CN113487068A - Short-term wind power prediction method based on long-term and short-term memory module - Google Patents

Short-term wind power prediction method based on long-term and short-term memory module Download PDF

Info

Publication number
CN113487068A
CN113487068A CN202110685441.XA CN202110685441A CN113487068A CN 113487068 A CN113487068 A CN 113487068A CN 202110685441 A CN202110685441 A CN 202110685441A CN 113487068 A CN113487068 A CN 113487068A
Authority
CN
China
Prior art keywords
wind power
short
term
training
long
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
CN202110685441.XA
Other languages
Chinese (zh)
Other versions
CN113487068B (en
Inventor
黄文聪
潘风
杨子潇
朱自铭
张凤顺
王浩源
朱禛浩
张惠雯
胡宇博
常雨芳
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.)
Hubei University of Technology
Original Assignee
Hubei University of Technology
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 Hubei University of Technology filed Critical Hubei University of Technology
Priority to CN202110685441.XA priority Critical patent/CN113487068B/en
Publication of CN113487068A publication Critical patent/CN113487068A/en
Application granted granted Critical
Publication of CN113487068B publication Critical patent/CN113487068B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a short-term wind power prediction method based on a long-term and short-term memory module, which comprises the following steps: acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data; carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components; inputting a plurality of connotation modal components into the long-short term memory model to obtain an initial prediction result; determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence; a target prediction result is determined based on the corrected error sequence and the initial prediction result. According to the method, the wind power preprocessing data are decomposed by adopting improved empirical mode decomposition, reconstruction errors caused by the addition of white noise can be reduced, the problem of mode splitting can be suppressed, and an autoregressive moving average model is adopted to correct an error sequence, so that a target prediction result is closer to an actual value, and a more accurate prediction result is obtained.

