CN112766076A - Power load ultra-short-term prediction method, system, equipment and storage medium - Google Patents

Power load ultra-short-term prediction method, system, equipment and storage medium Download PDF

Info

Publication number
CN112766076A
CN112766076A CN202011638086.2A CN202011638086A CN112766076A CN 112766076 A CN112766076 A CN 112766076A CN 202011638086 A CN202011638086 A CN 202011638086A CN 112766076 A CN112766076 A CN 112766076A
Authority
CN
China
Prior art keywords
load
principal component
components
remainder
kelm
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
CN202011638086.2A
Other languages
Chinese (zh)
Other versions
CN112766076B (en
Inventor
丁云飞
叶子军
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.)
Shanghai Dianji University
Original Assignee
Shanghai Dianji 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 Shanghai Dianji University filed Critical Shanghai Dianji University
Priority to CN202011638086.2A priority Critical patent/CN112766076B/en
Publication of CN112766076A publication Critical patent/CN112766076A/en
Application granted granted Critical
Publication of CN112766076B publication Critical patent/CN112766076B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • 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 a method, a system, equipment and a storage medium for ultra-short-term prediction of a power load, wherein the method specifically comprises the following steps: collecting a power load time sequence signal S (t), carrying out MEEMD decomposition on the signal S (t), and removing abnormal signals in the signal S (t) to obtain S' (t); performing EMD on S' (t) to obtain a remainder and a plurality of first IMF components, calculating principal component contribution rates corresponding to the remainder and the plurality of first IMF components, and substituting the remainder, the plurality of first IMF components and the corresponding principal component contribution rates into the trained extreme learning machine to obtain a plurality of predicted load components; and superposing the plurality of predicted load components to obtain the predicted total load of S (t). Compared with the prior art, the invention has the advantages of high speed, high precision, high reliability and the like.

