CN115130762A - Load ultra-short term prediction method and system based on least square support vector machine - Google Patents

Load ultra-short term prediction method and system based on least square support vector machine Download PDF

Info

Publication number
CN115130762A
CN115130762A CN202210789034.8A CN202210789034A CN115130762A CN 115130762 A CN115130762 A CN 115130762A CN 202210789034 A CN202210789034 A CN 202210789034A CN 115130762 A CN115130762 A CN 115130762A
Authority
CN
China
Prior art keywords
load
ultra
prediction
error
prediction result
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.)
Pending
Application number
CN202210789034.8A
Other languages
Chinese (zh)
Inventor
撖奥洋
牛庆达
周生奇
刘同同
纪永尚
吕宏媛
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.)
QINGDAO POWER SUPPLY Co OF STATE GRID SHANDONG ELECTRIC POWER Co
State Grid Corp of China SGCC
Original Assignee
QINGDAO POWER SUPPLY Co OF STATE GRID SHANDONG ELECTRIC POWER Co
State Grid Corp of China SGCC
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 QINGDAO POWER SUPPLY Co OF STATE GRID SHANDONG ELECTRIC POWER Co, State Grid Corp of China SGCC filed Critical QINGDAO POWER SUPPLY Co OF STATE GRID SHANDONG ELECTRIC POWER Co
Priority to CN202210789034.8A priority Critical patent/CN115130762A/en
Publication of CN115130762A publication Critical patent/CN115130762A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G06F17/12—Simultaneous equations, e.g. systems of linear equations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Economics (AREA)
  • Operations Research (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Algebra (AREA)
  • General Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Water Supply & Treatment (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Computing Systems (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Power Engineering (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The utility model belongs to the technical field of electric power system, concretely relates to load ultra-short term prediction method and system based on least square support vector machine, include: acquiring historical loads of load points; constructing a prediction model by adopting a least square support vector machine; predicting the ultra-short-term load of a load point based on the acquired historical load and the constructed prediction model to obtain a first load ultra-short-term prediction result; calculating the error of the obtained first load ultra-short-term prediction result; predicting the obtained error to obtain an error prediction result; and obtaining a load ultra-short term prediction result based on the obtained first load ultra-short term prediction result and the error prediction result.