Description

Short-term wind power prediction method based on long-term and short-term memory module
Technical Field
The application relates to the field of prediction, in particular to a short-term wind power prediction method based on a long-term and short-term memory module.
Background
With the rapid development of economy in China, the traditional energy sources of coal, petroleum and natural gas are rapidly consumed, and only in a not too long period, the traditional energy sources of coal, petroleum and natural gas are finally consumed. New energy and renewable energy will be of great concern to countries around the world. Among various new energy sources, wind power receives increasing attention. Due to the fact that wind power generation is intermittent and unstable, impact on a power grid is large after grid connection, prediction of wind power can help electric power related departments to correspondingly adjust changes of the wind power, adverse effects of wind power grid connection on operation of a power system can be reduced, and prediction of the wind power is important and meaningful.
At present, there are many methods for predicting wind power, wherein the prediction of wind power by using a neural network becomes a current hot spot. However, the learning ability of the traditional neural network on long-term dependence information is insufficient, the problems of gradient disappearance and the like can occur, and the wind power data is random and fluctuating, so that the wind power data is not accurate to predict directly. In addition, the influence of actual meteorological factors is considered, and the wind power is predicted by utilizing the neural network to have some deviation compared with an actual value.
Therefore, the prior art is in need of improvement.
Disclosure of Invention
The invention aims to solve the technical problem that the accuracy rate of wind power prediction in the prior art is not high. The invention provides a short-term wind power prediction method based on a long-term and short-term memory module, which is used for carrying out improved empirical mode decomposition processing on preprocessed data, reducing reconstruction errors caused by white noise increase, and correcting an error sequence by adopting an autoregressive moving average model, so that a target prediction result is closer to an actual value, and a more accurate prediction result is obtained.
In a first aspect, an embodiment of the present invention provides a short-term wind power prediction method based on a long-term and short-term memory module, including:
acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data;
carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
determining a target prediction result based on the corrected error sequence and the initial prediction result.
As a further improved technical solution, the performing improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of content modal components specifically includes:
adding two groups of white noise signals to the preprocessed data respectively to obtain two groups of wind power noise signals;
respectively carrying out improved empirical mode decomposition processing on the two groups of wind power noise signals to obtain two groups of component sequences and residual components;
performing set average processing on the two component sequences to obtain an average component sequence, wherein the average component sequence comprises a plurality of average components;
determining a plurality of abnormal components in the average component sequence, and eliminating the plurality of abnormal components;
and taking the residual component and the plurality of average components which are not eliminated as a plurality of connotative modal components.
As a further improved technical solution, the determining a plurality of abnormal components in the average component sequence and removing the plurality of abnormal components specifically includes:
determining an entropy value of each average component in the average component sequence, and taking a plurality of average components of which the entropy values are larger than a preset threshold value as a plurality of abnormal components;
and eliminating the abnormal components.
As a further improved technical scheme, an initial autoregressive moving average model is trained on the basis of training wind power data and the long-short term memory model to obtain the autoregressive moving average model.
As a further improved technical solution, the training of the initial autoregressive moving average model based on the training wind power data and the long-short term memory model to obtain the autoregressive moving average model specifically includes:
determining a training prediction result based on training wind power data and the long-short term memory model;
determining a training error sequence based on the training prediction and the training wind power data;
when the training error sequence is stable, determining the order according to a minimum information criterion;
and determining a training residual sequence based on the training wind power data, the order and the initial autoregressive moving average model, and when the training residual sequence is not a white noise sequence, continuing to execute the step of determining the order according to a minimum information criterion when the training error sequence is stable until the training residual sequence is the white noise sequence, thus obtaining the autoregressive moving average model.
As a further improved technical scheme, the long-short term memory model is obtained by training an initial long-short term memory model based on a training data set and a learning rate, wherein the learning rate is determined based on a whale optimization algorithm.
As a further improved technical solution, the determining a target prediction result based on the corrected error sequence and the initial prediction result specifically includes:
and adding the corrected error sequence and the initial prediction result to obtain a target prediction result.
As a further improved technical solution, the preprocessing the raw wind power data to obtain preprocessed data specifically includes:
acquiring a maximum value and a minimum value corresponding to the original wind power data;
calculating a first difference between the raw wind power data and the maximum value;
calculating a second difference between the raw wind power data and the minimum value;
and taking the ratio between the first difference and the second difference as the preprocessing data.
In a second aspect, the present invention provides a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:
acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data;
carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
determining a target prediction result based on the corrected error sequence and the initial prediction result.
In a third aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:
acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data;
carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
