CN114742298A - Ultra-short-term power prediction method for wind power generation - Google Patents

Ultra-short-term power prediction method for wind power generation Download PDF

Info

Publication number
CN114742298A
CN114742298A CN202210378788.4A CN202210378788A CN114742298A CN 114742298 A CN114742298 A CN 114742298A CN 202210378788 A CN202210378788 A CN 202210378788A CN 114742298 A CN114742298 A CN 114742298A
Authority
CN
China
Prior art keywords
data
wind
model
term
error
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210378788.4A
Other languages
Chinese (zh)
Inventor
宋哲
田海亭
严合群
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.)
Anhui Jianchi Intelligent Technology Co ltd
Original Assignee
Anhui Jianchi Intelligent Technology 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 Anhui Jianchi Intelligent Technology Co ltd filed Critical Anhui Jianchi Intelligent Technology Co ltd
Priority to CN202210378788.4A priority Critical patent/CN114742298A/en
Publication of CN114742298A publication Critical patent/CN114742298A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • 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
    • H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

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

Abstract

The invention discloses an ultra-short-term power prediction method for wind power generation, which relates to the technical field of wind power generation and comprises the following steps of firstly carrying out data splicing on cabin data files of each fan folder of a wind power plant in units of days so as to construct a complete wind speed time sequence with 30s as a time interval for two years, then carrying out repair processing on abnormal values and missing values of the time sequence, and then carrying out visual analysis on data. The method improves the capability of predicting the existing ultra-short-term wind power, and brings remarkable safety value and economic benefit.