Description

Power load ultra-short-term prediction method, system, equipment and storage medium
Technical Field
The invention relates to the field of power system planning, in particular to a power load ultra-short-term prediction method, a system, equipment and a storage medium.
Background
The prediction of the power load requires that the deviation between the predicted value and the actual load is smaller than the AGC (automatic Generation control) regulation capacity of the power grid. Because the current power grid AGC adjusting capacity is generally small and provides higher requirements for the precision of a power load prediction model, establishing a high-performance real-time power load prediction model is always a hot point of research in academia and industry.
Load prediction methods such as a neural network, a support vector machine, an extreme learning machine and the like are widely applied to power load prediction, load time sequence signals have typical non-stationarity, and when a single prediction method is adopted to process a large amount of data, the running time is long and large errors are easy to generate.
Some solutions are also provided in the prior art, and a Chinese patent CN201911128175.X provides an electric load prediction method and a system, which comprises decomposing an original load sequence by using a lumped empirical mode decomposition algorithm; calculating approximate entropy of each modal component and combining to obtain a reconstructed new sequence; each new subsequence is predicted by a load prediction model of the extreme learning machine; and superposing the prediction results of each subsequence to obtain a final prediction value. The prediction analysis of the actual power grid load data by using the method provided by the invention shows that the method decomposes the original load sequence by using a lumped empirical mode decomposition method, thereby overcoming the mode aliasing problem caused by discontinuity of signals of the original load sequence.
However, in the scheme, the original load sequence is decomposed by adopting a lumped empirical mode decomposition algorithm, but the method has white noise residue, cannot truly restore the electric power load time sequence signal, and has low prediction accuracy.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a method, a system, equipment and a storage medium for predicting the power load in an ultra-short period, which have the advantages of high speed, high precision and high reliability.
The purpose of the invention can be realized by the following technical scheme
A power load ultra-short term prediction method specifically comprises the following steps:
s1: collecting a power load time sequence signal S (t), carrying out MEEMD decomposition on the signal S (t), and removing abnormal signals in the signal S (t) to obtain S' (t);
s2: EMD decomposition is carried out on S' (t) to obtain a remainder and a plurality of first IMF components;
s3: calculating principal component contribution rates corresponding to the remainder and the plurality of first IMF components;
s4: substituting the remainder, a plurality of first IMF components and corresponding principal component contribution rates into the trained KELM model to obtain a plurality of predicted load components;
s5: superposing a plurality of predicted load components to obtain the predicted total load of S (t);
the KELM model introduces a kernel function into the ELM, can map input space sample data to a high-dimensional feature space, solves the problem of data prediction nonlinearity, can improve the convergence speed and accuracy of the model, simultaneously introduces principal component contribution rate into the kernel function, becomes a principal component fusion-based kernel extreme learning machine, can add one principal component contribution rate calculation to each feature parameter in the traditional extreme learning machine, and can eliminate the defect that parameters with different correlation degrees have the same influence on results.
Further, the MEEMD combines CEEMD with PE-based signal randomness detection, and is an improved ensemble empirical mode decomposition method, after an abnormal component of the MEEMD decomposition is detected, the ememd decomposition is directly performed, so that a mode aliasing phenomenon can be effectively suppressed, a reconstruction error is reduced, a calculation amount is reduced, and the completeness is good, and the specific process of the step S1 is as follows:
s101: adding a group of white noise signals S with zero mean value into the power load timing signal S (t)+(t) and S-(t)；
S102: will S+(t) and S-(t) EMD decomposition to obtain a second IMF component I+(t) and I-(t)；
S103: will I+(t) and I-(t) performing ensemble averaging to obtain an ensemble average value IP(t) judgment of IP(t) if the average value is larger than the set threshold value, if so, repeating the step S101, otherwise, eliminating the obtained P-1 integrated average values in the step S (t) to obtain S' (t).
Further, the principal component contribution rate is a variance contribution rate;
step S3 specifically includes:
s301: taking the remainder and a plurality of first IMF components as samples to form a feature matrix;
s302: transforming the feature matrix into a correlation matrix:
s303: solving eigenvalues corresponding to a plurality of principal components according to the correlation matrix;
s304: and solving the variance contribution rate of each principal component according to the characteristic value.