determining a target prediction result based on the corrected error sequence and the initial prediction result.
Compared with the prior art, the embodiment of the invention has the following advantages:
in the embodiment of the invention, original wind power data are obtained and are preprocessed to obtain preprocessed data; carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components; inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result; determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence; determining a target prediction result based on the corrected error sequence and the initial prediction result. According to the method, the preprocessed data is subjected to improved empirical mode decomposition, reconstruction errors caused by white noise increase can be reduced, an autoregressive moving average model is adopted to correct an error sequence, a target prediction result is closer to an actual value, and a more accurate prediction result is obtained
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart illustrating a short-term wind power prediction method based on a long-term and short-term memory module according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating an initial autoregressive moving average model trained to obtain the autoregressive moving average model, as embodied in the present invention;
FIG. 3 is a flowchart illustrating a short-term wind power prediction method based on a long-term and short-term memory module, according to an embodiment of the present invention;
fig. 4 is a schematic diagram of an internal structure of a computer device according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The inventor researches and discovers that with the rapid development of economy in China, the traditional energy sources of coal, petroleum and natural gas are rapidly consumed, and only in a not too long period, the traditional energy sources of coal, petroleum and natural gas are finally consumed. New energy and renewable energy will be of great concern to countries around the world. Among various new energy sources, wind power receives increasing attention. Due to the fact that wind power generation is intermittent and unstable, impact on a power grid is large after grid connection, prediction of wind power can help electric power related departments to correspondingly adjust changes of the wind power, adverse effects of wind power grid connection on operation of a power system can be reduced, and prediction of the wind power is important and meaningful.
At present, there are many methods for predicting wind power, wherein the prediction of wind power by using a neural network becomes a current hot spot. However, the learning ability of the traditional neural network on long-term dependence information is insufficient, the problems of gradient disappearance and the like can occur, and the wind power data is random and fluctuating, so that the wind power data is not accurate to predict directly. In addition, the influence of actual meteorological factors is considered, and the wind power is predicted by utilizing the neural network to have some deviation compared with an actual value.
In order to solve the problems, the method acquires original wind power data, and preprocesses the original wind power data to obtain preprocessed data; carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components; inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result; determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence; determining a target prediction result based on the corrected error sequence and the initial prediction result. When wind power preprocessing data are decomposed, a MEEMD method is adopted for decomposition, reconstruction errors caused by white noise increase can be reduced, the problem of modal splitting can be suppressed, when a long-short-term memory neural network is used for wind power prediction, a whale optimization algorithm is adopted for optimizing the super-parameter learning rate of the network, the convergence rate of an LSTM model can be increased, the wind power prediction accuracy can be improved, an ARMA model is adopted for correcting an error sequence, a target prediction result is closer to an actual value, and a more accurate prediction result is obtained.
The short-term wind power prediction method based on the long and short-term memory module can be executed on electronic equipment, and the electronic equipment can be realized in various forms, such as a PC (Personal computer), a server, a mobile phone, a tablet computer, a palm computer, a Personal Digital Assistant (PDA) and the like. In addition, the functions realized by the method can be realized by calling the program code by a processor in the electronic equipment, and the program code can be saved in a computer storage medium.
Various non-limiting embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Referring to fig. 1, a method for predicting short-term wind power based on a long-term and short-term memory module in an embodiment of the present invention is shown, including the following steps:
and S1, acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data.
In the embodiment of the invention, the original wind power data are wind power data corresponding to historical moments, and the original wind power data are used for short-term wind power prediction. The preprocessing is specifically normalization processing, and specifically, the preprocessing of the original wind power data to obtain preprocessed data includes: acquiring a maximum value and a minimum value corresponding to the original wind power data; calculating a first difference between the raw wind power data and the maximum value; calculating a second difference between the raw wind power data and the minimum value; and taking the ratio between the first difference and the second difference as the preprocessing data.
In the embodiment of the invention, in order to unify dimensions, the raw wind power data is converted into a numerical value between [0,1] through preprocessing, and the preprocessing process can be expressed by formula (1).
Figure BDA0003124442550000061
Wherein v is2Is pre-processed data, v1Is the raw wind power data, vminIs the minimum value of the raw wind power data, vmaxIs the maximum value of the raw wind power data. V in formula (1)1-vminIs the first difference, vmax-vminIs the second difference.
And S2, performing improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of content modal components.
In the embodiment of the present invention, an improved Empirical mode Decomposition (MEEMD), also called an improved lumped average Empirical mode Decomposition, is improved, so that a mode aliasing phenomenon can be effectively suppressed, white noise can be eliminated, and a real signal can be truly restored. The preprocessed data is decomposed by improving empirical Mode to obtain a plurality of connotation Mode components (IMF).
Specifically, step S2 includes:
and S21, adding two groups of white noise signals to the preprocessed data respectively to obtain two groups of wind power noise signals.