Description

Ultra-short-term power prediction method for wind power generation
Technical Field
The invention relates to the technical field of wind power generation, in particular to an ultra-short-term power prediction method for wind power generation.
Background
In recent years, with the expansion of installed sites of onshore wind turbine generators, the wind turbine generators installed in areas with more sudden changes of weather are affected by meteorological changes more obviously, when the wind conditions suddenly change, the wind turbine generators installed in areas with more sudden changes of weather are affected by meteorological changes more obviously due to the hysteresis of a control system, even the wind turbine generators are turned over, so that the wind turbine generators installed in areas with more sudden changes of weather are affected by meteorological changes more obviously in recent years, when the wind conditions suddenly change, the wind turbine generators are easily affected by overload and even turned over due to the hysteresis of the control system, so that great economic loss is caused, meanwhile, the accuracy of the existing ultra-short-term wind power prediction is poor, so that the reference value of a wind power prediction system to power grid scheduling is not large, and a large amount of generated electricity evaluation plans can be generated by a main generator, the unit price of common laser radar and other wind speed measurement products is high, the influence of weather is large, batch application deployment is difficult to realize, and reliable foresight is difficult to achieve under a large time space scale.
Therefore, reliable ultra-short-term wind condition prediction is urgently needed, the ultra-short-term wind condition prediction is a worldwide problem, if wind speed and wind direction data of each unit in a short time in the future can be predicted through big data and an artificial intelligence technology, the control foresight of the wind generation units can be improved, the load safety of the wind generation units can be improved, meanwhile, the capability of the conventional ultra-short-term wind power prediction is improved, the remarkable safety value and economic benefit can be brought, and therefore an ultra-short-term power prediction method for wind power generation is urgently needed to change the current situation.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides an ultra-short-term power prediction method for wind power generation. The method has the advantages of improving the capability of predicting the existing ultra-short-term wind power and bringing remarkable safety value and economic benefit.
In order to achieve the purpose, the invention adopts the following technical scheme:
an ultra-short-term power prediction method for wind power generation comprises the following steps:
the method comprises the following steps: firstly, carrying out data splicing on cabin data files in units of days under each fan folder of a wind power plant, so as to construct a complete wind speed time sequence with 30s as a time interval for two years, and then repairing abnormal values and missing values of the time sequence;
step two: then, carrying out visual analysis on the data, wherein the visual analysis mainly comprises the visualization of the fluctuation and distribution condition of the wind speed and the wind direction of the cabin, the correlation analysis of the data variable of the cabin, the comparative analysis of the trend of the wind speed and the wind direction of the cabin and the wind speed and the wind direction of meteorological data and the analysis of the relation between the wind speed and the season of the cabin;
step three: then basic data preparation is carried out on a subsequently constructed model, and the part can briefly introduce the data which can be input by the model and divide the data into a training set and a verification set;
step four: establishing a unified neural network model aiming at all data of all fans of all wind fields, and when the neural network model is established, predicting wind speed data of each unit within 10 minutes in the future by using the frequency converter power grid side active power, the external temperature, the wind direction and the wind speed data of each unit in the last 1 hour, wherein the time resolution is 30 seconds, and the 30 seconds are taken as one moment to obtain a model equation;
step five: inputting corresponding data according to a model equation, wherein the dimensionality of each input data is (120,4), 22517519 data are totally obtained, the verification set comprises 22580 data, the training set comprises 22517439 data, and after model training is finished, the verification set data of a certain wind field and a certain fan are selected randomly to carry out model evaluation and draw and see a prediction result, so that the ultra-short term prediction of wind power generation is completed.
Preferably, the model equation is
Figure BDA0003591810170000031
(x1(t-1)，x2(t-1)，x3(t-1)，y(t-1))，
...，
(x1(t-119)，x2(t-119)，x3(t-119), y (t-119))), wherein
Figure BDA0003591810170000032
For the wind speed, x, to be predicted at 20 moments in the future1(t)，x2(t)，x3And (t), y (t) respectively represent the active power, the external temperature, the wind direction and the wind speed of the power grid side of the frequency converter at the time t, and the input data of the model are the active power, the external temperature, the wind direction and the wind speed of the power grid side of the frequency converter at 120 moments from the moment t-119 to the moment t. f (-) describes the internal relationship between the frequency converter grid side active power, the external temperature, the wind direction and the wind speed data of each unit in the last 1 hour and the wind speed data of the next 10 minutes.