Further, the extreme learning machine is a KELM model, and the training process specifically includes:
setting initialization parameters, wherein the parameters comprise iteration times, stop conditions, kernel parameters gamma and connection weights beta of the hidden layer and the output layer;
carrying out iterative solution on the KELM by adopting a training set until beta is not changed any more;
wherein, the kernel function K (x) for solving the KELMi,xj) The calculation formula of (2) is as follows:
Figure RE-GDA0002982597800000031
therein, Dis* p(x, y) is the Minkowski distance, calculated as:
Figure RE-GDA0002982597800000032
wherein (x)j，yj) For the KELM input, Ci is the variance contribution of the principal component, d is the dimension, and p is the norm.
A power load ultra-short-term prediction system comprises a signal acquisition module, a signal filtering module and a load prediction module:
the signal acquisition module is used for acquiring a power load time sequence signal S (t);
the signal filtering module is used for performing MEEMD decomposition on the S (t), eliminating abnormal signals in the S (t) and obtaining S' (t);
the load prediction module comprises a load component prediction unit and a load total prediction unit;
the load component prediction unit performs EMD decomposition on S' (t) to obtain a remainder and a plurality of first IMF components, calculates principal component contribution rates corresponding to the remainder and the plurality of first IMF components, substitutes the remainder, the plurality of first IMF components and the corresponding principal component contribution rates into the trained KELM model, and correspondingly obtains a plurality of predicted load components;
the total load forecasting unit superposes a plurality of forecasting load components to obtain a forecasting total load of S (t);
the KELM model introduces a kernel function into the ELM, can map input space sample data to a high-dimensional feature space, solves the problem of data prediction nonlinearity, can improve the convergence speed and accuracy of the model, simultaneously introduces principal component contribution rate into the kernel function, becomes a principal component fusion-based kernel extreme learning machine, can add one principal component contribution rate calculation to each feature parameter in the traditional extreme learning machine, and can eliminate the defect that parameters with different correlation degrees have the same influence on results.
Furthermore, the MEEMD combines CEEMD and PE-based signal randomness detection, is an improved set empirical mode decomposition method, and is directly subjected to EMD decomposition after an abnormal component of CEEMD decomposition is detected, so that modal confusion can be effectively inhibited, reconstruction errors are reduced, calculation amount is reduced, and completeness is good;
the specific process of the signal filtering module for obtaining S' (t) is as follows:
s601: the signal filtering module adds a group of white noise signals S with zero mean value in the power load time sequence signals S (t)+(t) and S-(t)；
S602: the signal filtering module is used for filtering S+(t) and S-(t) EMD decomposition to obtain a second IMF component I+(t) and I-(t)；
S603: will I+(t) and I-(t) performing ensemble averaging to obtain an ensemble average value IP(t) judgment of IP(t) if it is greater than the set threshold valueIf so, repeating the step S101, otherwise, eliminating the obtained P-1 integrated mean values in the step S (t) to obtain S' (t).
Further, the principal component contribution rate is a variance contribution rate;
the calculation process of the variance contribution rate specifically comprises the following steps:
s701: taking the remainder and a plurality of first IMF components as samples to form a feature matrix;
s702: transforming the feature matrix into a correlation matrix:
s703: solving eigenvalues corresponding to a plurality of principal components according to the correlation matrix;
s704: and solving the variance contribution rate of each principal component according to the characteristic value.
Furthermore, the extreme learning machine is a KELM model, the system further comprises a model training module, and the training module is used for training the KELM model;
the training process specifically comprises the following steps:
setting initialization parameters, wherein the parameters comprise iteration times, stop conditions, kernel parameters gamma and connection weights beta of the hidden layer and the output layer;
carrying out iterative solution on the KELM by adopting a training set until beta is not changed any more;
wherein, the kernel function K (x) for solving the KELMi,xj) The calculation formula of (2) is as follows:
Figure RE-GDA0002982597800000041
therein, Dis* p(x, y) is the Minkowski distance, calculated as:
Figure RE-GDA0002982597800000042
wherein (x)j，yj) For the input of KELM, Ci is the variance contribution of the principal component, d is the dimension, and p is
An ultra-short term prediction device for electrical loads, comprising a memory and a processor, wherein the memory stores a computer program, and the processor calls the program instructions to execute the prediction method.
A computer readable storage medium comprising a computer program executable by a processor to implement any of the prediction methods.
Compared with the prior art, the invention has the following beneficial effects:
(1) according to the method, the power load time sequence signal is subjected to MEEMD decomposition, the abnormal second IMF component with white noise is removed according to the arrangement entropy, the influence of the white noise on reconstruction errors is weakened, the power load time sequence signal with the abnormal signal is removed has better completeness and high reliability, the calculated amount is reduced, the prediction precision and speed are improved, the first IMF component of the power load time sequence signal with the abnormal signal removed is stable and does not have sharp waves and outliers, the fitting degree is better, and the prediction errors are reduced;
(2) according to the method, the principal component contribution rate is added to the characteristic parameters in the prediction process of the extreme learning machine, the defect that the parameters with different correlation degrees have the same influence on the result can be eliminated, and the prediction accuracy and the convergence speed of the model are improved;
(3) the extreme learning machine is a KELM model, a kernel function is introduced into the ELM by the KELM model, a kernel algorithm is added, input space sample data can be mapped to a high-dimensional characteristic space, the problem of data prediction nonlinearity is solved, and the convergence speed and the prediction accuracy of the model are improved.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
Example 1
An ultra-short term prediction method for a power load, as shown in fig. 1, specifically includes:
101) collecting power load time sequence signal S (t), adding a group of white noise signals S (t) with zero mean value into the power load time sequence signal S (t)+(t) and S-(t)；
102) Will S+(t) and S-(t) EMD decomposition to obtain a second IMF component I+(t) and I-(t)；
103) Will I+(t) and I-(t) performing ensemble averaging to obtain an ensemble average value IP(t)；
104) Judgment of IP(t) whether the value is larger than a set threshold value, if so, executing a step 101), otherwise, executing a step 105);
105) eliminating P-1 integration average values obtained in S (t) to obtain S' (t);
106) EMD decomposition is carried out on S' (t) to obtain a remainder and a plurality of first IMF components;
107) calculating principal component contribution rates corresponding to the remainder and the plurality of first IMF components;
108) substituting the remainder, a plurality of first IMF components and corresponding principal component contribution rates into the trained KELM model to obtain a plurality of predicted load components;
109) and superposing the plurality of predicted load components to obtain the predicted total load of S (t).
The MEEMD combines CEEMD and PE-based signal randomness detection, is an improved ensemble empirical mode decomposition method, directly performs EMD decomposition after detecting abnormal components of CEEMD decomposition, can effectively inhibit mode confusion, reduces reconstruction errors, reduces calculation amount and has better completeness;
the KELM model introduces a kernel function into the ELM, can map input space sample data to a high-dimensional feature space, solves the problem of data prediction nonlinearity, can improve the convergence speed and accuracy of the model, simultaneously introduces principal component contribution rate into the kernel function to form a principal component fusion-based kernel extreme learning machine, can add one principal component contribution rate calculation to each feature parameter in the conventional extreme learning machine, and can eliminate the defect that parameters with different correlation degrees have the same influence on results.
The principal component contribution rate is the variance contribution rate, and the calculation process of the principal component contribution rate specifically comprises the following steps:
s301: taking the remainder and a plurality of first IMF components as samples to form a feature matrix;
s302: transforming the feature matrix into a correlation matrix:
s303: solving eigenvalues corresponding to a plurality of principal components according to the correlation matrix;
s304: and solving the variance contribution rate of each principal component according to the characteristic value.
The training process of the KELM model specifically comprises the following steps:
setting initialization parameters, wherein the parameters comprise iteration times, stop conditions, kernel parameters gamma and connection weights beta of the hidden layer and the output layer;
calculating Minkowski distance Dis* p(x, y) and the calculation formula is as follows:
Figure BDA0002879135060000061
wherein (x)j，yj) Ci is the variance contribution rate of the principal component, d is the dimension, and p is the norm;
carrying out iterative solution on the KELM by adopting a training set until beta is not changed any more;
wherein, a kernel function K (x) for solving KELMi,xj) The calculation formula of (2) is as follows:
Figure BDA0002879135060000071
example 2
A power load ultra-short-term prediction system comprises a signal acquisition module, a signal filtering module, a load prediction module and a model training module:
the signal acquisition module is used for acquiring a power load time sequence signal S (t);
the signal filtering module is used for performing MEEMD decomposition on the S (t), eliminating abnormal signals in the S (t) and obtaining S' (t);
the model training module is used for training a KELM model;