In the embodiment of the invention, the average value of the two groups of white noise signals is 0, and the absolute values of the white noise signals in each group are equal. White noise signal ni(t) and-ni(t) adding to the preprocessed data x (t), two sets of wind power noise signals can be obtained, respectively:
Figure BDA0003124442550000071
and
Figure BDA0003124442550000072
and S22, respectively carrying out improved empirical mode decomposition processing on the two groups of wind power noise signals to obtain two component quantity sequences and residual components.
In an embodiment of the present invention, the improved empirical mode decomposition is an improved algorithm combining empirical mode decomposition and ensemble empirical mode decomposition. The two groups of wind power noise signals comprise positive wind power noise signals and negative wind power noise signals, the positive wind power noise signals are subjected to MEEMD processing to obtain positive component sequences and positive residual components, and the negative wind power noise signals are subjected to MEEMD processing to obtain negative component sequences and negative residual components. The process of performing MEEMD processing on the positive (negative) wind power noise signal to obtain a positive (negative) component sequence and a positive (negative) residual component is the prior art, and is not described herein again.
In the embodiment of the present invention, the residual component is determined based on the positive residual component and the negative residual component, and specifically, an average value of the positive residual component and the negative residual component is calculated to obtain the residual component.
And S23, carrying out collection average processing on the two component quantity sequences to obtain an average component sequence.
In the embodiment of the present invention, the average component sequence includes a plurality of average components, and the average components are arranged in order of frequency from high to low to obtain the average component sequence.
S24, determining a plurality of abnormal components in the average component sequence, and removing the plurality of abnormal components.
In the embodiment of the invention, when the abnormal component occurs, the improved empirical mode decomposition processing is not needed, so the arrangement entropy detection method is adopted to remove the abnormal component in the average component sequence. Specifically, an entropy value of each average component in the average component sequence is determined, a plurality of average components with entropy values larger than a preset threshold value are used as a plurality of abnormal components, and the plurality of abnormal components are removed.
Further, the entropy of each average component in the average component sequence may be determined in turn to detect whether each average component is an abnormal component, and the determining the entropy of each average component in the average component sequence in turn refers to determining the entropy of each average component in turn according to the order of frequencies from high to low. And for an average component in the average component sequence, if the entropy value of the average component is greater than the preset threshold, determining that the average component corresponding to the entropy value is an abnormal component, continuously detecting whether the next average component is the abnormal component or not, and removing all average components before the average component until a certain average component is not the abnormal component. The preset threshold may be set to 0.6.
And S25, taking the residual component and the plurality of average components which are not eliminated as a plurality of connotation modal components.
In the embodiment of the present invention, after the processing in step S24, a plurality of average components that are not eliminated are obtained. For a plurality of Intrinsic Mode Functions (IMF) components, the IMF components may be arranged in order from high to low in frequency to obtain an IMF component sequence.
And S3, inputting the plurality of connotation modal components into the long-short term memory model to obtain an initial prediction result.
In the embodiment of the invention, the Long-Short Term Memory (LSTM) model is a variant of a Recurrent Neural Network (RNN), and can effectively solve the problem of gradient explosion or disappearance of a simple Recurrent Neural Network. The core concepts of LSTM are cell states and various gate structures. The cell state corresponds to a path through which relevant information can be transmitted, allowing the information to be passed on in series. In the training process of the neural network, information can be added or deleted through the gate structure, and different neural networks can decide which information is memorized or forgotten through the gate structure on the unit state.
In the embodiment of the invention, the initial predictor corresponding to each content modal component is determined through the long-term and short-term memory model, and the initial predictor corresponding to each content modal component is summed and reconstructed to obtain the initial prediction result.
In the embodiment of the invention, the long-short term memory model is obtained by training an initial long-short term memory model based on a training data set and a learning rate, wherein the learning rate is determined based on a whale optimization algorithm.
In the embodiment of the invention, the LSTM model has a plurality of hyper-parameters, wherein one of the more important hyper-parameters is the learning rate. The learning rate is used as a hyper-parameter for controlling the network learning speed, if the learning rate is too small, the convergence speed of the LSTM model will become slow, if the learning rate is too large, the LSTM model cannot be converged normally, and in order to enable the LSTM to be converged quickly, the hyper-parameter learning rate of the LSTM is optimized by using a whale optimization algorithm.
The whale optimization algorithm is a new group intelligent optimization algorithm for solving the optimization problem, simulates the behaviors of the whale surrounding, hunting and hunting search, and has the advantages of simplicity in operation, few parameters and strong ability of jumping out of local optimum.
The following describes a process for determining the learning rate of the LSTM model by whale optimization algorithm, including:
