CN115222138A - Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid - Google Patents

Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid Download PDF

Info

Publication number
CN115222138A
CN115222138A CN202210901508.3A CN202210901508A CN115222138A CN 115222138 A CN115222138 A CN 115222138A CN 202210901508 A CN202210901508 A CN 202210901508A CN 115222138 A CN115222138 A CN 115222138A
Authority
CN
China
Prior art keywords
data
prediction
photovoltaic
power generation
sequence
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.)
Withdrawn
Application number
CN202210901508.3A
Other languages
Chinese (zh)
Inventor
沈中元
朱晔
徐伟明
赵家振
章思亮
苏丹
王霈霈
张泽宇
黄竹青
鲍双双
胡杨军
沈梦笑
姜琪健
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.)
Jiaxing Hengchuang Electric Power Design And Research Institute Co ltd
Jiaxing Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
Original Assignee
Jiaxing Hengchuang Electric Power Design And Research Institute Co ltd
Jiaxing Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
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 Jiaxing Hengchuang Electric Power Design And Research Institute Co ltd, Jiaxing Power Supply Co of State Grid Zhejiang Electric Power Co Ltd filed Critical Jiaxing Hengchuang Electric Power Design And Research Institute Co ltd
Priority to CN202210901508.3A priority Critical patent/CN115222138A/en
Publication of CN115222138A publication Critical patent/CN115222138A/en
Withdrawn legal-status Critical Current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
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • 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)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Tourism & Hospitality (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Mathematics (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • Databases & Information Systems (AREA)
  • Molecular Biology (AREA)
  • Algebra (AREA)
  • Power Engineering (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)

Abstract

The invention relates to the technical field of power interval prediction, in particular to a photovoltaic short-term power interval prediction method based on an EEMD-LSTM microgrid, which comprises the following steps: s1, data input: inputting photovoltaic power generation related data including information such as meteorological factors, power generation system parameters, photovoltaic output and the like; s2, data preprocessing: the integration of multi-source data is mainly to integrate and summarize data in a multi-file or database so as to analyze data characteristics of different levels, and a KNN algorithm is adopted to detect missing data. According to the photovoltaic short-term power interval prediction method based on the EEMD-LSTM microgrid, through the photovoltaic output sample feature extraction method based on the XGboost, model training can be avoided by directly utilizing a large number of features, factor screening can be performed on factors simultaneously by considering influence on photovoltaic power more comprehensively, and modeling efficiency is improved.