Preferably, the neural network model is a long-short memory neural network LSTM, which is a variant of a recurrent neural network, the LSTM is specially designed to solve the long-term dependence problem of RNN, the LSTM is also the structure of RNN, but the repeated modules have a different structure, the key of the LSTM is the unit state, the horizontal line runs through the top of the graph, the unit state is just like a conveyor belt, the past information is directly sent to the next moment through the conveyor belt, and only a few minor linear interactions exist.
Preferably, there are a number of "gates" in the LSTM that can selectively pass information, and the first step of the LSTM is to determine which information needs to be discarded from the cell state, which is determined by a sigmoid layer, also called a forgetting gate, whose formula is
ft＝σ(Wf·[ht-1，xt]+bf)
Wherein, WfRepresents a weight matrix, [ h ]t-1，xt]Representing the splicing together of two matrices, bfRepresents a bias term, sigma is sigmoid function, ftIs the output of the forgetting gate.
Preferably, the next step of the LSTM is to decide which letters to sendInformation is stored in a cellular state, which comprises two parts: first there is a sigmoid layer called the entry gate, which will decide which information we will update, and next a tanh layer creates a vector of candidate values
Figure BDA0003591810170000041
It determines which information can be added to the cell state, the input gate being calculated as: i.e. it＝σ(Wi·[ht-1，xt]+bi)
Figure BDA0003591810170000042
We will now turn cell state Ct-1Is updated to Ct: first used cell state Ct-1Corresponding point times ftDiscarding the information that has been determined to be forgotten, and adding it
Figure BDA0003591810170000043
The final constituent cell state CtThe calculation formula is as follows:
Figure BDA0003591810170000044
where denotes the Hadamard product, i.e. the respective multiplication of the corresponding elements.
Preferably, the LSTM further includes an output gate, the output gate controls how much information is output to the external state at the current moment, a sigmoid layer is firstly run to determine which information of the cell state is to be output, then the cell state is passed through a tanh layer and then multiplied by the output of the sigmoid layer, and finally we can output the part we want to output, and the calculation formula of the output gate is: ot＝σ(Wo·[ht-1，xt]+bo)
ht＝ot*tanh(Ct)
In the formula: hidden state h in recurrent neural networkstThe history information is stored and can be regarded as a memory, in the simple recurrent neural network, the hidden state is rewritten at each moment, so that the simple recurrent neural network can be regarded as a short-term memory, the long-term memory can be regarded as a network parameter, the experience learned from training data is hidden, the updating period is far slower than that of the short-term memory, and in the LSTM network, the memory unit CtA memory unit C capable of capturing a key message at a certain moment and capable of storing the key message for a certain time intervaltThe life cycle of the stored information is longer than that of short-term memory htBut much shorter than long-term memory and is therefore called long-short term memory.
Preferably, the model evaluation comprises a mean square error evaluation, a mean absolute error evaluation, a variance evaluation of absolute errors, a mean absolute percentage error evaluation, and a variance evaluation of absolute percentage errors.
Preferably, the mean square error estimation calculation formula is:
Figure BDA0003591810170000061
the mean square error is the most common regression loss function, also called L2 loss, which means the mean value of the sum of squares of the distance between the predicted value and the true value, and the smaller the mean square error is, the better the model prediction effect is; the average absolute error evaluation calculation formula is as follows:
Figure BDA0003591810170000062
the average absolute error is another loss function used for a regression model, which is also called L1 loss, and means that the sum of absolute values of differences between a predicted value and a true value is taken as an average value, and the smaller the average absolute error is, the better the model prediction effect is; in the formula: m is the number of samples, yiIn order to be the true value of the value,
Figure BDA0003591810170000063
is a predicted value.
Preferably, the variance estimation of the absolute error is calculated by the following formula:
Figure BDA0003591810170000064
the variance of the absolute error is the variance of the data sequence describing the absolute value of the difference between the predicted value and the true value, wherein: m is the number of samples, yiIn order to be the true value of the value,
Figure BDA0003591810170000065
is a predicted value.
Preferably, the average absolute percentage error evaluation calculation formula is:
Figure BDA0003591810170000066
the average absolute percentage error not only considers the error between the predicted value and the true value, but also considers the ratio of the error to the true value, that is, the process of comparing the average absolute percentage error with the original data exists, it can be seen that the map and the mae are similar in calculation, only one denominator is added in the calculation of the map, the smaller the value of the map is, the better the model is represented, the perfect model is represented when the map is 0, and the inferior model is represented when the map is greater than 1; the variance evaluation calculation formula of the absolute percentage error is as follows:
Figure BDA0003591810170000071