creating an LSTM model, and setting the number of hidden units of an LSTM layer to be 288; initializing the learning rate, dimensionality, iteration times and population number of a whale algorithm, and setting preset iteration times; determining a training data set, inputting training connotation modal components in the training data set into the initial long-short term memory model to obtain a training predicted value, wherein the training data set comprises a plurality of groups of training connotation modal components and actual values corresponding to each group of training connotation modal components; determining fitness corresponding to each population respectively based on a training predicted value and an actual value corresponding to the training content modal component, selecting the minimum fitness of all the fitness, and taking the smaller value of the minimum fitness and the preset global optimal fitness as target fitness; updating the learning rate corresponding to the next population by using a whale optimization algorithm; and continuing to execute the step of determining the plurality of fitness degrees based on the actual values respectively corresponding to all the training predicted values and the plurality of training content modal components until the iteration times reach the preset iteration times, so as to obtain the learning rate of the LSTM model.
In the embodiment of the invention, because an Adaptive motion Estimation (Adam) optimizer is provided with a momentum term, the training speed of the LSTM model can be accelerated, an LSTM solver is set as Adam, and the Adam is the optimizer for comprehensively considering the mean value of the gradient and the variance of the gradient without centralization and calculating the update step length. The threshold value of the gradient is set to 1, the initial learning rate may be set to 0.005, the preset number of iterations may be set to 250, and the learning rate decrease number may be set to 125.
In the embodiment of the present invention, a training data set is determined, that is, a process of determining a plurality of sets of training content modal components and an actual value corresponding to each set of training content modal components is determined. Each group of training content modal components comprises a plurality of training content modal components, the training content modal components are determined based on one training wind power data, and the real prediction result corresponding to the training wind power data is the actual value corresponding to the group of training content modal components. Determining a plurality of training content modal components based on one training wind power data is consistent with the process of determining a plurality of content modal components in step S2, and is not described herein again.
In the embodiment of the present invention, multiple sets of content modal components may be determined based on multiple wind power data, 80% of the multiple sets of content modal components are used as multiple sets of training content modal components, and 20% of the multiple sets of content modal components are used as multiple sets of testing content modal components. And the multiple groups of test connotation modal components are used for testing whether the long-term and short-term memory model obtained by training can be put into use or not.
In the embodiment of the invention, after the learning rate is determined based on a whale optimization algorithm, the learning rate is adopted, the initial long-short term memory model is trained based on the training data set to obtain the long-short term memory model, and the model structure of the initial long-short term memory model is the same as that of the long-short term memory model.
And S4, determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence.
In the embodiment of the invention, the result of predicting the wind power by the LSTM model has some deviation compared with the actual value in consideration of the influence of actual meteorological factors. And the ARMA model corrects the initial prediction result to obtain an accurate wind power prediction result. The Auto Regression Moving Average (ARMA) model is an important method for a stationary time series model and for studying time series. The ARMA model combines two models, Auto Regression (AR) and Moving Average (MA). The ARMA model can be expressed by equation (2).
Xt＝γ1Xt-1+γ2Xt-2+…+γpXt-p+αt-β1αt-1-β2tt-2-…-βqαt-q (2)
Wherein the content of the first and second substances,p and q are the order of the ARMA model; gamma ray1，γ2,…,γpIs an autoregressive parameter; beta is a1,β2,…,βqIs a moving average parameter; alpha is alphatParameters are estimated for the pending.
In the embodiment of the invention, the difference between the original wind power data and the initial prediction result is calculated to obtain an error sequence, and the ARMA model is adopted to correct the error sequence to obtain a corrected error sequence.
In the embodiment of the invention, an initial autoregressive moving average model is trained on the basis of training wind power data and the long-short term memory model to obtain the autoregressive moving average model. The model structure of the initial autoregressive moving average model is the same as that of the autoregressive moving average model.
Specifically, training an initial autoregressive moving average model based on training wind power data and the long-short term memory model to obtain the autoregressive moving average model includes:
determining a training prediction result based on training wind power data and the long-short term memory model; determining a training error sequence based on the training prediction and the training wind power data; when the training error sequence is stable, determining the order according to a minimum information criterion; and determining a training residual sequence based on the training wind power data, the order and the initial autoregressive moving average model, and when the training residual sequence is not a white noise sequence, continuing to execute the step of determining the order according to a minimum information criterion when the training error sequence is stable until the training residual sequence is the white noise sequence, thus obtaining the autoregressive moving average model.
In the embodiment of the present invention, the process of determining the training prediction result based on the training wind power data and the long-short term memory model is the same as the process of step S1 to step S3, and therefore, the process of determining the training prediction result based on the training wind power data and the long-short term memory model may refer to the description of step S1 to step S3. And determining a training error sequence based on the training prediction result and the training wind power data, specifically, calculating a difference between the training prediction result and the training wind power data to obtain the training error sequence.
In the embodiment of the invention, after the training error sequence is obtained, the stationarity of the training error sequence is firstly determined, and if the training error sequence is stable, the order is determined. And if the training error sequence is not stable, carrying out stabilization processing on the training error sequence. The stationarity of the training error sequence can be determined using existing methods, for example by the unit root test (ADF).