Description

Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid
Technical Field
The invention relates to the technical field of power interval prediction, in particular to a photovoltaic short-term power interval prediction method based on an EEMD-LSTM microgrid.
Background
With the proposal of carbon peak reaching and carbon neutral targets, china plays an increasingly important role in green treatment for dealing with world environment and climate change, large-scale development and utilization of renewable new energy, particularly solar energy, is inevitable, photovoltaic power generation is introduced into a plurality of comprehensive energy experiment parks at present, for a comprehensive energy microgrid system of one park, because the power load of the system is limited, the photovoltaic power generation has very strong intermittency and randomness, and the quantity of the generated power has great influence on the safe and stable operation of the system, therefore, reliable photovoltaic power generation prediction plays an important role in the microgrid power system, and the scheme of the prior art for predicting the photovoltaic power of the microgrid has the following steps:
(1) A power prediction method based on physical modeling;
(2) A power prediction method based on statistical modeling;
(3) Conventional artificial intelligence power prediction methods.
The prior art has the defects that:
(1) The influence factors of the photovoltaic power generation power are not fully considered;
(2) The prediction accuracy is low;
(3) Uncertainty in point prediction error, etc.
Therefore, a photovoltaic short-term power prediction method meeting actual requirements and the like needs to be established, and powerful support is provided for stable operation and accurate control of the microgrid.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a photovoltaic short-term power interval prediction method based on an EEMD-LSTM microgrid, which has the advantages that the XGboost method is adopted, the characteristics of photovoltaic power samples can be accurately extracted, the calculation complexity of a model is obviously reduced, the photovoltaic power prediction accuracy is improved by combining the EEMD method and the LSTM method, the uncertainty problem of photovoltaic power point prediction errors is solved by adopting an error fitting interval prediction method, the problems of incomplete consideration of photovoltaic power generation power influence factors, low prediction accuracy, uncertainty of point prediction errors and the like are solved.
(II) technical scheme
In order to achieve the purpose, the invention provides the following technical scheme: a photovoltaic short-term power interval prediction method based on an EEMD-LSTM microgrid comprises the following steps:
s1, data input: inputting photovoltaic power generation related data including information such as meteorological factors, power generation system parameters, photovoltaic output and the like;
s2, data preprocessing: integrating multi-source data mainly by integrating and summarizing data in multiple files or databases to analyze data characteristics of different levels, detecting missing data by adopting a KNN algorithm, constructing a unitary linear regression model by using MIC (many integrated computer) to complete missing data, and then standardizing the data;
s3, photovoltaic output sample feature extraction: performing correlation analysis on photovoltaic power generation output factors, screening all factors by adopting an XGboost algorithm, and extracting photovoltaic power generation sample characteristics;
s4, EEMD decomposition: processing photovoltaic power generation output sequence data by adopting a modal decomposition (EEMD) method, and decomposing the data into a plurality of IMF components and residual components;
s5, LSTM parameter setting and training: adopting a first-order two-layer structure LSTM, setting the batch size to be 96, determining the time length of an input variable by one point in 15 minutes based on MIC, determining the number of nodes of an LSTM layer by selecting sample characteristics and the time length of the input variable, determining the number of nodes of a hidden layer by an empirical formula, and performing model training by using decomposed sequence data;
s6, interval prediction: and superposing and reducing the prediction result of the decomposition sequence to obtain a photovoltaic power generation output prediction result, fitting the error between the prediction result and an actual value by utilizing t distribution to obtain an error probability density function, and representing the possible occurrence range of the prediction error by using a confidence interval.
Preferably, the XGboost algorithm screens all variables to extract characteristics of the photovoltaic power generation sample, the XGboost uses a decision tree as a bottom model, and the output is as shown in a formula (1):
Figure BDA0003771043130000031
secondly, the optimization goal of the XGboost model is to minimize the structural risk, and an objective function is as follows:
Figure BDA0003771043130000032
the first item of
Figure BDA0003771043130000033
Is the residual, second term
Figure BDA0003771043130000034
Is a regular term.