the load prediction module comprises a load component prediction unit and a load total prediction unit;
the load component prediction unit performs EMD decomposition on S' (t) to obtain a remainder and a plurality of first IMF components, calculates principal component contribution rates corresponding to the remainder and the plurality of first IMF components, substitutes the remainder, the plurality of first IMF components and the corresponding principal component contribution rates into the trained KELM model, and correspondingly obtains a plurality of predicted load components;
and the total load predicting unit superposes a plurality of predicted load components to obtain the predicted total load of S (t).
The KELM model introduces a kernel function into the ELM, can map input space sample data to a high-dimensional feature space, solves the problem of data prediction nonlinearity, can improve the convergence speed and accuracy of the model, simultaneously introduces principal component contribution rate into the kernel function to form a principal component fusion-based kernel extreme learning machine, can add one principal component contribution rate calculation to each feature parameter in the conventional extreme learning machine, and can eliminate the defect that parameters with different correlation degrees have the same influence on results.
The MEEMD combines CEEMD and PE-based signal randomness detection, is an improved ensemble empirical mode decomposition method, directly performs EMD decomposition after detecting abnormal components of CEEMD decomposition, can effectively inhibit mode confusion, reduces reconstruction errors, reduces calculation amount and has better completeness;
the specific process of the signal filtering module for obtaining S' (t) is as follows:
s601: the signal filtering module adds a group of white noise signals S with zero mean value in the power load timing signals S (t)+(t) and S-(t)；
S602: the signal filtering module will S+(t) and S-(t) EMD decomposition to obtain the secondIMF fraction I+(t) and I-(t)；
S603: will I+(t) and I-(t) performing ensemble averaging to obtain an ensemble average value IP(t) judgment of IP(t) if the average value is larger than the set threshold value, if so, repeating the step S101, otherwise, eliminating the obtained P-1 integrated average values in the step S (t) to obtain S' (t).
The principal component contribution rate is the variance contribution rate;
the calculation process of the variance contribution rate specifically comprises the following steps:
s701: taking the remainder and a plurality of first IMF components as samples to form a feature matrix;
s702: transforming the feature matrix into a correlation matrix:
s703: solving eigenvalues corresponding to a plurality of principal components according to the correlation matrix;
s704: and solving the variance contribution rate of each principal component according to the characteristic value.
The training process of the KELM model specifically comprises the following steps:
setting initialization parameters, wherein the parameters comprise iteration times, stop conditions, kernel parameters gamma and connection weights beta of the hidden layer and the output layer;
carrying out iterative solution on the KELM by adopting a training set until beta is not changed any more;
wherein, a kernel function K (x) for solving KELMi,xj) The calculation formula of (2) is as follows:
Figure RE-GDA0002982597800000081
therein, Dis* p(x, y) is the Minkowski distance, calculated as:
Figure RE-GDA0002982597800000082
wherein (x)j，yj) For the KELM input, Ci is the variance contribution of the principal component, d is the dimension, and p is the norm.
Example 3
An ultra-short term prediction device of power load comprises a memory and a processor, wherein the memory stores a computer program, and the processor calls the program instruction to execute the prediction method in the embodiment 1.
Example 4
A computer-readable storage medium comprising a computer program executable by a processor to implement the prediction method of embodiment 1.
Embodiment 1, embodiment 2, embodiment 3, and embodiment 4 provide a method, a system, a device, and a storage medium for ultra-short term prediction of power load, which perform MEEMD decomposition on a power load timing signal, remove an abnormal second IMF component added with white noise according to an arrangement entropy, weaken the influence of the addition of white noise on a reconstruction error, so that the power load timing signal from which an abnormal signal is removed has better completeness, reduce a calculation amount, improve prediction accuracy and speed, stabilize the first IMF component of the power load timing signal from which the abnormal signal is removed without spike and outlier, have better fitting degree, reduce a prediction error, and add a principal component contribution rate to characteristic parameters in a prediction process of a limit learning machine, can eliminate the defect that different parameters of correlation have the same influence on a result, the prediction accuracy and the convergence speed of the model are improved, the extreme learning machine is a KELM model, the KELM model introduces a kernel function into the ELM, a kernel algorithm is added, input space sample data can be mapped to a high-dimensional feature space, the problem of data prediction nonlinearity is solved, and the convergence speed and the prediction accuracy of the model are improved.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims (10)

