data is subjected to linear model representation in a delay coordinate space, and the original power grid load data in the delay coordinate space is subjected to sliding window sampling to construct a sliding window matrix. The specific construction method comprises the following steps:
constructing a Hankel matrix H, and promoting the power grid load data to a delay coordinate space for linear model representation, wherein the expression of the matrix H is as follows:
in the formula: the shape of H is [ m-d, d ], m is the number of load points, and d is the dimension of the delay embedding. The delay embedding time is long enough to contain the periodicity of the data, so d is determined according to the sampling period of the grid load data. Data superposition time is promoted to a delay coordinate space through delay embedding, and data dimension is promoted, so that a prediction model is more suitably and accurately expressed.
Sampling data in the matrix H by a sliding window to construct a sliding window matrix, wherein the expression of the sliding window matrix is as follows:
in the formula:
representing the nth column load data of the H matrix.
Step three: and decomposing the sliding window matrix by using singular value decomposition and dynamic mode decomposition to construct a power grid load linear model. The specific construction method comprises the following steps:
decomposing the sliding window matrix X by using singular value decomposition, wherein the decomposition expression is as follows:
。
in the formula: u is a left singular matrix, and the delay embedding can generate a principal component track left singular matrix; each column of the U-matrix represents the principal component of the delay coordinate space, considered as a learned time-frequency basis, representing the observed dynamics in low rank,
for focusing matrix， 
Representing the transpose of the right singular matrix V.
Fidelity is optimal in the least-squares sense by boosting the dimensionality recompression signal to a low rank time-frequency representation, i.e., the dynamics observed by the low rank representation.
Decomposing the sliding window matrix by using dynamic mode decomposition, wherein the decomposition expression is as follows:
The method can be extended to learning-driven linear models by delayed coordinate dynamic pattern decomposition, which are quite efficient even with highly non-linear dynamics.
The singular value decomposition and dynamic mode decomposition expression are combined to calculate a linear operator matrix A, and the expression of the linear operator matrix A is as follows:
(ii) a From the above formula, the matrix a contains a large amount of data and is usually a high-dimensional matrix, and the obtained high-dimensional matrix contains part of redundant information, which results in an excessively long calculation time when performing correlation calculation.
Thus, extracting r matrices of truncated rank
Data and truncating the rank r matrices
Projecting the data to eigen-orthogonal decomposition mode arranged according to the sequence of the eigenvectors to obtain an approximate matrix
Where the truncation rank r is constant.
From approximation matrices
Acquiring an optimized power grid load linear model generated by a dynamic mode decomposition method, wherein the expression of the optimized power grid load linear model is as follows:
。
the present embodiment can be extended to learning-driven linear models by a delayed coordinate dynamic pattern decomposition method, which is quite efficient even for highly non-linear dynamics. The linear model generated by the dynamic mode decomposition method has the advantage that the dynamics can be understood by the eigenvalues and eigenvectors of the matrix, with interpretability.
Although the mathematical process of identifying dynamic pattern decomposition patterns and feature values is purely linear, the system itself may be nonlinear, as evidenced by the Koopman operator theoretical basis that a nonlinear system may be described by a set of pattern and feature value pairs.
Step four: and (4) subtracting the original power grid load data from the power grid load data obtained by the power grid load linear model to obtain power grid load deviation data.
The expression of the power grid load deviation data is as follows:
。
in the formula:
in the form of the original load data,
and decomposing the power grid load data obtained by the power grid load linear model for the dynamic mode.
And the power grid load data obtained by decomposing the power grid load linear model in the dynamic mode corresponds to the matrix data of each sliding window one to one.
Step five: and acquiring a trend factor, a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to the randomness factor according to the sliding window matrix.
Because the sliding window matrix is a time delay matrix, and data is redundant for the time prediction neural network, all unique elements in the sliding window matrix, namely the power grid load value at each moment, are extracted, and a time sequence is formed according to the time sequence, namely the time sequence of the power grid load value corresponding to the sliding window is obtained.
And then carrying out fluctuation decomposition on the power grid load value time sequence corresponding to the sliding window, wherein the optimal method for fluctuation decomposition adopts STL decomposition, and the STL has the advantages of insensitivity to abnormal values and robustness. The specific decomposition method is as follows:
STL decomposes time-series data Y at a certain time into trend components based on a local weighted regression method
Periodic component of
And remainder
The expression of the time-series data Y is:
and then, for each component, carrying out analysis and judgment by adopting multi-fractal detrended fluctuation.