Description

Load ultra-short term prediction method and system based on least square support vector machine
Technical Field
The disclosure belongs to the technical field of power systems, and particularly relates to a load ultra-short term prediction method and system based on a least square support vector machine.
Background
The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.
The ultra-short-term load prediction is used for predicting the load change condition from 5min to 1h in the future, and has the remarkable characteristics of high prediction speed and high precision. The traditional method applied to ultra-short-term load prediction at present mainly has a time sequence method, a curve extrapolation method, a load derivation method and the like; the ultra-short-term prediction method based on the modern intelligent technology comprises methods such as an artificial neural network and a support vector machine. The traditional prediction method has the advantages of simple model, high prediction speed and higher prediction precision on a stable load sequence, but is not suitable for load sequence prediction with higher volatility because only the data fitting, namely the searching of a load sequence rule, is taken into consideration; the artificial intelligence prediction method is adaptive to the problems of nonlinearity and uncertainty due to strong adaptive and learning capabilities, but has the defects of long learning time and easy falling into local minimum, so that the model is not converged.
Disclosure of Invention
In order to solve the problems, the invention provides a load ultra-short term prediction method and system based on a least square support vector machine, which comprehensively considers the calculation speed and precision and adopts the least square support vector machine to predict according to the characteristics of high calculation speed of the least square support vector machine and better prediction effect on a load sequence with large volatility compared with the traditional prediction method; starting from the aspect of errors, predicting the errors predicted by the least square support vector machine by utilizing the strong nonlinear mapping capability of the BP neural network, and correcting the predicted values through an error correction link to improve the prediction precision.
According to some embodiments, a first aspect of the present disclosure provides a load ultra-short term prediction method based on a least square support vector machine, which adopts the following technical solutions:
a load ultra-short term prediction method based on a least square support vector machine comprises the following steps:
acquiring historical loads of load points;
constructing a prediction model by adopting a least square support vector machine;
predicting the ultra-short-term load of the load point based on the acquired historical load and the constructed prediction model to obtain a first load ultra-short-term prediction result;
calculating the error of the obtained first load ultra-short-term prediction result;
predicting the obtained error to obtain an error prediction result;
and obtaining a load ultra-short term prediction result based on the obtained first load ultra-short term prediction result and the error prediction result.
As a further technical limitation, in the process of constructing a prediction model, a least square support vector machine regression algorithm based on a fast leave-one method is adopted to cut the acquired historical load sample into a plurality of sample subsets, the sample subsets are trained through a fast leave-one method cross-validation method, and iteration is repeated until all the sample subsets are trained.
As a further technical limitation, the process of predicting the ultra-short term load of the load point is as follows:
inputting a training sample set of historical loads and a testing sample set of historical loads;
performing cross validation by adopting a quick leave-one-out method, and searching for an optimal parameter pair;
judging whether the searched parameters are optimal values, if so, carrying out the next step, and otherwise, returning to the previous step to continue cross validation;
outputting an optimal parameter pair based on a test sample set of historical loads;
and establishing a prediction model, inputting sample data, and predicting to obtain a first load ultra-short-term prediction result.
As a further technical limitation, the obtained first load ultra-short term prediction result is cross-validated by using a FLOO method to obtain an error of the first load ultra-short term prediction result.
And as a further technical limitation, in the process of predicting the obtained error, introducing a prediction error correction link to correct the error of the obtained first load ultra-short term prediction result, and performing rolling prediction by using a BP neural network algorithm to obtain an error prediction result.
Further, the training process of the BP neural network is as follows:
forming a training sample set of the BP neural network based on the error of the obtained first load ultra-short term prediction result;
and constructing a BP neural network, and training the BP neural network based on the formed training sample set of the BP neural network to obtain an error prediction result.
As a further technical limitation, the load very-short term predictor is an accumulated sum of the first load very-short term predictor and the error predictor.
According to some embodiments, a second aspect of the present disclosure provides a load ultra-short term prediction system based on a least square support vector machine, which adopts the following technical solutions:
a load ultra-short term prediction system based on a least square support vector machine comprises:
an acquisition module configured to acquire a historical load of the load point;
a modeling module configured to construct a predictive model using a least squares support vector machine;
the first prediction module is configured to predict the ultra-short-term load of the load point based on the acquired historical load and the constructed prediction model, and obtain a first load ultra-short-term prediction result;
an error calculation module configured to calculate an error of the obtained first load ultra-short term prediction result;