the variance of the absolute percentage error is the variance of the data sequence describing the absolute percentage error between the predicted value and the true value, wherein: m is the number of samples, yiIn order to be the true value of the value,
Figure BDA0003591810170000072
is a predicted value.
The invention has the beneficial effects that: according to the method, cabin data are processed, a training set and a verification set are divided, then a neural network model is established, model training is carried out on input data, after model training is finished, data of the verification set of a certain wind field and a certain fan are selected randomly to draw and see a prediction result, the prediction error of a wind power prediction model is smaller, the accuracy is higher, the prediction error is closer to the actual situation, ultra-short-term prediction of wind power can be better carried out, powerful reference is provided for power scheduling of the wind field, the control foresight of a wind turbine generator is improved, the load safety of the wind turbine generator is improved, meanwhile, the capability of the existing ultra-short-term wind power prediction is improved, and remarkable safety value and economic benefit are brought.
Drawings
FIG. 1 is a schematic diagram of a prediction flow structure of an ultra-short-term power prediction method for wind power generation according to the present invention;
FIG. 2 is a schematic diagram of a data preparation process of the ultra-short-term power prediction method for wind power generation according to the present invention;
FIG. 3 is a schematic structural diagram of a repetitive module in an LSTM of the ultra-short-term power prediction method for wind power generation according to the present invention;
FIG. 4 is a schematic diagram of the state change structure of an LSTM module of the ultra-short-term power prediction method for wind power generation according to the present invention;
FIG. 5 is a schematic structural diagram of an LSTM module forgetting gate of the ultra-short-term power prediction method for wind power generation according to the present invention;
FIG. 6 is a schematic structural diagram of an LSTM module input gate of the ultra-short-term power prediction method for wind power generation according to the present invention;
fig. 7 is a schematic structural diagram of an output gate of an LSTM module of the ultrashort-term power prediction method for wind power generation according to the present invention.
Detailed Description
The technical solution of the present patent will be described in further detail with reference to the following embodiments.
Referring to fig. 1-2, a method for ultra-short term power prediction for wind power generation includes the following steps:
the method comprises the following steps: firstly, carrying out data splicing on cabin data files in units of days under each fan folder of a wind power plant, so as to construct a complete wind speed time sequence with 30s as a time interval for two years, and then repairing abnormal values and missing values of the time sequence;
step two: then, carrying out visual analysis on the data, wherein the visual analysis mainly comprises the visualization of the fluctuation and distribution condition of the wind speed and the wind direction of the cabin, the correlation analysis of the data variable of the cabin, the comparative analysis of the trend of the wind speed and the wind direction of the cabin and the wind speed and the wind direction of meteorological data and the analysis of the relation between the wind speed and the season of the cabin;
step three: then basic data preparation is carried out on a subsequently constructed model, and the part can briefly introduce the data which can be input by the model and divide the data into a training set and a verification set;
step four: establishing a unified neural network model aiming at all data of all fans of all wind fields, and when the neural network model is established, predicting wind speed data of each unit within 10 minutes in the future by using the frequency converter power grid side active power, the external temperature, the wind direction and the wind speed data of each unit in the last 1 hour, wherein the time resolution is 30 seconds, and the 30 seconds are taken as one moment to obtain a model equation;
step five: inputting corresponding data according to a model equation, wherein the dimensionality of each input data is (120,4), 22517519 data are totally obtained, the verification set comprises 22580 data, the training set comprises 22517439 data, and after model training is finished, the verification set data of a certain wind field and a certain fan are selected randomly to carry out model evaluation and draw and see a prediction result, so that the ultra-short term prediction of wind power generation is completed.
In this embodiment, the model equation is
Figure BDA0003591810170000091
(x1(t-1)，x2(t-1)，x3(t-1)，y(t-1))，
...，
(x1(t-119)，x2(t-119)，x3(t-119), y (t-119))), wherein
Figure BDA0003591810170000092
For the wind speed, x, to be predicted at 20 moments in the future1(t)，x2(t)，x3And (t), y and t) respectively represent the active power, the external temperature, the wind direction and the wind speed of the power grid side of the frequency converter at the moment t, and the input data of the model comprise the active power, the external temperature, the wind direction and the wind speed of the power grid side of the frequency converter at 120 moments from the moment t-119 to the moment t. f (-) describes the frequency converter for the last 1 hour of each unitThe grid side active power, the outside temperature, the wind direction and the wind speed data of the future 10 minutes.