When the training error sequence is stable, judging and identifying the order of the training error sequence through an autocorrelation function (ACF) and a partial autocorrelation function (PACF) truncation, and taking a minimum Information Criterion (Akaike Information Criterion, AIC) as a fixed-order standard. After the ARMA order is determined, inputting training wind power data into an ARMA model with the determined order to obtain a fitting sequence, obtaining a training residual sequence according to the training wind power data and the fitting sequence, and if the training residual sequence is a white noise sequence, taking the determined order of the ARMA model as an optimal order to obtain a trained autoregressive moving average model; and if the training residual sequence is not a white noise sequence, determining that the order of the ARMA model is not the optimal order, and continuously executing the step of determining the order according to the minimum information criterion when the training error sequence is stable until the training residual sequence is the white noise sequence to obtain the trained autoregressive moving average model.
For ease of understanding, referring to fig. 2, training an initial autoregressive moving average model to obtain the autoregressive moving average model includes:
step 11, determining a training prediction result based on training wind power data and the long-short term memory model, and determining a training error sequence based on the training prediction result and the training wind power data;
step 12, judging whether the training error sequence is stable, if not, entering step 13, and if so, entering step 14;
step 13, carrying out stabilization processing on the training error sequence, and entering step 12;
step 14, determining the order;
step 15, inputting training wind power data into an initial autoregressive moving average model with a determined order to obtain a fitting sequence;
step 16, obtaining a training residual sequence according to the training wind power data and the fitting sequence;
step 17, judging whether the training residual sequence is a white noise sequence, if so, entering step 18, and if not, entering step 14;
and step 18, obtaining the trained autoregressive moving average model.
And S5, determining a target prediction result based on the corrected error sequence and the initial prediction result.
In the embodiment of the invention, the corrected error sequence and the initial prediction result are added to obtain a target prediction result.
For ease of understanding, referring to fig. 3, in one embodiment, the method for predicting short-term wind power based on the long-term and short-term memory module includes:
step 101, acquiring original wind power data;
step 102, preprocessing the original wind power data to obtain preprocessed data;
103, performing improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
104, inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
step 105, determining an error sequence based on the original wind power data and the initial prediction result;
step 106, inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
and step 107, determining a target prediction result based on the corrected error sequence and the initial prediction result.
In one embodiment, the invention verifies that the short-term wind power prediction method based on the long-term and short-term memory module can obtain more accurate prediction results. The method comprises the steps of collecting 10 days of data of a certain wind power plant, sampling once every 15 minutes, collecting 960 groups of data in total, dividing the first 80% of the data into a training set, dividing the second 20% of the data into a test set, performing simulation test by Matlab2020b software, and displaying the final simulation result, wherein the result errors of LSTM prediction optimized by whale algorithm are corrected by an ARMA model and are smaller than the result errors of the LSTM prediction optimized by whale algorithm, wherein the Mean Absolute Percentage Error (MAPE) of the LSTM optimized by whale algorithm corrected by the ARMA model is reduced by 6.125% compared with the MAPE of the LSTM optimized by whale algorithm, and the Root Mean Square Error (RMSE) of the LSTM optimized by whale algorithm corrected by the ARMA model is reduced by 6.225% compared with the RMSE of the LSTM optimized by whale algorithm. Therefore, the method can fully show that the ARMA model is adopted to correct the LSTM model optimized by the whale algorithm, the error of wind power prediction is smaller, and a more accurate prediction result can be obtained.
In the embodiment of the invention, original wind power data are obtained and are preprocessed to obtain preprocessed data; carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components; inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result; determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence; determining a target prediction result based on the corrected error sequence and the initial prediction result. According to the method, the preprocessed data are decomposed by adopting the MEEMD method, so that reconstruction errors caused by white noise increase can be reduced, and the problem of modal splitting can be suppressed; when the long-term and short-term memory neural network is used for predicting the wind power, the whale optimization algorithm is adopted to optimize the over-parameter learning rate of the network, the convergence speed of the LSTM model can be increased, and the wind power prediction accuracy can be improved; and correcting the error sequence by adopting an ARMA (autoregressive moving average) model, so that the target prediction result is closer to the actual value, and a more accurate prediction result is obtained.
The embodiment of the invention also provides computer equipment which can be a terminal, and the internal structure of the computer equipment is shown in figure 4. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a short-term wind power prediction method based on a long-term and short-term memory module. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the illustration in fig. 4 is merely a block diagram of a portion of the structure associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
The embodiment of the present invention further provides a computer device, which includes a memory and a processor, where the memory stores a computer program, and the processor implements the following steps when executing the computer program:
acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data;
carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
determining a target prediction result based on the corrected error sequence and the initial prediction result.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the following steps:
acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data;
carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
determining a target prediction result based on the corrected error sequence and the initial prediction result.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