The simplified objective function is to sum the loss function values of each sample to obtain the following equation:
Figure BDA0003771043130000035
by rewriting the above formula, we can rewrite the objective function as a unitary quadratic function about the leaf node coefficient w, and can find the optimal w and the minimum value of the objective function as:
Figure BDA0003771043130000036
Figure BDA0003771043130000037
Figure BDA0003771043130000038
Figure BDA0003771043130000039
and traversing the feature division points of all the features by using a greedy algorithm, searching a maximum gain node according to layers for expansion, and then adding a new decision tree in the model to automatically acquire the importance of the features for feature screening.
Preferably, standard normal distribution white noise n is added to the photovoltaic power generation output sequence data x (t) i (t) obtaining new sequence data x i (t)＝x(t)+n i (t), where i represents the number of white noise additions, i =1, 2.., M, new contribution sequence data x that will contain white noise i (t) EMD decomposition is carried out in sequence to obtain the representation forms of IMF component and residual component
Figure BDA0003771043130000041
Wherein J represents the number of IMF components, c (t) represents the IMF components, r (t) represents the remaining components, each IMF component c is calculated M times i，j (t) has a statistical average of
Figure BDA0003771043130000042
As the jth IMF component of the original force sequence.
And evaluating the power generation output decomposition effect by using a signal-to-noise ratio (SNR) index. When the SNR value is larger, the ratio of the decomposed sequence to the noise sequence is large, and the decomposition effect is good; when the smaller SNR value indicates that the ratio of the decomposed sequence to the noise sequence is small, the decomposition effect is poor, and the SNR is defined as follows:
Figure BDA0003771043130000043
wherein S is j Representing the energy of the jth IMF component of the power generation output sequence, N representing the energy of noise, and N representing the length of the power generation output sequence;
and taking M when the SNR value is maximum as the EEMD decomposition times, and finally completing the mode decomposition of the original output sequence.
Preferably, a first-order two-layer structure is adopted, and the specific parameters of the model are set as follows:
the training times are set to 1000 times, and the number of LSTM layer nodes is set to 50; the batch size is set to 72, the average absolute error function is selected as a loss function, adam is used as an optimization function, tanh is used as an activation function, and due to the fact that the data volume is large, dropout is set to be 0.2 for relieving the overfitting phenomenon.
Preferably, the prediction result of the decomposition sequence is subjected to superposition reduction to obtain the prediction result of the photovoltaic power generation output, and the error between the prediction result and the actual value is fitted by utilizing t distribution to obtain an error probability density function;
the range in which prediction errors are likely to occur is represented by a confidence interval, and when the significance level is α and the confidence is 1- α, the confidence interval of the error distribution can be represented by (μ -Z) α/2 σ，μ+Z α/2 σ), where μ represents the mean value of the errors, σ represents the standard deviation of the errors, Z α/2 Is corresponding to standard score (z-score);
and evaluating the prediction result of the prediction interval by using the coverage rate and the width of the prediction interval, wherein the PICP represents the ratio of the true value between the upper boundary and the lower boundary of the prediction interval, and the formula is as follows:
Figure BDA0003771043130000051
wherein
Figure BDA0003771043130000052
n is the number of samples, L i ，U i Lower and upper bound values are predicted for the ith.
PINAW represents the average distance between the upper and lower boundaries of the prediction interval and is used for reflecting the accuracy of the prediction interval, and the calculation formula is as follows:
Figure BDA0003771043130000053
(III) advantageous effects
Compared with the prior art, the invention provides a photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid, which has the following beneficial effects:
according to the photovoltaic short-term power interval prediction method based on the EEMD-LSTM microgrid, model training can be avoided by directly utilizing a large number of features through a photovoltaic output sample feature extraction method based on the XGboost, influence on photovoltaic power can be considered more comprehensively, factor screening can be carried out on factors simultaneously, modeling efficiency is improved, a photovoltaic power short-term prediction model is established based on the EEMD-LSTM, photovoltaic power prediction accuracy is improved, finally, the uncertainty problem of photovoltaic power point prediction errors is solved by adopting an error fitting interval prediction method, and prediction interval coverage rate and prediction interval width are used for reflecting the accuracy of a prediction interval.
Drawings