Referring to fig. 3 and 4, the neural network model is a long-short memory neural network LSTM, which is a variant of the recurrent neural network, and is specifically designed to solve the long-term dependence problem of RNN, and the LSTM is also the structure of RNN, but the repeated modules have a different structure, and the key of the LSTM is the unit state, the horizontal line runs through the top of the graph, the unit state is like a conveyor belt, and the past information is directly sent to the next moment through the conveyor belt, and only has small linear interaction.
Referring to fig. 5, there are a number of "gates" in the LSTM that can selectively pass information, and the first step of the LSTM is to determine which information needs to be discarded from the cell state, which is determined by a sigmoid layer, also called a forgetting gate, whose formula is
ft＝σ(Wf·[ht-1，xt]+bf)
Wherein, WfRepresents a weight matrix, [ h ]t-1，xt]Representing the splicing together of two matrices, bfRepresents a bias term, sigma is sigmoid function, ftIs the output of the forgetting gate.
Referring to fig. 6, the next step of LSTM is to decide which information to store in the cellular state, which consists of two parts: first there is a sigmoid layer called the entry gate, which will decide which information we will update, and next a tanh layer creates a vector of candidate values
Figure BDA0003591810170000101
It determines which information can be added to the cell state, the input gate is calculated as:
it＝σ(Wi·[ht-1，xt]+bi)
Figure BDA0003591810170000111
we will now turn cell state Ct-1Is updated to Ct: first used cell state Ct-1Corresponding point times ftDiscarding the information that has been determined to be forgotten, and adding it
Figure BDA0003591810170000112
Final constituent cell state CtThe calculation formula is as follows:
Figure BDA0003591810170000113
where denotes the Hadamard product, i.e. the respective multiplication of the corresponding elements.
Referring to fig. 7, the LSTM further includes an output gate, the output gate controls how much information is output to the external state at the current time, a sigmoid layer is first run to determine which information of the cell state is to be output, then the cell state is passed through a tanh layer and then multiplied by the output of the sigmoid layer, and finally we can output the portion we want to output, and the calculation formula of the output gate is: ot＝σ(Wo·[ht-1，xt]+bo)
ht＝o1*tanh(Ct)
In the formula: hidden state h in recurrent neural networkstThe history information is stored and can be regarded as a memory, in the simple recurrent neural network, the hidden state is rewritten at every moment, so that the simple recurrent neural network can be regarded as a short-term memory, the long-term memory can be regarded as a network parameter, the experience learned from the training data is hidden, the updating period is far slower than that of the short-term memory, and in the LSTM network, the memory unit CtA memory unit C capable of capturing a key message at a certain moment and capable of storing the key message for a certain time intervaltThe life cycle of the stored information is longer than that of short-term memory htBut much shorter than long-term memory and is therefore called long-short term memory.
In the present embodiment, the model evaluation includes a mean square error evaluation, a mean absolute error evaluation, a variance evaluation of absolute errors, a mean absolute percentage error evaluation, and a variance evaluation of absolute percentage errors.
In this embodiment, the mean square error estimation calculation formula is:
Figure BDA0003591810170000121
the mean square error is the most common regression loss function, also called L2 loss, which means the mean value of the sum of squares of the distance between the predicted value and the true value, and the smaller the mean square error is, the better the model prediction effect is; the mean absolute error evaluation calculation formula is:
Figure BDA0003591810170000122
the average absolute error is another loss function used for a regression model, which is also called L1 loss, and means that the sum of absolute values of differences between a predicted value and a true value is taken as an average value, and the smaller the average absolute error is, the better the model prediction effect is; in the formula: m is the number of samples, yiIn order to be the true value of the value,
Figure BDA0003591810170000123
is a predicted value.
In the present embodiment, the variance estimation calculation formula of the absolute error is:
Figure BDA0003591810170000124
the variance of the absolute error is the variance of the data sequence describing the absolute value of the difference between the predicted value and the true value, wherein: m is the number of samples, yiIn order to be the true value of the value,
Figure BDA0003591810170000131
is a predicted value.
In this embodiment, the average absolute percentage error estimation calculation formula is:
Figure BDA0003591810170000132
the average absolute percentage error not only takes into account the error between the predicted value and the true value, but also takes into account the ratio of the error to the true value, i.e.The process of comparing the original data with the original data exists, and it can be seen that mape and mae are similar in calculation, but a denominator is added in the calculation of mape, the smaller the value of mape is, the better the model is, mape is 0 and represents a perfect model, and mape is greater than 1 and represents a poor model; the variance estimation of the absolute percentage error is calculated as:
Figure BDA0003591810170000133
the variance of the absolute percentage error is the variance of the data sequence describing the absolute percentage error between the predicted value and the true value, wherein: m is the number of samples, yiIn order to be the true value of the value,
Figure BDA0003591810170000134
is a predicted value.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.