a second prediction module configured to predict the obtained error to obtain an error prediction result;
a prediction module configured to derive a load ultra-short term prediction result based on the derived first load ultra-short term prediction result and the error prediction result.
According to some embodiments, a third aspect of the present disclosure provides a computer-readable storage medium, which adopts the following technical solutions:
a computer readable storage medium, on which a program is stored, which when executed by a processor implements the steps in the method for load ultra-short term prediction based on a least squares support vector machine according to the first aspect of the present disclosure.
According to some embodiments, a fourth aspect of the present disclosure provides an electronic device, which adopts the following technical solutions:
an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in the method for load ultra-short term prediction based on least squares support vector machine according to the first aspect of the present disclosure when executing the program.
Compared with the prior art, the beneficial effect of this disclosure is:
the invention provides a load ultrashort term algorithm based on a least square support vector machine, which introduces a prediction error correction link because the prediction error of the algorithm on the load inflection point is larger, adopts a BP neural network algorithm to predict the prediction error of the least square support vector machine, the prediction result is the sum of the calculation results of two algorithms, and combines an actual example to carry out simulation, thereby verifying that the prediction precision of the algorithm is improved after the correction link is added.
Drawings
The accompanying drawings, which are included to provide a further understanding of the disclosure, illustrate embodiments of the disclosure and together with the description serve to explain the disclosure and are not to limit the disclosure.
FIG. 1 is a flowchart of a load ultra-short term prediction method based on a least squares support vector machine in a first embodiment of the disclosure;
FIG. 2 is a flow chart of a least squares support vector machine in a first embodiment of the present disclosure;
FIG. 3 is a flowchart of BP neural network training in a first embodiment of the present disclosure;
FIG. 4 is a flow chart of ultra-short term load prediction in one embodiment of the present disclosure;
FIG. 5 is a schematic view of a working day load curve according to a first embodiment of the disclosure;
FIG. 6 is a schematic diagram of a weekend load curve according to a first embodiment of the disclosure;
FIG. 7 is a schematic diagram of a 12-month 5-day load ultrashort-term prediction curve according to a first embodiment of the disclosure;
FIG. 8 is a schematic diagram of the 12-month and 5-day overload short-term prediction error in the first embodiment of the disclosure;
FIG. 9 is a schematic diagram of a saturday load ultrashort term prediction curve according to a first embodiment of the disclosure;
FIG. 10 is a schematic diagram of the Saturday load ultrashort term prediction error in accordance with one embodiment of the present disclosure;
FIG. 11 is a schematic diagram of a 12-month and 7-day prediction curve and an actual value curve in the first embodiment of the present disclosure;
FIG. 12 is a schematic diagram of a prediction error of 12 months and 7 days in the first embodiment of the disclosure;
FIG. 13 is a diagram illustrating an error corrected prediction curve according to a first embodiment of the present disclosure;
FIG. 14 is a schematic illustration of a prediction error in a first embodiment of the disclosure;
fig. 15 is a block diagram of a load ultra-short term prediction system based on a least square support vector machine in the second embodiment of the present disclosure.
Detailed Description
The present disclosure is further illustrated by the following examples in conjunction with the accompanying drawings.
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
The embodiments and features of the embodiments in the present disclosure may be combined with each other without conflict.
Example one
The embodiment of the disclosure introduces a load ultra-short term prediction method based on a least square support vector machine.
A load ultra-short term prediction method based on a least square support vector machine as shown in fig. 1 includes:
acquiring historical loads of load points;
constructing a prediction model by adopting a least square support vector machine;
predicting the ultra-short-term load of the load point based on the acquired historical load and the constructed prediction model to obtain a first load ultra-short-term prediction result;
calculating the error of the obtained first load ultra-short term prediction result;
predicting the obtained error to obtain an error prediction result;
and obtaining a load ultra-short term prediction result based on the obtained first load ultra-short term prediction result and the error prediction result.
Next, a detailed description is given of the prediction method in the present embodiment.
A Support Vector Machine (SVM) method is a prediction method established on the statistical theory, the training problem is a classical quadratic programming problem essentially, a local optimal solution can be avoided, a unique global optimal solution is provided, and a plurality of mature algorithms in the optimization theory can be utilized. When the problem of nonlinearity is solved, the SVM can map a nonlinear sample set to a high-dimensional feature space through nonlinear mapping to obtain a linearly separable data set, and a kernel function is used for replacing corresponding inner product operation in the high-dimensional space. The Support Vector Regression (SVR) algorithm is widely applied to load prediction of a power system by introducing an insensitive loss function and a kernel function, and has good prediction performance and popularization capability.