Firstly, sequence components are analyzed by utilizing multi-fractal detrending fluctuation:
the flow of the multi-fractal detrending fluctuation analysis is as follows, and the specific calculation details are well known and will not be described herein.
Step 1: is provided with
For a time series, calculating
And constructing a new sequence X (k) according to the accumulated deviation of the mean value.
Step 2: will be provided with
Divided into non-overlapping lengths of
Is/are as follows
A plurality of equal-length subsequences. To ensure that sequence information is not lost, then the slave is
The tail end is divided forward once to obtain
A sub-sequence.
And step 3: and performing polynomial construction on each section of data by using a least square method, and calculating a corresponding variance mean value.
And 4, step 4: is determined for
Of subsequences
An order fluctuation function.
And 5: calculate the first
Order of ripple function
And time scale
And obtaining a generalized Hurst index through the power law relation between the two.
Step 6: calculating singular index
And multi-fractal spectra
。
The value of q is when
When the temperature of the water is higher than the set temperature,
mainly reflects the large fluctuation condition in the signal time sequence; when in use
When the temperature of the water is higher than the set temperature,
mainly reflecting the small fluctuation conditions in the signal time series. And the value empirical value Start of q is-3, the Step length (Step) is 0.2, the cutoff is 3.2(Stop), and the total length is 31.
The resulting singular index and fractal multiplet spectra are 30 lengths, and q =0 is deleted because it does not converge here.
Multifractal spectroscopy
The internal dynamics of the time series can be finely characterized.
Singular index
The size of (a) determines the degree of smoothness or irregularity of the wave process in a certain part.
The decomposition factor is a combination of the following three factors:
one is a trending factor, which is a contributing factor to cover the entire time series.
The second is a periodic fluctuation factor, which is an influence factor that appears and disappears repeatedly at a certain time interval.
Finally, a randomness factor, which is a time-independent influencing factor.
Therefore, the singular index and the multi-fractal spectrum sequence of the trend factor, the periodic fluctuation factor and the randomness factor are finally obtained. The fluctuation condition and fluctuation irregularity degree of the original sequence can be effectively reflected through the singular indexes and the multi-fractal spectrum of the three sequences, and further, the neural network can learn better characteristics.
Step five: and then training a time sequence prediction neural network on the singular index time sequence and the multi-fractal spectrum time sequence of the three sequences.
The time-series prediction neural network can adopt
Etc. neural networks.
Taking TCN as an example, the detailed steps of training are as follows:
the implementer will have the TCN last terminating FC, i.e., the fully connected layer, to output the prediction target.
The input data is a singular index time sequence and a multi-fractal spectrum time sequence extracted from a sliding window time sequence of the power grid load sliding matrix X, and the output data is F.
And (3) extracting the characteristics of the singular index time sequence and the multi-fractal spectrum data of the window through a time convolution neural network, and fitting the load deviation value of the power grid by using a fully-connected network.
The input shape of TCN is [ B, N, 6 ]]The output shape is [ B,1 ]]B is
N represents the length of the singular index and the multifractal spectrum,
represents the sum of 3 singular indexes and 3 fractal spectral values, and 1 represents 1 load deviation value.
The loss function uses the mean square error.
And finally, fitting the power grid load deviation value according to the variation trend through the TCN.
Step six: and combining the power grid load linear model obtained in the third step and the power grid load deviation data output by the time sequence prediction neural network obtained in the fifth step to obtain a power grid load prediction model.
Wherein the combined model expression is as follows:
；
in the formula:
for the drive signal, the model is a forced linear model, so that all model errors are present in the drive signal, and can be used to determine the differenceA common event.
And predicting the power grid load data through the power grid load prediction model.
In summary, in the embodiment, real-time power grid load data is constructed into a forced linear model, power grid load early warning information is provided for workers, the workers can conveniently understand, control and predict a power grid system, and the method has the advantages of high prediction accuracy and high robustness.
Example 2
As shown in fig. 2, this embodiment provides a power grid load prediction system based on a neural network and dynamic mode decomposition, and the system includes an original data acquisition module, an original data processing module, a sliding window matrix construction module, a power grid load linear model construction module, a singular index time sequence and multi-fractal spectrum time sequence construction module, a time sequence prediction neural network training module, a power grid load prediction model construction module, and a power grid load data prediction module.
The original data acquisition module is used for acquiring original power grid load data.