Claims (10)

1. An ultra-short-term power prediction method for wind power generation is characterized by comprising the following steps:
the method comprises the following steps: firstly, carrying out data splicing on cabin data files in units of days under each fan folder of a wind power plant, so as to construct a complete wind speed time sequence with 30s as a time interval for two years, and then repairing abnormal values and missing values of the time sequence;
step two: then, carrying out visual analysis on the data, wherein the visual analysis mainly comprises the visualization of the fluctuation and distribution condition of the wind speed and the wind direction of the engine room, the correlation analysis of the variable of the data of the engine room, the comparative analysis of the trend of the wind speed and the wind direction of the engine room and the wind speed and the wind direction of meteorological data and the analysis of the relation between the wind speed and the season of the engine room;
step three: then basic data preparation is carried out on a subsequently constructed model, and the part can briefly introduce the data which can be input by the model and divide the data into a training set and a verification set;
step four: establishing a unified neural network model aiming at all data of all fans of all wind fields, and when the neural network model is established, predicting wind speed data of each unit within 10 minutes in the future by using the frequency converter power grid side active power, the external temperature, the wind direction and the wind speed data of each unit in the last 1 hour, wherein the time resolution is 30 seconds, and the 30 seconds are taken as one moment to obtain a model equation;
step five: inputting corresponding data according to a model equation, wherein the dimensionality of each input data is (120,4), 22517519 data are totally obtained, the verification set comprises 22580 data, the training set comprises 22517439 data, and after model training is finished, the verification set data of a certain wind field and a certain fan are selected randomly to carry out model evaluation and draw and see a prediction result, so that the ultra-short term prediction of wind power generation is completed.
2. The ultra-short term power prediction method for wind power generation as claimed in claim 1, wherein the model equation is
Figure FDA0003591810160000021
(x1(t-1)，x2(t-1)，x3(t-1)，y(t-1))，
…，
(x1(t-119)，x2(t-119)，x3(t-119), y (t-119))), wherein
Figure FDA0003591810160000022
For the wind speed, x, to be predicted at 20 moments in the future1(t)，x2(t)，x3(t), y (t) respectively represent the frequency converter power grid side active power, the external temperature, the wind direction and the wind speed at the moment t, the input data of the model are the frequency converter power grid side active power at 120 moments from the moment t-119 to the moment t, the external temperature, the wind direction and the wind speed, and f (·) describes the frequency converter power grid side active power, the external temperature, the wind direction and the wind speed data and the frequency converter power grid side active power of each unit in the last 1 hourInternal relationship between wind speed data for the next 10 minutes.
3. The ultrashort term power prediction method for wind power generation as claimed in claim 1, wherein the neural network model is long-short term memory neural network LSTM, which is a variant of recurrent neural network, LSTM is specially designed to solve the long term dependence problem of RNN, LSTM is also the structure of RNN, but the repeated modules have a different structure, the key of LSTM is unit state, the horizontal line runs through the top of the graph, the unit state is just like a conveyor belt, the past information is directly sent to the next time through the conveyor belt, and there are only small linear interactions.
4. The ultra-short term power prediction method for wind power generation as claimed in claim 3, wherein there are a plurality of "gates" in the LSTM, which can selectively let information pass, the first step of LSTM is to decide which information needs to be discarded from the unit state, which is determined by a sigmoid layer also called forgetting gate, the formula of which is that
ft＝σ(Wf·[ht-1，xt]+bf)
Wherein, WfRepresents a weight matrix, [ h ]t-1，xt]Representing the splicing together of two matrices, bfRepresents a bias term, sigma is sigmoid function, ftIs the output of the forgetting gate.
5. The ultrashort term power prediction method for wind power generation as claimed in claim 4, wherein the next step of the LSTM is to decide which information to store in the cell state, which includes two parts: first there is a sigmoid layer called the entry gate, which will decide which information we will update, and next a tanh layer creates a vector of candidate values
Figure FDA0003591810160000031
It decides which information to useCan be added to the cell state, the input gate being calculated as: i.e. it＝σ(Wi·[ht-1，xt]+bi)
Figure FDA0003591810160000032
We will now turn the cell state Ct-1Is updated to Ct: first used cell state Ct-1Corresponding point times ftDiscarding the information that has been determined to be forgotten, and adding it
Figure FDA0003591810160000033
The final constituent cell state CtThe calculation formula is as follows:
Figure FDA0003591810160000041
where denotes the Hadamard product, i.e. the respective multiplication of the corresponding elements.
6. The ultrashort-term power prediction method for wind power generation as claimed in claim 4, wherein the LSTM further comprises an output gate, the output gate controls how much information is output to the external state at the current moment, a sigmoid layer is firstly operated to determine which information of the cell state is output, then the cell state passes through a tanh layer and then is multiplied by the output of the sigmoid layer, and finally, the output gate outputs the portion which we want to output, and the calculation formula of the output gate is as follows: ot＝σ(Wo·[ht-1，xt]+bo)
ht＝ot*tanh(Ct)
In the formula: hidden state h in recurrent neural networkstThe history information is stored, and the method can be regarded as a memory, and in a simple recurrent neural network, the hidden state can be rewritten at each moment, so that the method can be regarded as a short-term memoryMemory, long-term memory can be considered as a network parameter, and the experience learned from training data is implicit, the update period of the long-term memory is far slower than that of short-term memory, and in the LSTM network, the memory cell CtA memory unit C capable of capturing a key message at a certain moment and capable of storing the key message for a certain time intervaltThe life cycle of the stored information is longer than that of short-term memory htBut much shorter than long-term memory and is therefore referred to as long-short term memory.
7. The ultra-short term power prediction method for wind power generation as claimed in claim 1, wherein the model estimation comprises mean square error estimation, mean absolute error estimation, variance estimation of absolute error, mean absolute percentage error estimation and variance estimation of absolute percentage error.
8. The ultra-short term power prediction method for wind power generation as claimed in claim 7, wherein the mean square error estimation calculation formula is:
Figure FDA0003591810160000051
the mean square error is the most common regression loss function, also called L2 loss, which means the mean value of the sum of squares of the distance between the predicted value and the true value, and the smaller the mean square error is, the better the model prediction effect is; the average absolute error evaluation calculation formula is as follows:
Figure FDA0003591810160000052
the average absolute error is another loss function used for a regression model, which is also called L1 loss, and means that the sum of absolute values of differences between a predicted value and a true value is taken as an average value, and the smaller the average absolute error is, the better the model prediction effect is; in the formula: m is the number of samples, yiIn order to be the true value of the value,
Figure FDA0003591810160000053
is a predicted value.
9. The ultrashort-term power prediction method for wind power generation as claimed in claim 8, wherein the variance estimation of the absolute error is calculated by the formula:
Figure FDA0003591810160000054
the variance of the absolute error is the variance of the data sequence describing the absolute value of the difference between the predicted value and the true value, wherein: m is the number of samples, yiIn order to be the true value of the value,
Figure FDA0003591810160000055
is a predicted value.
10. The ultra-short term power prediction method for wind power generation as claimed in claim 9, wherein the average absolute percentage error estimation calculation formula is:
Figure FDA0003591810160000061
the average absolute percentage error not only considers the error between the predicted value and the true value, but also considers the ratio of the error to the true value, that is, the process of comparing the average absolute percentage error with the original data exists, it can be seen that the map and the mae are similar in calculation, only one denominator is added in the calculation of the map, the smaller the value of the map is, the better the model is represented, the perfect model is represented when the map is 0, and the inferior model is represented when the map is greater than 1; the variance evaluation calculation formula of the absolute percentage error is as follows:
Figure FDA0003591810160000062
the variance of the absolute percentage error is the variance of the data sequence describing the absolute percentage error between the predicted value and the true value, wherein: m is the number of samples, yiIn order to be the true value of the value,
Figure FDA0003591810160000063
is a predicted value.
CN202210378788.4A 2022-04-12 2022-04-12 Ultra-short-term power prediction method for wind power generation Pending CN114742298A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210378788.4A CN114742298A (en) 2022-04-12 2022-04-12 Ultra-short-term power prediction method for wind power generation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210378788.4A CN114742298A (en) 2022-04-12 2022-04-12 Ultra-short-term power prediction method for wind power generation