The short-term load prediction based on the SVM has higher prediction precision than the traditional method, is established on the basis of a Vapnik-Chervonenkis Dimension theory and a structural risk minimization principle, has ideal effect on solving the practical problems of small samples, nonlinearity, high dimensional number, local minimum points and the like, has the characteristics of high fitting precision, strong popularization capability, optimal global situation and the like, and fully considers various factors influencing the load. The method has the disadvantages that the selection of self-selection parameters and kernel functions is considered and is generally determined by experience, so that the prediction effect is influenced. Therefore, many improved algorithms are proposed, such as SVM based on linear programming, least squares Support Vector Machine LSSVM, Weighted Support Vector Machine (W-SVM for short), and the like; the performance of the SVM can be improved to a certain extent, wherein LS-SVM is the most common method.
In this embodiment, a least squares support vector machine regression algorithm based on the fast leave-one-out method is adopted.
Given training set
Figure BDA0003733006820000081
Wherein x is i Is called the ith input quantity, y i Is said to correspond to x i Target value of (a) < l >Is the number of samples. The objective of the regression problem is to determine the optimal function f (x) such that the regression function f (x) has the form:
Figure BDA0003733006820000082
in the least square support vector machine, the optimization problem corresponding to the regression problem is
Figure BDA0003733006820000083
It can be seen that the constraint is different compared to the support vector machine, where the constraint is an equation and the support vector machine is an equation, where e i The loss function is represented by a quadratic function, also unlike the support vector machine. The corresponding lagrangian function of the least squares support vector machine is:
Figure BDA0003733006820000091
the optimal value condition of equation (3) is:
Figure BDA0003733006820000092
writing equation (4) to matrix form and eliminating w and e yields the following equation
Figure BDA0003733006820000093
In the formula (I), the compound is shown in the specification,
Figure BDA0003733006820000094
α＝[α 1 ,α 2 ,…α l ] T ；Y＝[y 1 ,y 2 ,…y N ] T (ii) a α and b can be obtained by the formula (5).
Proved and proved that the prediction algorithm is fineThe degree has no relation to the choice of kernel functions, and the common kernel functions are: linear kernel: k (x) i ,x j )＝x T i x j A polynomial kernel: k (x) i ,x j )＝(x T i x j ) d And a Gaussian kernel:
Figure BDA0003733006820000095
this example selects the gaussian radial basis RBF kernel.
The least squares support vector machine regression model is
Figure BDA0003733006820000096
Wherein alpha is i (i ═ 1,2, …, l) and b are each
Figure BDA0003733006820000101
As shown in fig. 2, the algorithm steps of the least squares support vector machine are as follows:
(1) inputting a training sample set M (x) i ,y i ) And test sample set N (x) i ,y i )；
(2) Performing cross validation by adopting a quick leave-one-out method, and searching for optimal parameter pairs gam and sig 2;
(3) judging whether the parameters are optimal, if so, performing the next step, otherwise, returning to the second step to continue the cross validation;
(4) testing the sample, and outputting optimal parameter pairs gam and sig 2;
(5) and establishing a prediction model, inputting sample data, and predicting.
The optimization algorithm of the least square support vector machine adopts the quick leave-one-out cross validation. Cross Validation (CV), also known as cycle estimation, is mainly used in modeling performance evaluation applications; the principle is as follows: in a given number of samples, it is cut into a smaller number of samples of subsets. Then taking out most samples for modeling, leaving a small part of samples for testing the established model, evaluating the reliability of the model according to the error of the testing result, and repeating the above processes until all samples are used for modeling and prediction.
Leave One Out Cross Validation (Leave One Out Cross Validation, LOOCV for short), i.e. Leave One Out; the method is characterized in that only one sample is left for detection and all the remaining samples are used for modeling when verification is performed each time; this process is then repeated until all samples have been used for one test. Therefore, the leave-one-out method is suitable for data detection with a small number of samples. The quick leave-one-out method has small calculation amount, so that the online selection of the model parameters becomes possible. The parameters that the prediction algorithm in this embodiment needs to optimally select are γ, σ.
When the number of samples is l, for the convenience of calculation, the number is recorded
Figure BDA0003733006820000111
P＝H -1 A matrix of
Figure BDA0003733006820000112
Is re-partitioned into
Figure BDA0003733006820000113
Then the corresponding equations (9) and (10) hold
Figure BDA0003733006820000114
Figure BDA0003733006820000115
In the formula (I), the compound is shown in the specification,
Figure BDA0003733006820000116
and
Figure BDA0003733006820000117
respectively represent alpha l And y l Remove the first row, i.e.
Figure BDA0003733006820000118
Figure BDA0003733006820000119
When cross-validation is performed using FLOO, the parameters of the LSSVR model after the ith iteration are changed to
Figure BDA00037330068200001110
And
Figure BDA00037330068200001111
then when the 1 st sample is excluded, then the equation (11) holds
Figure BDA00037330068200001112
Thus, the FLOO-based estimate of the new model parameters for the first sample is
Figure BDA00037330068200001113
The formula (9) and the formula (10) are firstly put into the formula (12) and are arranged to obtain the compound
Figure BDA00037330068200001114
Meanwhile, for the formula (8), the inverse theorem of the block matrix holds that the formula (14)
Figure BDA0003733006820000121
In the formula
Figure BDA0003733006820000122
Thus, cross-validation with FLOOThe prediction error of 1 sample can be expressed as
Figure BDA0003733006820000123
In the formula
Figure BDA0003733006820000124
Represents Q -1 Row 1, column 1 elements. Note that when solving the linear equation set of equation (5), changing the order of the equations does not affect the final solution, so the prediction error for the ith sample based on the FLOO method can be expressed as
Figure BDA0003733006820000125