1. A power load ultra-short term prediction method is characterized by comprising the following steps:
s1: collecting a power load time sequence signal S (t), carrying out MEEMD decomposition on the signal S (t), and removing abnormal signals in the signal S (t) to obtain S' (t);
s2: EMD decomposition is carried out on S' (t) to obtain a remainder and a plurality of first IMF components;
s3: calculating principal component contribution rates corresponding to the remainder and the plurality of first IMF components;
s4: substituting the remainder, a plurality of first IMF components and corresponding principal component contribution rates into the trained extreme learning machine to obtain a plurality of predicted load components;
s5: and superposing the plurality of predicted load components to obtain the predicted total load of S (t).
2. The ultra-short term prediction method for electrical load according to claim 1, wherein the specific process of step S1 is as follows:
s101: adding a group of white noise signals S with zero mean value into the power load timing signal S (t)+(t) and S-(t)；
S102: will S+(t) and S-(t) EMD decomposition to obtain a second IMF component I+(t) and I-(t)；
S103: will I+(t) and I-(t) performing ensemble averaging to obtain an ensemble average value IP(t) judgment of IP(t) whether the average value is larger than the set threshold value or not, if so, repeating the step S101, otherwise, eliminating the obtained P-1 integrated average values in the step S (t) to obtain S' (t).
3. The ultra-short term power load prediction method as claimed in claim 1, wherein the principal component contribution rate is a variance contribution rate;
step S3 specifically includes:
s301: taking the remainder and a plurality of first IMF components as samples to form a feature matrix;
s302: transforming the feature matrix into a correlation matrix:
s303: solving eigenvalues corresponding to a plurality of principal components according to the correlation matrix;
s304: and solving the variance contribution rate of each principal component according to the characteristic value.
4. The ultra-short term prediction method for electrical load as claimed in claim 1, wherein the extreme learning machine is a KELM model, and the training process specifically comprises:
setting initialization parameters, wherein the parameters comprise iteration times, stop conditions, kernel parameters gamma and connection weights beta of the hidden layer and the output layer;
carrying out iterative solution on the KELM by adopting a training set until beta is not changed any more;
wherein, the kernel function K (x) for solving the KELMi,xj) The calculation formula of (2) is as follows:
Figure RE-FDA0002982597790000021
therein, Dis* p(x, y) is the Minkowski distance, calculated as:
Figure RE-FDA0002982597790000022
wherein (x)j，yj) For the KELM input, Ci is the variance contribution of the principal component, d is the dimension, and p is the norm.
5. An ultra-short term prediction system for electrical loads, comprising:
the signal acquisition module is used for acquiring a power load time sequence signal S (t);
the signal filtering module is used for performing MEEMD decomposition on the S (t), eliminating abnormal signals in the S (t) and obtaining S' (t);
the load prediction module comprises a load component prediction unit and a load total amount prediction unit;
the load component prediction unit performs EMD decomposition on S' (t) to obtain a remainder and a plurality of first IMF components, calculates principal component contribution rates corresponding to the remainder and the plurality of first IMF components, and substitutes the remainder, the plurality of first IMF components and the corresponding principal component contribution rates into the trained extreme learning machine to correspondingly obtain a plurality of predicted load components;
and the total load predicting unit superposes a plurality of predicted load components to obtain the predicted total load of S (t).
6. The ultra-short term prediction system for electrical load as claimed in claim 5, wherein the specific process of obtaining S' (t) by the signal filtering module is:
s601: the signal filtering module adds a group of white noise signals S with zero mean value in the power load time sequence signals S (t)+(t) and S-(t)；
S602: the signal filtering module is used for filtering S+(t) and S-(t) EMD decomposition to obtain a second IMF component I+(t) and I-(t)；
S603: will I+(t) and I-(t) performing ensemble averaging to obtain an ensemble average value IP(t) judgment of IP(t) whether the average value is larger than the set threshold value or not, if so, repeating the step S101, otherwise, eliminating the obtained P-1 integrated average values in the step S (t) to obtain S' (t).
7. The ultra-short term prediction system for electrical load as claimed in claim 5, wherein the principal component contribution rate is variance contribution rate;
the calculation process of the variance contribution rate specifically comprises the following steps:
s701: taking the remainder and a plurality of first IMF components as samples to form a feature matrix;
s702: transforming the feature matrix into a correlation matrix:
s703: solving eigenvalues corresponding to a plurality of principal components according to the correlation matrix;
s704: and solving the variance contribution rate of each principal component according to the characteristic value.
8. The ultra-short term prediction system for electrical load as claimed in claim 5, wherein the extreme learning machine is a KELM model, the system further comprises a model training module, the training module is used for training the KELM model;
the training process specifically comprises the following steps:
setting initialization parameters, wherein the parameters comprise iteration times, stop conditions, kernel parameters gamma and connection weights beta of the hidden layer and the output layer;
carrying out iterative solution on the KELM by adopting a training set until beta is not changed any more;
wherein, the kernel function K (x) for solving the KELMi,xj) The calculation formula of (2) is as follows:
Figure RE-FDA0002982597790000031
therein, Dis* p(x, y) is the Minkowski distance, calculated as:
Figure RE-FDA0002982597790000032
wherein (x)j，yj) For the KELM input, Ci is the variance contribution of the principal component, d is the dimension, and p is the norm.
9. An ultra-short term prediction device for electrical loads, comprising a memory and a processor, wherein the memory stores a computer program, and the processor calls the program instructions to execute the prediction method according to any one of claims 1 to 4.
10. A computer-readable storage medium comprising a computer program executable by a processor to implement the prediction method of any one of claims 1 to 4.
CN202011638086.2A 2020-12-31 2020-12-31 Ultra-short-term prediction method, system, equipment and storage medium for power load Active CN112766076B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011638086.2A CN112766076B (en) 2020-12-31 2020-12-31 Ultra-short-term prediction method, system, equipment and storage medium for power load