Claims (10)

1. A short-term wind power prediction method based on a long-term and short-term memory module is characterized by comprising the following steps:
acquiring original wind power data, and preprocessing the original wind power data to obtain preprocessed data;
carrying out improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of connotation modal components;
inputting the plurality of connotation modal components into a long-term and short-term memory model to obtain an initial prediction result;
determining an error sequence based on the original wind power data and the initial prediction result, and inputting the error sequence into an autoregressive moving average model to obtain a corrected error sequence;
determining a target prediction result based on the corrected error sequence and the initial prediction result.
2. The method for short-term wind power prediction based on a long-term and short-term memory module as claimed in claim 1, wherein the performing improved empirical mode decomposition processing on the preprocessed data to obtain a plurality of content modal components specifically comprises:
adding two groups of white noise signals to the preprocessed data respectively to obtain two groups of wind power noise signals;
respectively carrying out improved empirical mode decomposition processing on the two groups of wind power noise signals to obtain two groups of component sequences and residual components;
performing set average processing on the two component sequences to obtain an average component sequence, wherein the average component sequence comprises a plurality of average components;
determining a plurality of abnormal components in the average component sequence, and eliminating the plurality of abnormal components;
and taking the residual component and the plurality of average components which are not eliminated as a plurality of connotative modal components.
3. The method for short-term wind power prediction based on long-term and short-term memory module as claimed in claim 2, wherein the determining several abnormal components in the average component sequence and removing the several abnormal components specifically comprises:
determining an entropy value of each average component in the average component sequence, and taking a plurality of average components of which the entropy values are larger than a preset threshold value as a plurality of abnormal components;
and eliminating the abnormal components.
4. The long short term memory module based short term wind power prediction method of claim 1, wherein the autoregressive moving average model is obtained by training an initial autoregressive moving average model based on training wind power data and the long short term memory model.
5. The method for short-term wind power prediction based on a long-term and short-term memory module as claimed in claim 4, wherein the training of the initial autoregressive moving average model based on the training wind power data and the long-term and short-term memory model to obtain the autoregressive moving average model specifically comprises:
determining a training prediction result based on training wind power data and the long-short term memory model;
determining a training error sequence based on the training prediction and the training wind power data;
when the training error sequence is stable, determining the order according to a minimum information criterion;
and determining a training residual sequence based on the training wind power data, the order and the initial autoregressive moving average model, and when the training residual sequence is not a white noise sequence, continuing to execute the step of determining the order according to a minimum information criterion when the training error sequence is stable until the training residual sequence is the white noise sequence, thus obtaining the autoregressive moving average model.
6. A long-short term memory module based short term wind power prediction method as claimed in claim 1, characterized in that the long-short term memory model is trained on an initial long-short term memory model based on a training data set, learning rate, wherein the learning rate is determined based on whale optimization algorithm.
7. The method for short-term wind power prediction based on a long-term and short-term memory module as claimed in claim 1, wherein the determining a target prediction result based on the corrected error sequence and the initial prediction result specifically comprises:
and adding the corrected error sequence and the initial prediction result to obtain a target prediction result.
8. The method for predicting short-term wind power based on the long-term and short-term memory module as claimed in any one of claims 1 to 7, wherein the pre-processing the original wind power data to obtain pre-processed data specifically comprises:
acquiring a maximum value and a minimum value corresponding to the original wind power data;
calculating a first difference between the raw wind power data and the maximum value;
calculating a second difference between the raw wind power data and the minimum value;
and taking the ratio between the first difference and the second difference as the preprocessing data.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the long-short term memory module based short term wind power prediction method of any of claims 1 to 8.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method for short-term wind power prediction based on a long-short-term memory module of any one of claims 1 to 8.
CN202110685441.XA 2021-06-21 2021-06-21 Short-term wind power prediction method based on long-term and short-term memory module Active CN113487068B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110685441.XA CN113487068B (en) 2021-06-21 2021-06-21 Short-term wind power prediction method based on long-term and short-term memory module