Q is obtained by adopting an inversion formula of a block matrix for Q of the formula (8) -1 And P l In relation to (2)
Figure BDA0003733006820000126
In the formula
Figure BDA0003733006820000127
Is a scalar quantity. Thus, Q -1 And P l The previous element on the diagonal satisfies the following relation
Figure BDA0003733006820000128
In the formula
Figure BDA0003733006820000129
By substituting formula (18) for formula (16)
Figure BDA00037330068200001210
Thus, the FLOO-based full-sample total error is
Figure BDA00037330068200001211
Due to alpha i And P l,ii All the terms are known, and the calculation amount of s and omicron is small, so
Figure BDA00037330068200001212
Has a computational complexity of about o (l). FLOO-based computing
Figure BDA0003733006820000131
Only one inversion calculation (i.e. P) is performed l ) The calculated amount is 1/l of the common LOO algorithm.
The BP neural network has the advantages of strong autonomous learning ability, no need of establishing accurate mathematical model and physical model before network training, and realization of any complex nonlinear mapping function through training. In this embodiment, a BP neural network rolling prediction method is used to predict the error of the load, and the error of the same time of the current day is predicted by using the error of the same time 4 days before the prediction day as an input according to the similarity and regularity of the predicted error of the load, so that in order to ensure the consistency of the network model, the input training sample at the t time of the ith day is set as X in the selection of the training sample structure i,t ＝(e i-4,t ,e i-3,t ,e i-2,t ,e i-1,t ) In the formula: e.g. of the type i-4,t To predict the error at the same time 4 days before the day, e i-3,t To predict the same time prediction error 3 days before the day, e i-2,t To predict the error at the same time 2 days before the day, e i-1,t The error is predicted at the same time 1 day before the prediction day.
It can be seen that the number of inputs used to train the BP neural network at each time is 4, and then the BP neural network input layer is set to include 4 neurons, where the 4 input layer neurons correspond to the 4 network inputs, i.e., e i-4,t ,e i-3,t ,e i-2,t ,e i-1,t : the output layer comprises a neuron, and 1 output layer neuron corresponds to 1 output e i,t I.e. byAnd (5) error prediction value at the time t on the ith day. As the selection of the number of the hidden layers is not supported by a reliable theory, the training effect is optimal when the number of the hidden layer neurons is 8 through repeated experiments, and finally the BP neural network structure is determined to be 4-8-1 type, namely 4 input neurons, 8 hidden layer neurons and 1 output neuron. The training process of the BP neural network is to find the prediction error of the previous 4 days at the time t as input to obtain the nonlinear mapping relation with the prediction error of the predicted day at the time t as output.
The training process of the BP neural network at each prediction time shown in fig. 3 is as follows:
preparing historical load data L of 96 moments every day for continuous n days i,t Reading historical data and predicting each time of n days by a least square support phasor machine to obtain a corresponding load predicted value L' i,t ；
Obtaining the prediction error e of each time of n days i,t ＝L i,t -L' i,t And storing the prediction error;
forming a trained sample set X of the BP neural network from the stored prediction errors according to a defined training sample structure i,t ；
96 BP neural networks corresponding to 96 moments are respectively established for each moment, and the input is X i,t Output is e i,t Training the BP neural network;
and saving 96 trained BP neural networks to prepare for a load prediction stage.
On the basis of completing the BP network training, the method of the present embodiment may be used to perform ultra-short term load prediction at time t, and the basic flow is shown in fig. 4; the operation flow of ultra-short-term load prediction is as follows:
preparing a prediction sample for a least square support vector machine and an error prediction sample for a BP neural network model at the t moment of the ith day;
inputting a prediction sample at the t moment of the ith day to obtain a load prediction value l at the t moment i,t The predicted value is not corrected by error;
loading the trained BP neural network corresponding to the time t toTaking the error prediction sample as network input to obtain an error prediction value e at the ith day t i,t ；
Load prediction value l obtained by LSSVM prediction i,t Error prediction value e obtained by corresponding BP neural network i,t Adding to obtain final ultra-short-term load predicted value L at the ith day t i,t ＝l i,t +e i,t 。
Adopting three-phase total active power data of a 35kv bus load point in a certain distribution network area to carry out example analysis of the prediction method introduced in the embodiment, obtaining a group of data every 15min, wherein 96 groups of data are totally obtained in one day, the sample data is divided into two parts, one part is a working day of 12.3-12.7 days, the other part is three consecutive weekends of 12.1-12.2, 12.8-12.9 and 12.15-12.6, and active power changes of the load point working day and the weekend are shown in fig. 5 and fig. 6; during the working day, the load change has obvious regularity, and in the time period of about 9 am and 1 to 3 pm of each day, the active power is very large and has a peak value, and at about 11 am and 6 pm of each day, the active power is suddenly reduced and has a valley value, and the two moments are two points with larger fluctuation. During weekends, it can be seen that the load variation for saturday is the same as for weekdays, where sunday loads are generally lower than for weekdays.
The load point load is predicted based on the least square support vector machine algorithm in the embodiment, the prediction period is 15min, and the previous t is 1 ,t 2 ,t 3 ,t 4 ]Predicting the next time t by the data of four time points 5 And sliding the data, with t 5 True value of time and first three t 2 ,t 3 ,t 4 The time is then predicted to be the value of the next time. For the weekdays, 96 data of 12 months and 4 days are adopted as training data, 96 data of 12 months and 5 days are adopted as test data for prediction, and for weekends, 96 load data of 12 months, 1 day and saturday on the previous weekend are adopted as training data, and 96 load data of 12 months, 8 days and saturday are adopted as test data; the prediction curves and prediction errors are shown in fig. 7, 8, 9 and 10, respectively.