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011638086.2A CN112766076B (en) 2020-12-31 2020-12-31 Ultra-short-term prediction method, system, equipment and storage medium for power load

Publications (2)

Publication Number Publication Date
CN112766076A true CN112766076A (en) 2021-05-07
CN112766076B CN112766076B (en) 2023-05-12

Family

ID=75698146

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011638086.2A Active CN112766076B (en) 2020-12-31 2020-12-31 Ultra-short-term prediction method, system, equipment and storage medium for power load

Country Status (1)

Country Link
CN (1) CN112766076B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106600059A (en) * 2016-12-13 2017-04-26 北京邮电大学 Intelligent power grid short-term load predication method based on improved RBF neural network
CN109034490A (en) * 2018-08-13 2018-12-18 广东工业大学 A kind of Methods of electric load forecasting, device, equipment and storage medium
CN109146183A (en) * 2018-08-24 2019-01-04 广东工业大学 Short-term impact load forecasting model method for building up based on signal decomposition and intelligent optimization algorithm

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106600059A (en) * 2016-12-13 2017-04-26 北京邮电大学 Intelligent power grid short-term load predication method based on improved RBF neural network
CN109034490A (en) * 2018-08-13 2018-12-18 广东工业大学 A kind of Methods of electric load forecasting, device, equipment and storage medium
CN109146183A (en) * 2018-08-24 2019-01-04 广东工业大学 Short-term impact load forecasting model method for building up based on signal decomposition and intelligent optimization algorithm

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
NILS JAKOB JOHANNESEN ET AL.: "Deregulated Electric Energy Price Forecasting in NordPool Market using Regression Techniques", 《2019ISPEC》 *
赵睿智等: "基于 MEEMD-KELM 的短期风电功率预测", 《电测与仪表》 *

Also Published As

Publication number Publication date
CN112766076B (en) 2023-05-12

Similar Documents

Publication Publication Date Title
Yang et al. Remaining useful life prediction based on denoising technique and deep neural network for lithium-ion capacitors
CN110826803A (en) Electricity price prediction method and device for electric power spot market
CN111426905B (en) Power distribution network common bus transformation relation abnormity diagnosis method, device and system
CN107844670A (en) The computational methods of sample size needed for a kind of harmonic wave statistics
CN113627685B (en) Wind driven generator power prediction method considering wind power internet load limit
CN106650060B (en) Photovoltaic cell internal resistance attenuation coefficient prediction method
CN112766076A (en) Power load ultra-short-term prediction method, system, equipment and storage medium
CN112906967A (en) Desulfurization system slurry circulating pump performance prediction method and device
Sun et al. Short-term power load prediction based on VMD-SG-LSTM
Lopez-Ramirez et al. Fpga-based online voltage/current swell segmentation and measurement
Liu et al. A key-term separation based least square method for Hammerstein SOC estimation model
CN115587326B (en) Noise environment wind power plant data correction method
Uddin et al. An intelligent short-circuit fault classification scheme for power transmission line
CN111505420B (en) Online monitoring and diagnosing method and system for state of line arrester
Yue et al. Signal acquisition analyzer based on MSP432
CN113341278B (en) System and method for evaluating insulation performance of gas insulation voltage transformer
CN117313021B (en) Power equipment abnormality detection analysis method, system, terminal and medium
Yin et al. Generalized accelerated failure time frailty model for systems subject to imperfect preventive maintenance
Wang et al. An Automatic Identification Framework for Complex Power Quality Disturbances Based on Ensemble CNN
EP4116892A1 (en) Learning processing program, information processing device, and learning processing method
CN111965424A (en) Novel prediction compensation method for wide area signal of power system
CN116044798A (en) Fault diagnosis method and device for photovoltaic inverter fan and electronic equipment
CN117992851A (en) Power grid power quality disturbance classification method and device, electronic equipment and storage medium
CN117972312A (en) Power transformation data processing method and device
CN116911155A (en) Concentration prediction method, device, equipment and storage medium for online monitoring of dissolved gas in transformer oil

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