Fig. 1 is a schematic diagram of a photovoltaic short-term power interval prediction process based on an EEMD-LSTM microgrid provided by the invention.
Detailed Description
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.
Referring to fig. 1, a photovoltaic short-term power interval prediction method based on EEMD-LSTM micro grid includes the following steps:
s1, data input: inputting photovoltaic power generation related data including information such as meteorological factors, power generation system parameters, photovoltaic output and the like;
s2, data preprocessing: integrating multi-source data mainly by integrating and summarizing data in a multi-file or database to analyze data characteristics of different layers, detecting missing data by adopting a KNN algorithm, constructing a unitary linear regression model by using MIC (many integrated circuits) to complete missing data, and then standardizing the data;
s3, photovoltaic output sample characteristic extraction: performing correlation analysis on photovoltaic power generation output factors, screening all factors by adopting an XGboost algorithm, and extracting photovoltaic power generation sample characteristics;
s4, EEMD decomposition: processing photovoltaic power generation output sequence data by adopting a modal decomposition (EEMD) method, and decomposing the data into a plurality of IMF components and residual components;
s5, LSTM parameter setting and training: adopting a first-order two-layer structure LSTM, setting the batch size to be 96, determining the time length of an input variable by one point in 15 minutes based on MIC, determining the number of nodes of an LSTM layer by selecting sample characteristics and the time length of the input variable, determining the number of nodes of a hidden layer by an empirical formula, and performing model training by using decomposed sequence data;
s6, interval prediction: and superposing and reducing the prediction result of the decomposition sequence to obtain a photovoltaic power generation output prediction result, fitting the error between the prediction result and an actual value by utilizing t distribution to obtain an error probability density function, and representing the possible occurrence range of the prediction error by using a confidence interval.
In the photovoltaic short-term power prediction model building process, the traditional method usually selects input characteristic variables based on experience, or directly utilizes a large number of characteristics to start model training without considering the correlation between the input variables and the output variables, and the conditions of low precision and time consumption of training exist, the XGboost algorithm is adopted to screen the variables, the characteristics of a photovoltaic power generation sample are extracted, the XGboost uses a decision tree as a bottom model, and the output is shown as a formula (1):
Figure BDA0003771043130000071
secondly, the optimization goal of the XGboost model is to minimize the structural risk, and the objective function is as follows:
Figure BDA0003771043130000072
the first item
Figure BDA0003771043130000073
Is the residual, the second term
Figure BDA0003771043130000074
Is a regular term;
the simplified objective function is to sum the loss function values of each sample to obtain the following equation:
Figure BDA0003771043130000075
by rewriting the above formula, we can rewrite the objective function as a unitary quadratic function about the leaf node coefficient w, and can find the optimal w and the minimum value of the objective function as:
Figure BDA0003771043130000076
Figure BDA0003771043130000077
Figure BDA0003771043130000078
Figure BDA0003771043130000079
traversing feature division points of all features by using a greedy algorithm, searching a maximum gain node according to layers for expansion, then adding a new decision tree in the model, automatically acquiring the importance of the features, screening the features, expressing the importance of one feature on the times of taking the feature as a division node, and expressing that the feature has obvious discrimination on a sample if the times are more, and finally averaging the importance of elements in all decision trees in the model to obtain the importance of each feature;
in order to solve the problem of low prediction accuracy of the traditional method, the invention introduces the EEMD method to carry out modal decomposition on the photovoltaic power generation sequence, and carries out prediction based on each component.
Because photovoltaic power generation output is related to various factors, output sequence data has non-stationarity and nonlinearity, and the method adopts a modal decomposition (EEMD) method to process the sequence data, compared with other decomposition methods, the EEMD method has a better decomposition effect while inhibiting modal aliasing.
Adding standard normal distribution white noise n into photovoltaic power generation output sequence data x (t) i (t) obtaining new sequence data x i (t)＝x(t)+n i (t) where i represents the number of white noise additions, i =1, 2.. M, new contribution sequence data x that will contain white noise i (t) EMD decomposition is carried out in sequence to obtain the representation forms of IMF component and residual component
Figure BDA0003771043130000081