As can be seen from fig. 7, 8, 9 and 10, the LSSVM prediction algorithm has a good prediction effect, substantially conforms to the variation trend of the load, and has relatively small prediction errors, the average relative error rate of 12 months and 5 days is 3.514%, the average relative error rate of 12 months and 6 days is 4.413%, and the average relative error rate of 12 months and 7 days is 4.443%. Other predictions do not necessarily show that the relative error rate is substantially maintained between 3% and 5%. For the load prediction of 12 months, 8 days and saturday, the load change of 12 months, 2 days and saturday is quite similar to the load change of 12 months and 8 days, so the prediction effect is good, the average relative error rate is 2.3657%, and the maximum relative error rate is 6.7263%. However, the least squares support vector machine predicts that the prediction error is still large at the inflection point in a time period with large fluctuation, and the relative error rate even reaches 12%, so that the error is corrected by using the BP neural network, and the prediction curves are shown in fig. 11, fig. 12, fig. 13 and fig. 14.
As can be seen from fig. 11, 12, 13, and 14, after error correction is added on day 7 of 12 months, the average relative error rate is reduced from 4.443% to 1.832%, the maximum relative error rate is reduced from 11.815% to 6.731%, the prediction accuracy is significantly improved, the relative error rate is substantially below 4%, the prediction error at almost every moment is reduced, and particularly the prediction accuracy at the inflection point is greatly improved, and the error reduction is significant. However, the relative error rate at the inflection point is still relatively high in value. By combining the above results, it can be known that the ultra-short-term prediction accuracy of the load in the power distribution network is lower than that of the power distribution network, which has a great relationship with the randomness of the load accessed by the power distribution network, the load change regularity in the power distribution network is not as strong as that in a large power distribution network, and especially the load change condition difference is great on weekends, so that the selection of the load data with great similarity to the prediction day as the training sample is crucial to the improvement of the prediction accuracy. The load ultra-short-term prediction is used as a core link of situation prediction, the prediction accuracy directly determines the future situation prediction accuracy of the power distribution network, the current prediction accuracy cannot completely meet the demand of the distribution network situation prediction, and the improvement of the load ultra-short-term prediction accuracy is still the key work of the situation prediction.
In the load ultrashort-term algorithm based on the least square support vector machine, because the prediction error of the algorithm at the load inflection point is large, a prediction error correction link is introduced, the prediction error of the least square support vector machine is predicted by adopting a BP neural network algorithm, the prediction result is the sum of the calculation results of the two algorithms, and the prediction precision of the algorithm is improved after the correction link is added through simulation analysis.
Example two
The second embodiment of the disclosure introduces a load ultra-short term prediction system based on a least square support vector machine.
Fig. 15 shows a load ultra-short term prediction system based on a least squares support vector machine, which includes:
an acquisition module configured to acquire a historical load of the load point;
a modeling module configured to construct a predictive model using a least squares support vector machine;
the first prediction module is configured to predict the ultra-short-term load of the load point based on the acquired historical load and the constructed prediction model, and obtain a first load ultra-short-term prediction result;
an error calculation module configured to calculate an error of the obtained first load ultra-short term prediction result;
the second prediction module is configured to predict the obtained error to obtain an error prediction result;
a prediction module configured to derive a load ultra-short term prediction result based on the derived first load ultra-short term prediction result and the error prediction result.
The detailed steps are the same as those of the load ultra-short term prediction method based on the least square support vector machine provided in the first embodiment, and are not described again here.
EXAMPLE III
The third embodiment of the disclosure provides a computer-readable storage medium.
A computer readable storage medium, on which a program is stored, which when executed by a processor, implements the steps in the method for load ultra-short term prediction based on least squares support vector machine according to one embodiment of the present disclosure.
The detailed steps are the same as those of the load ultra-short term prediction method based on the least square support vector machine provided in the first embodiment, and are not described again here.
Example four
The fourth embodiment of the disclosure provides an electronic device.
An electronic device comprising a memory, a processor and a program stored in the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method for ultra-short term prediction of load based on least squares support vector machine according to an embodiment of the present disclosure.
The detailed steps are the same as the load ultra-short term prediction method based on the least square support vector machine provided in the first embodiment, and are not described herein again.
The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure, and various modifications and changes may be made to the present disclosure by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.