And the original data processing module is used for converting the original power grid load data into a high-dimensional space to obtain the high-dimensional power grid load data.
The sliding window matrix construction module is used for sampling the sliding window of the high-dimensional power grid load data and constructing a sliding window matrix;
and the power grid load linear model building module is used for decomposing the sliding window matrix by using singular value decomposition and dynamic mode decomposition to build a power grid load linear model.
The singular index time sequence and multi-fractal spectrum time sequence construction module is used for acquiring a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a trend factor, a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a randomness factor in the sliding window according to the sliding window matrix.
The time sequence prediction neural network training module is used for carrying out time sequence prediction on input data of the neural network by using the singular index time sequence and the multi-fractal spectrum time sequence corresponding to the tendency factor, the periodic fluctuation factor and the randomness factor, and taking power grid load deviation data as output data of the time sequence prediction neural network to carry out neural network training.
And the power grid load prediction model construction module is used for acquiring a power grid load prediction model according to the power grid load linear model and the prediction power grid deviation data output by the time sequence prediction neural network.
And the power grid load data prediction module is used for predicting the power grid load data through the power grid load prediction model.
The above embodiments are merely illustrative of the present invention, and should not be construed as limiting the scope of the present invention, and all designs identical or similar to the present invention are within the scope of the present invention.
Claims (7)
1. A power grid load prediction method based on a neural network and dynamic mode decomposition is characterized by specifically comprising the following steps:
collecting original power grid load data;
converting the original power grid load data into a high-dimensional space to obtain high-dimensional power grid load data;
constructing a sliding window matrix according to the high-dimensional power grid load data;
decomposing each sliding window matrix by using singular value decomposition and dynamic mode decomposition to construct a power grid load linear model;
obtaining power grid load deviation data by subtracting the original power grid load data from the power grid load data obtained by the power grid load linear model;
acquiring a trend factor, a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a randomness factor according to the sliding window matrix;
using a singular index time sequence and a multi-fractal spectrum time sequence corresponding to the trend factor, the periodic fluctuation factor and the randomness factor as input data of time sequence prediction neural network training, and using power grid load deviation data as output data of the time sequence prediction neural network training to perform neural network training;
and constructing a power grid load prediction model by using the power grid load linear model and the predicted power grid load deviation data output by the time sequence prediction neural network, and acquiring the predicted power grid load data according to the power grid load prediction model.
2. The power grid load prediction method based on the neural network and the dynamic pattern decomposition as claimed in claim 1, wherein the construction method of the sliding window matrix is as follows:
the collected original charge data are promoted to a delay coordinate space by constructing a Hankel matrix H; wherein the expression of the matrix H is:
in the formula: the shape of H is [ m-d, d ], m is the number of load points, d is the dimension of delay embedding;
sampling data in the matrix H by a sliding window to construct a sliding window matrix, wherein the expression of the sliding window matrix is as follows:
in the formula:
representing the nth column load data of the H matrix.
3. The power grid load prediction method based on the neural network and the dynamic pattern decomposition as claimed in claim 1, wherein the expression of the power grid load linear model is as follows:
in the formula: a is a linear operator matrix.
4. The power grid load prediction method based on the neural network and the dynamic mode decomposition according to claim 3, wherein the power grid load linear model is optimized, specifically:
obtaining a linear operator matrix according to the power grid load linear model, wherein the expression of the linear operator matrix is as follows:
extracting truncated rank r matrixes
Data and truncating the rank r matrices
Projecting the data to eigen-orthogonal decomposition mode arranged according to the sequence of the eigenvectors to obtain an approximate matrix
；
From approximation matrices
Obtaining an optimized power grid load linear model, wherein the expression of the optimized power grid load linear model is as follows:
。
5. the power grid load prediction method based on the neural network and the dynamic mode decomposition as claimed in claim 1, wherein the method for acquiring the singular index time sequence and the multi-fractal spectrum time sequence corresponding to the trend factor, the periodic fluctuation factor and the stochastic factor decomposition factor comprises:
extracting a power grid load value sequence of each moment of each sliding window matrix, arranging the power grid load value sequences in time, acquiring a time sequence of the power grid load values corresponding to the sliding windows, performing fluctuation decomposition on the time sequence of the power grid load values corresponding to the sliding windows, acquiring a trend component, a periodic component and a remainder component, and analyzing and acquiring a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a trend factor, a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a randomness factor corresponding to each component by utilizing multi-fractal trend fluctuation.