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110685441.XA CN113487068B (en) 2021-06-21 2021-06-21 Short-term wind power prediction method based on long-term and short-term memory module

Publications (2)

Publication Number Publication Date
CN113487068A true CN113487068A (en) 2021-10-08
CN113487068B CN113487068B (en) 2023-03-24

Family

ID=77935593

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110685441.XA Active CN113487068B (en) 2021-06-21 2021-06-21 Short-term wind power prediction method based on long-term and short-term memory module

Country Status (1)

Country Link
CN (1) CN113487068B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836756A (en) * 2021-11-29 2021-12-24 山东华尚电气有限公司 Intelligent monitoring method and system for annealing process of three-dimensional wound core transformer

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103400230A (en) * 2013-08-08 2013-11-20 上海电机学院 Wind power forecast system and method
CN105956722A (en) * 2016-06-22 2016-09-21 广东工业大学 Short-term wind power prediction method and apparatus
CN109636066A (en) * 2019-01-04 2019-04-16 广东工业大学 A kind of wind power output power prediction technique based on fuzzy time series data mining
CN110298494A (en) * 2019-06-11 2019-10-01 东南大学 A kind of wind power forecasting method based on Segment Clustering and Combinatorial Optimization
WO2020203854A1 (en) * 2019-03-29 2020-10-08 三菱重工業株式会社 Power generation amount prediction device, power generation amount prediction method, and program
CN112906935A (en) * 2020-12-21 2021-06-04 湘潭大学 Method for predicting ultra-short-term power of wind power plant
CN112989693A (en) * 2021-03-02 2021-06-18 上海电机学院 Wind power prediction method based on SSA-GRU-MSAR

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103400230A (en) * 2013-08-08 2013-11-20 上海电机学院 Wind power forecast system and method
CN105956722A (en) * 2016-06-22 2016-09-21 广东工业大学 Short-term wind power prediction method and apparatus
CN109636066A (en) * 2019-01-04 2019-04-16 广东工业大学 A kind of wind power output power prediction technique based on fuzzy time series data mining
WO2020203854A1 (en) * 2019-03-29 2020-10-08 三菱重工業株式会社 Power generation amount prediction device, power generation amount prediction method, and program
CN110298494A (en) * 2019-06-11 2019-10-01 东南大学 A kind of wind power forecasting method based on Segment Clustering and Combinatorial Optimization
CN112906935A (en) * 2020-12-21 2021-06-04 湘潭大学 Method for predicting ultra-short-term power of wind power plant
CN112989693A (en) * 2021-03-02 2021-06-18 上海电机学院 Wind power prediction method based on SSA-GRU-MSAR

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
陈籽君: "基于MEEMD-GAELM组合模型的短期风电功率预测", 《水电能源科学》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836756A (en) * 2021-11-29 2021-12-24 山东华尚电气有限公司 Intelligent monitoring method and system for annealing process of three-dimensional wound core transformer
CN113836756B (en) * 2021-11-29 2022-03-18 山东华尚电气有限公司 Intelligent monitoring method and system for annealing process of three-dimensional wound core transformer

Also Published As

Publication number Publication date
CN113487068B (en) 2023-03-24

Similar Documents

Publication Publication Date Title
CN112215442B (en) Method, system, device and medium for predicting short-term load of power system
JP2009225550A (en) Power demand predicting method and system, and power generation prediction method
CN111144644B (en) Short-term wind speed prediction method based on variation variance Gaussian process regression
CN113657661A (en) Enterprise carbon emission prediction method and device, computer equipment and storage medium
CN113487068B (en) Short-term wind power prediction method based on long-term and short-term memory module
JP7062144B1 (en) Power demand forecasting device and power demand forecasting method
Zhang et al. Interval prediction of ultra-short-term photovoltaic power based on a hybrid model
CN114841199A (en) Power distribution network fault diagnosis method, device, equipment and readable storage medium
CN114330934A (en) Model parameter self-adaptive GRU new energy short-term power generation power prediction method
Aizpurua et al. Probabilistic forecasting informed failure prognostics framework for improved RUL prediction under uncertainty: A transformer case study
Sen Time Series Prediction based on Improved Deep Learning
CN112307672A (en) BP neural network short-term wind power prediction method based on cuckoo algorithm optimization
CN115935822A (en) Method, system, device and medium for predicting long-term degradation trend of computer parameters
Li et al. Short-term probabilistic load forecasting method based on uncertainty estimation and deep learning model considering meteorological factors
CN113011674A (en) Photovoltaic power generation prediction method and device, electronic equipment and storage medium
CN112883649A (en) Power load prediction method, system, computer equipment and storage medium
Shen et al. An interval analysis scheme based on empirical error and mcmc to quantify uncertainty of wind speed
Fu et al. Enhanced machine learning model via twin support vector regression for streamflow time series forecasting of hydropower reservoir
Feng Garlic price forecast based on the combined model of time-frequency decomposition and neural network
Chen et al. Short-term wind power prediction based on BP neural network improved by t-tent-SSA algorithm
CN112685900B (en) Power load simulation method for representing impact load power characteristics
Riquelme-Dominguez et al. A machine learning-based methodology for short-term kinetic energy forecasting with real-time application: Nordic Power System case
US11960805B2 (en) Architecture exploration and compiler optimization using neural networks
CN111461418B (en) Wind speed prediction method, system, electronic equipment and storage medium
Zhang et al. Probabilistic forecasting of power grid project cost control index based on a variational Bayesian deep learning model

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