Claims (10)

1. A load ultra-short term prediction method based on a least square support vector machine is characterized by comprising the following steps:
acquiring historical loads of load points;
constructing a prediction model by adopting a least square support vector machine;
predicting the ultra-short-term load of the load point based on the acquired historical load and the constructed prediction model to obtain a first load ultra-short-term prediction result;
calculating the error of the obtained first load ultra-short-term prediction result;
predicting the obtained error to obtain an error prediction result;
and obtaining a load ultra-short term prediction result based on the obtained first load ultra-short term prediction result and the error prediction result.
2. The load ultra-short term prediction method based on the least square support vector machine as claimed in claim 1, characterized in that in the process of constructing the prediction model, the least square support vector machine regression algorithm based on the fast leave-one method is adopted to cut the acquired historical load sample into a plurality of sample subsets, the sample subsets are trained by the fast leave-one method cross-validation method, and the iteration is repeated until all the sample subsets are trained.
3. The method for ultra-short term load prediction based on least squares support vector machine as claimed in claim 1, wherein the process of predicting the ultra-short term load of the load point is:
inputting a training sample set of historical loads and a testing sample set of historical loads;
performing cross validation by adopting a quick leave-one-out method, and searching for an optimal parameter pair;
judging whether the searched parameters are optimal values, if so, carrying out the next step, and otherwise, returning to the previous step to continue cross validation;
outputting an optimal parameter pair based on a test sample set of historical loads;
and establishing a prediction model, inputting sample data, and predicting to obtain a first load ultra-short-term prediction result.
4. The method as claimed in claim 1, wherein the first super short term load prediction result is cross-validated by using a FLOO method to obtain the error of the first super short term load prediction result.
5. The load ultra-short term prediction method based on the least square support vector machine as claimed in claim 1, characterized in that in the process of predicting the obtained error, a prediction error correction link is introduced to correct the error of the obtained first load ultra-short term prediction result, and rolling prediction is performed through a BP neural network algorithm to obtain an error prediction result.
6. The method for ultra-short term load prediction based on least squares support vector machine as claimed in claim 5, wherein the training process of the BP neural network is:
forming a training sample set of the BP neural network based on the error of the obtained first load ultra-short term prediction result;
and constructing a BP neural network, training the BP neural network based on the formed training sample set of the BP neural network, and obtaining an error prediction result.
7. The method as claimed in claim 1, wherein the super short term prediction result is a cumulative sum of the first super short term prediction result and the error prediction result.
8. A load ultra-short term prediction system based on a least square support vector machine is characterized by comprising:
an acquisition module configured to acquire a historical load of the load point;
a modeling module configured to construct a predictive model using a least squares support vector machine;
the first prediction module is configured to predict the ultra-short-term load of the load point based on the acquired historical load and the constructed prediction model, and obtain a first load ultra-short-term prediction result;
an error calculation module configured to calculate an error of the obtained first load ultra-short term prediction result;
a second prediction module configured to predict the obtained error to obtain an error prediction result;
a prediction module configured to derive a load ultra-short term prediction result based on the derived first load ultra-short term prediction result and the error prediction result.
9. A computer-readable storage medium, on which a program is stored, which, when being executed by a processor, carries out the steps of the method for ultra-short term prediction of a load based on a least squares support vector machine according to any one of claims 1 to 7.
10. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps in the method for load ultra-short term prediction based on a least squares support vector machine as claimed in any one of claims 1-7.
CN202210789034.8A 2022-07-06 2022-07-06 Load ultra-short term prediction method and system based on least square support vector machine Pending CN115130762A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210789034.8A CN115130762A (en) 2022-07-06 2022-07-06 Load ultra-short term prediction method and system based on least square support vector machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210789034.8A CN115130762A (en) 2022-07-06 2022-07-06 Load ultra-short term prediction method and system based on least square support vector machine