6. The power grid load prediction method based on the neural network and the dynamic pattern decomposition as claimed in claim 1, wherein the power grid load prediction model has an expression as follows:
in the formula: is the drive signal.
7. A power grid load prediction system based on a neural network and dynamic mode decomposition is characterized by comprising an original data acquisition module, an original data processing module, a sliding window matrix construction module, a power grid load linear model construction module, a singular index time sequence and multi-fractal spectrum time sequence construction module, a time sequence prediction neural network training module, a power grid load prediction model construction module and a power grid load data prediction module;
the original data acquisition module is used for acquiring original power grid load data;
the original data processing module is used for converting original power grid load data into a high-dimensional space to obtain high-dimensional power grid load data;
the sliding window matrix construction module is used for sampling the sliding window of the high-dimensional power grid load data and constructing a sliding window matrix;
the power grid load linear model building module is used for decomposing the sliding window matrix by using singular value decomposition and dynamic mode decomposition to build a power grid load linear model;
the singular index time sequence and multi-fractal spectrum time sequence construction module is used for acquiring a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a trend factor, a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a periodic fluctuation factor and a singular index time sequence and a multi-fractal spectrum time sequence corresponding to a randomness factor in the sliding window according to the sliding window matrix;
the time sequence prediction neural network training module is used for taking a singular index time sequence and a multi-fractal spectrum time sequence corresponding to the tendency factor, the periodic fluctuation factor and the randomness factor as input data of time sequence prediction neural network training, taking power grid load deviation data as output data of the time sequence prediction neural network training, carrying out the neural network training and obtaining a time sequence prediction neural network;
the power grid load prediction model construction module is used for constructing a power grid load prediction model by utilizing a power grid load linear model and prediction power grid load deviation data output by a time sequence prediction neural network;
and the power grid load data prediction module is used for predicting the power grid load data through the power grid load prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111083651.8A CN113537649B (en) | 2021-09-16 | 2021-09-16 | Power grid load prediction method and system based on neural network and dynamic mode decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111083651.8A CN113537649B (en) | 2021-09-16 | 2021-09-16 | Power grid load prediction method and system based on neural network and dynamic mode decomposition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113537649A true CN113537649A (en) | 2021-10-22 |
| CN113537649B CN113537649B (en) | 2022-02-22 |
Family
ID=78093275
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111083651.8A Active CN113537649B (en) | 2021-09-16 | 2021-09-16 | Power grid load prediction method and system based on neural network and dynamic mode decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537649B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115158399A (en) * | 2022-06-14 | 2022-10-11 | 通号城市轨道交通技术有限公司 | Time sequence signal abnormity detection method and system |
| 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 |
| CN116631429A (en) * | 2023-07-25 | 2023-08-22 | 临沂金诺视讯数码科技有限公司 | Voice and video processing method and system based on VOLTE call |
| CN116739383A (en) * | 2023-06-30 | 2023-09-12 | 浙江东鸿电子股份有限公司 | Charging pile power load prediction evaluation method based on big data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160098052A1 (en) * | 2014-09-12 | 2016-04-07 | Tsinghua University | Partition method and device for power system |
| CN105760952A (en) * | 2016-02-15 | 2016-07-13 | 国网山东省电力公司电力科学研究院 | Load prediction method based on Kalman filtering and self-adaptive fuzzy neural network |
| CN109978217A (en) * | 2018-11-09 | 2019-07-05 | 国网浙江省电力有限公司绍兴供电公司 | Methods of electric load forecasting based on singular spectrum analysis and memory network |
| CN111967688A (en) * | 2020-09-02 | 2020-11-20 | 沈阳工程学院 | Power load prediction method based on Kalman filter and convolutional neural network |
| CN112803398A (en) * | 2021-01-08 | 2021-05-14 | 武汉数澎科技有限公司 | Load prediction method and system based on empirical mode decomposition and deep neural network |
-
2021