Wherein J represents the number of IMF components, c (t) represents the IMF components, r (t) represents the remaining components, each IMF component c is calculated M times i，j (t) a statistical average of
Figure BDA0003771043130000082
As the jth IMF component of the original force sequence.
Evaluating the decomposition effect of the generated output by adopting a signal-to-noise ratio (SNR) index, wherein the larger the SNR value is, the larger the ratio of the decomposition sequence to the noise sequence is, the good decomposition effect is obtained, and the smaller the SNR value is, the smaller the ratio of the decomposition sequence to the noise sequence is, the poor decomposition effect is obtained, and the SNR is defined as follows:
Figure BDA0003771043130000091
wherein S is j The energy of the jth IMF component of the power generation output sequence is shown, N is the energy of noise, and N is the length of the power generation output sequence.
Taking M when the SNR value is maximum as the EEMD decomposition times, and finally completing the mode decomposition of the original output sequence;
the structure of a first-order two-layer structure is adopted, and the specific parameters of the model are set as follows:
the training times are set to be 1000 times, the number of nodes of an LSTM layer is set to be 50, the batch size is set to be 72, an average absolute error function is selected as a loss function, adam is used as an optimization function, tanh is used as an activation function, and due to the fact that the data volume is large, dropout is set to be 0.2 in order to relieve the overfitting phenomenon.
In order to solve the problem of uncertainty of point prediction errors in the traditional method, the invention introduces an error fitting interval prediction method.
And (4) superposing and reducing the prediction result of the decomposition sequence to obtain the prediction result of the photovoltaic power generation output, and fitting the error between the prediction result and the actual value by utilizing t distribution to obtain an error probability density function.
The confidence interval represents the range in which the prediction error is likely to occur, and when the significance level is α and the confidence is 1- α, the confidence interval of the error distribution can be represented as (μ -Z) α/2 σ，μ+Z α/2 σ), where μ represents the mean value of the errors, σ represents the standard deviation of the errors, Z α/2 As corresponding to a standard score (z-score).
The interval prediction results were evaluated using the Prediction Interval Coverage (PICP), which represents the ratio of the true value between the upper and lower bounds of the prediction interval, and the prediction interval width (PINAW), as follows:
Figure BDA0003771043130000092
wherein
Figure BDA0003771043130000101
n is the number of samples, L i ，U i Lower and upper bound values are predicted for the ith.
PINAW represents the average distance between the upper and lower boundaries of the prediction interval and is used for reflecting the accuracy of the prediction interval, and the calculation formula is as follows:
Figure BDA0003771043130000102
the electrical components presented in the document are all electrically connected with an external master controller and 220V mains, and the master controller can be a conventional known device controlled by a computer or the like.
When the photovoltaic power generation system is used, the general idea of the proposal of the application is to firstly collect and arrange the relevant historical data of photovoltaic power generation, including meteorological data, power generation system parameters, photovoltaic output data and the like, and carry out data preprocessing, including data integration, data cleaning and data standardization; then analyzing the photovoltaic output influence factors and extracting the characteristics of the photovoltaic output sample; finally, photovoltaic power short-term interval prediction is carried out based on EEMD-LSTM and error fitting, and the specific flow is shown in figure 1.
In summary, according to the photovoltaic short-term power interval prediction method based on the EEMD-LSTM microgrid, model training can be avoided by directly utilizing a large number of features through the photovoltaic output sample feature extraction method based on the XGboost, the influence on photovoltaic power can be considered more comprehensively, factor screening can be carried out on factors at the same time, the modeling efficiency is improved, the photovoltaic power short-term prediction model is established based on the EEMD-LSTM, the photovoltaic power prediction accuracy is improved, finally, the uncertainty problem of photovoltaic power point prediction errors is solved by adopting an error fitting interval prediction method, and the prediction interval accuracy is reflected by using the prediction interval coverage rate and the prediction interval width.
It should be noted that, in this document, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, the use of the verb "comprise a" to define an element does not exclude the presence of another, same element in a process, method, article, or apparatus that comprises the element.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (5)