Publications (1)

Publication Number Publication Date
CN115130762A true CN115130762A (en) 2022-09-30

Family

ID=83382791

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210789034.8A Pending CN115130762A (en) 2022-07-06 2022-07-06 Load ultra-short term prediction method and system based on least square support vector machine

Country Status (1)

Country Link
CN (1) CN115130762A (en)

Similar Documents

Publication Publication Date Title
CN106600059B (en) Intelligent power grid short-term load prediction method based on improved RBF neural network
CN110647042B (en) Robot robust learning prediction control method based on data driving
CN113361197B (en) Method and system for predicting remaining service life of lithium battery
CN113156263B (en) Cable state monitoring method and storage medium
CN110689183B (en) Cluster photovoltaic power probability prediction method, system, medium and electronic device
CN112434848A (en) Nonlinear weighted combination wind power prediction method based on deep belief network
CN111723982A (en) Medium-and-long-term power load combined prediction method based on gray-Markov chain
CN105710137B (en) A kind of cold-rolling mill multivariable board-shape control method based on constrained optimization method
CN116484747A (en) Sewage intelligent monitoring method based on self-adaptive optimization algorithm and deep learning
CN113406503A (en) Lithium battery SOH online estimation method based on deep neural network
CN115201686A (en) Lithium ion battery health state assessment method under incomplete charging and discharging data
CN112036651A (en) Electricity price prediction method based on quantum immune optimization BP neural network algorithm
CN115308558A (en) Method and device for predicting service life of CMOS (complementary Metal oxide semiconductor) device, electronic equipment and medium
CN112036598A (en) Charging pile use information prediction method based on multi-information coupling
CN114564787A (en) Bayesian optimization method, device and storage medium for target-related airfoil design
CN115062528A (en) Prediction method for industrial process time sequence data
Hatherall et al. Load prediction based remaining discharge energy estimation using a combined online and offline prediction framework
CN115130762A (en) Load ultra-short term prediction method and system based on least square support vector machine
CN105976057A (en) Medium and long-term load forecasting method based on improved grey theory
CN116029183A (en) Power battery temperature prediction method based on iPSO-LSTM model
CN112581311B (en) Method and system for predicting long-term output fluctuation characteristics of aggregated multiple wind power plants
CN115101136A (en) Large-scale aluminum electrolysis cell global anode effect prediction method
CN114895206A (en) Lithium ion battery SOH estimation method based on RBF neural network of improved wolf optimization algorithm
CN112329995A (en) Optimal scheduling method and device for distributed energy storage cluster and computer equipment
CN113570129A (en) Method for predicting strip steel pickling concentration and computer readable storage medium

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