- 2021-09-16 CN CN202111083651.8A patent/CN113537649B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160098052A1 (en) * | 2014-09-12 | 2016-04-07 | Tsinghua University | Partition method and device for power system |
| CN105760952A (en) * | 2016-02-15 | 2016-07-13 | 国网山东省电力公司电力科学研究院 | Load prediction method based on Kalman filtering and self-adaptive fuzzy neural network |
| CN109978217A (en) * | 2018-11-09 | 2019-07-05 | 国网浙江省电力有限公司绍兴供电公司 | Methods of electric load forecasting based on singular spectrum analysis and memory network |
| CN111967688A (en) * | 2020-09-02 | 2020-11-20 | 沈阳工程学院 | Power load prediction method based on Kalman filter and convolutional neural network |
| CN112803398A (en) * | 2021-01-08 | 2021-05-14 | 武汉数澎科技有限公司 | Load prediction method and system based on empirical mode decomposition and deep neural network |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115158399A (en) * | 2022-06-14 | 2022-10-11 | 通号城市轨道交通技术有限公司 | Time sequence signal abnormity detection method and system |
| CN115158399B (en) * | 2022-06-14 | 2023-10-17 | 通号城市轨道交通技术有限公司 | Time sequence signal abnormality detection method and system |
| 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 |
| CN116739383A (en) * | 2023-06-30 | 2023-09-12 | 浙江东鸿电子股份有限公司 | Charging pile power load prediction evaluation method based on big data |
| CN116739383B (en) * | 2023-06-30 | 2024-02-23 | 浙江东鸿电子股份有限公司 | Charging pile power load prediction evaluation method based on big data |
| CN116631429A (en) * | 2023-07-25 | 2023-08-22 | 临沂金诺视讯数码科技有限公司 | Voice and video processing method and system based on VOLTE call |
| CN116631429B (en) * | 2023-07-25 | 2023-10-10 | 临沂金诺视讯数码科技有限公司 | Voice and video processing method and system based on VOLTE call |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113537649B (en) | 2022-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113537649B (en) | Power grid load prediction method and system based on neural network and dynamic mode decomposition | |
| CN109002915B (en) | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model | |
| CN107292446B (en) | Hybrid wind speed prediction method based on component relevance wavelet decomposition | |
| CN111027775A (en) | Step hydropower station generating capacity prediction method based on long-term and short-term memory network | |
| CN111815035B (en) | Fusion form clustering and TCN-attribute short-term load prediction method | |
| CN111191191B (en) | Construction method of combined model for accurately predicting deformation effect of concrete dam | |
| CN102478584B (en) | Wind power station wind speed prediction method based on wavelet analysis and system thereof | |
| CN111582551A (en) | Method and system for predicting short-term wind speed of wind power plant and electronic equipment | |
| CN109002781B (en) | Fault prediction method for energy storage converter | |
| CN115600722A (en) | Long-term prediction method for electric carbon factor of enterprise | |
| CN114118569A (en) | Wind power multi-step prediction method based on multi-mode multi-task Transformer network | |
| CN116151114A (en) | Method and system for predicting service life of fuel cell under meta-universe based on hybrid framework | |
| CN110555566B (en) | B-spline quantile regression-based photoelectric probability density prediction method | |
| CN113837434A (en) | Solar photovoltaic power generation prediction method and device, electronic equipment and storage medium | |
| Moustris et al. | Wind power forecasting using historical data and artificial neural networks modeling | |
| CN117251672A (en) | Regional industry multi-dimensional energy carbon emission calculation method and device based on electric power data | |
| CN117033923A (en) | Method and system for predicting crime quantity based on interpretable machine learning | |
| CN115186935B (en) | Electromechanical device nonlinear fault prediction method and system | |
| CN116167465A (en) | Solar irradiance prediction method based on multivariate time series ensemble learning | |
| CN116911419A (en) | Long time sequence prediction method based on trend correlation feature learning | |
| Ming-guang et al. | Short-term load combined forecasting method based on BPNN and LS-SVM | |
| CN115149528A (en) | Intelligent electric energy meter distributed prediction method based on big data non-intrusive technology | |
| Paulin et al. | SOLAR PHOTOVOLTAIC OUTPUT POWER FORECASTING USING BACK PROPAGATION NEURAL NETWORK. | |
| Kang et al. | Research on forecasting method for effluent ammonia nitrogen concentration based on GRA-TCN | |
| Zhang et al. | Wind farm wind power prediction method based on CEEMDAN and DE optimized DNN neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230306 Address after: Xingang Town, Qidong City, Nantong City, Jiangsu Province 226000 Patentee after: NANTONG HONGDA EXPERIMENT INSTRUMENTS Co.,Ltd. Address before: 226000 north side of Yingbin Road, Juxing Town, Wangbao Town, Qidong City, Nantong City, Jiangsu Province Patentee before: NANTONG HUAHAOJU ELECTRONIC TECHNOLOGY Co.,Ltd. |
|
| TR01 | Transfer of patent right |