1. The photovoltaic short-term power interval prediction method based on the EEMD-LSTM microgrid is characterized by comprising the following steps of:
s1, data input: inputting photovoltaic power generation related data including information such as meteorological factors, power generation system parameters, photovoltaic output and the like;
s2, data preprocessing: integrating multi-source data mainly by integrating and summarizing data in a multi-file or database to analyze data characteristics of different layers, detecting missing data by adopting a KNN algorithm, constructing a unitary linear regression model by using MIC (many integrated circuits) to complete missing data, and then standardizing the data;
s3, photovoltaic output sample feature extraction: performing correlation analysis on photovoltaic power generation output factors, screening all factors by adopting an XGboost algorithm, and extracting photovoltaic power generation sample characteristics;
s4, EEMD decomposition: processing photovoltaic power generation output sequence data by adopting a modal decomposition (EEMD) method, and decomposing the data into a plurality of IMF components and residual components;
s5, LSTM parameter setting and training: adopting a first-order two-layer structure LSTM, setting the batch size to be 96, determining the time length of an input variable by one point in 15 minutes based on MIC, determining the number of nodes of an LSTM layer by selecting sample characteristics and the time length of the input variable, determining the number of nodes of a hidden layer by an empirical formula, and performing model training by using decomposed sequence data;
s6, interval prediction: and superposing and reducing the prediction result of the decomposition sequence to obtain a photovoltaic power generation output prediction result, fitting the error between the prediction result and an actual value by utilizing t distribution to obtain an error probability density function, and representing the possible occurrence range of the prediction error by using a confidence interval.
2. The EEMD-LSTM microgrid-based photovoltaic short-term power interval prediction method as recited in claim 1, wherein the method comprises the following steps: the XGboost algorithm is used for screening all variables and extracting characteristics of a photovoltaic power generation sample, the XGboost uses a decision tree as a bottom layer model, and the output is as shown in a formula (1):
Figure FDA0003771043120000011
secondly, the optimization goal of the XGboost model is to minimize the structural risk, and the objective function is as follows:
Figure FDA0003771043120000021
the first item of
Figure FDA0003771043120000022
Is the residual, the second term
Figure FDA0003771043120000023
Is a regular term.
The simplified objective function is to sum the loss function values of each sample to obtain the following equation:
Figure FDA0003771043120000024
by rewriting the above formula, we can rewrite the objective function as a unitary quadratic function about the leaf node coefficient w, and can find the optimal w and the minimum value of the objective function as:
Figure FDA0003771043120000025
Figure FDA0003771043120000026
Figure FDA0003771043120000027
Figure FDA0003771043120000028
and traversing the feature division points of all the features by using a greedy algorithm, searching a maximum gain node according to layers for expansion, and then adding a new decision tree in the model to automatically acquire the importance of the features for feature screening.
3. The EEMD-LSTM microgrid-based photovoltaic short-term power interval prediction method as recited in claim 1, wherein the method comprises the following steps: adding standard normal distribution white noise n into photovoltaic power generation output sequence data x (t) i (t) obtaining new sequence data x i (t)＝x(t)+n i (t), where i represents the number of white noise additions, i =1, 2.., M, new contribution sequence data x that will contain white noise i (t) EMD decomposition is carried out in sequence to obtain the representation forms of IMF component and residual component
Figure FDA0003771043120000029
Wherein J represents the number of IMF components, c (t) represents the IMF components, r (t) represents the remaining components, each IMF component c is calculated M times i，j (t) a statistical average of
Figure FDA0003771043120000031
As the jth IMF component of the original force sequence.
And evaluating the power generation output decomposition effect by using a signal-to-noise ratio (SNR) index. When the SNR value is larger, the ratio of the decomposed sequence to the noise sequence is large, and the decomposition effect is good; when the smaller the SNR value is, the smaller the ratio of the decomposed sequence to the noise sequence is, the decomposition effect is poor, and the SNR is defined as follows:
Figure FDA0003771043120000032
wherein S is j Representing the energy of the jth IMF component of the power generation output sequence, N representing the energy of noise, and N representing the length of the power generation output sequence;
and taking M when the SNR value is maximum as the EEMD decomposition times, and finally completing the mode decomposition of the original output sequence.
4. The EEMD-LSTM microgrid-based photovoltaic short-term power interval prediction method as recited in claim 1, wherein the method comprises the following steps: the structure of a first-order two-layer structure is adopted, and the specific parameters of the model are set as follows:
the training times are set to 1000 times, and the number of LSTM layer nodes is set to 50; the batch size is set to 72, the mean absolute error function is selected as the loss function, adam is selected as the optimization function, and tanh is selected as the activation function. To alleviate the overfitting phenomenon, dropout is set to 0.2 due to the large amount of data.
5. The EEMD-LSTM microgrid-based photovoltaic short-term power interval prediction method as recited in claim 1, wherein the method comprises the following steps: superposing and reducing the prediction result of the decomposition sequence to obtain a prediction result of the photovoltaic power generation output, and fitting the error between the prediction result and an actual value by utilizing t distribution to obtain an error probability density function;
the range in which prediction errors are likely to occur is represented by a confidence interval, and when the significance level is α and the confidence is 1- α, the confidence interval of the error distribution can be represented by (μ -Z) α/2 σ，μ+Z α/2 σ), where μ represents the mean value of the errors, σ represents the standard deviation of the errors, Z α/2 Is corresponding to standard score (z-score);
and evaluating the prediction result of the prediction interval by using the coverage rate and the width of the prediction interval, wherein the PICP represents the ratio of the true value between the upper boundary and the lower boundary of the prediction interval, and the formula is as follows:
Figure FDA0003771043120000041
wherein
Figure FDA0003771043120000042
n is the number of samples, L i ，U i The ith prediction lower and upper bound values are used.
PINAW represents the average distance between the upper boundary and the lower boundary of the prediction interval and is used for reflecting the accuracy of the prediction interval, and the calculation formula is as follows:
Figure FDA0003771043120000043
CN202210901508.3A 2022-07-28 2022-07-28 Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid Withdrawn CN115222138A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210901508.3A CN115222138A (en) 2022-07-28 2022-07-28 Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210901508.3A CN115222138A (en) 2022-07-28 2022-07-28 Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid

Publications (1)

Publication Number Publication Date
CN115222138A true CN115222138A (en) 2022-10-21

Family

ID=83613383

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210901508.3A Withdrawn CN115222138A (en) 2022-07-28 2022-07-28 Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid

Country Status (1)

Country Link
CN (1) CN115222138A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117113267A (en) * 2023-10-25 2023-11-24 杭州海兴泽科信息技术有限公司 Prediction model training method based on big data and photovoltaic power generation performance detection method
CN117374972A (en) * 2023-11-28 2024-01-09 广东电网有限责任公司 Transformer-based photovoltaic short-term power interval prediction method, device and medium

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117113267A (en) * 2023-10-25 2023-11-24 杭州海兴泽科信息技术有限公司 Prediction model training method based on big data and photovoltaic power generation performance detection method
CN117113267B (en) * 2023-10-25 2024-02-09 杭州海兴泽科信息技术有限公司 Prediction model training method based on big data and photovoltaic power generation performance detection method
CN117374972A (en) * 2023-11-28 2024-01-09 广东电网有限责任公司 Transformer-based photovoltaic short-term power interval prediction method, device and medium

Similar Documents

Publication Publication Date Title
CN112949945B (en) Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network
CN110610280A (en) Short-term prediction method, model, device and system for power load
CN115222138A (en) Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid
CN107992447B (en) Feature selection decomposition method applied to river water level prediction data
CN112801388B (en) Power load prediction method and system based on nonlinear time series algorithm
CN112434848B (en) Nonlinear weighted combination wind power prediction method based on deep belief network
CN113554466A (en) Short-term power consumption prediction model construction method, prediction method and device
CN113988477A (en) Photovoltaic power short-term prediction method and device based on machine learning and storage medium
CN116258269A (en) Ultra-short-term load dynamic prediction method based on load characteristic decomposition
CN115495991A (en) Rainfall interval prediction method based on time convolution network
CN115423163A (en) Method and device for predicting short-term flood events of drainage basin and terminal equipment
CN115907131B (en) Method and system for constructing electric heating load prediction model in northern area
CN117520784A (en) Groundwater level multi-step prediction method based on convolution attention long-short-term neural network
CN116169670A (en) Short-term non-resident load prediction method and system based on improved neural network
CN116739172A (en) Method and device for ultra-short-term prediction of offshore wind power based on climbing identification
CN118095891A (en) Active power distribution network payload prediction method and system considering source load meteorological characteristic decoupling
CN113094448B (en) Analysis method and analysis device for residence empty state and electronic equipment
CN117933531A (en) Distributed photovoltaic power generation power prediction system and method
CN109858667A (en) It is a kind of based on thunder and lightning weather to the short term clustering method of loading effects
CN117033923A (en) Method and system for predicting crime quantity based on interpretable machine learning
CN116561569A (en) Industrial power load identification method based on EO feature selection and AdaBoost algorithm
CN116167465A (en) Solar irradiance prediction method based on multivariate time series ensemble learning
CN110009132A (en) A kind of short-term electric load fining prediction technique based on LSTM deep neural network
CN115222137A (en) GWO-LSSVM-based microgrid load short-term interval prediction method
CN118596922B (en) Residential electric vehicle load analysis method based on Elman neural network algorithm

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20221021