Publications (1)

Publication Number Publication Date
CN114742298A true CN114742298A (en) 2022-07-12

Family

ID=82281526

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210378788.4A Pending CN114742298A (en) 2022-04-12 2022-04-12 Ultra-short-term power prediction method for wind power generation

Country Status (1)

Country Link
CN (1) CN114742298A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117312726A (en) * 2023-11-29 2023-12-29 中国船舶集团有限公司第七〇七研究所 Capability assessment method during anchoring auxiliary power positioning operation

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117312726A (en) * 2023-11-29 2023-12-29 中国船舶集团有限公司第七〇七研究所 Capability assessment method during anchoring auxiliary power positioning operation
CN117312726B (en) * 2023-11-29 2024-02-06 中国船舶集团有限公司第七〇七研究所 Capability assessment method during anchoring auxiliary power positioning operation

Similar Documents

Publication Publication Date Title
Wang et al. Research and application of a hybrid forecasting framework based on multi-objective optimization for electrical power system
CN108280551B (en) Photovoltaic power generation power prediction method utilizing long-term and short-term memory network
Qu et al. A temporal distributed hybrid deep learning model for day-ahead distributed PV power forecasting
Abuella et al. Solar power forecasting using artificial neural networks
Eom et al. Feature-selective ensemble learning-based long-term regional PV generation forecasting
Wu et al. Stratification-based wind power forecasting in a high-penetration wind power system using a hybrid model
CN110942205A (en) Short-term photovoltaic power generation power prediction method based on HIMVO-SVM
CN116307291B (en) Distributed photovoltaic power generation prediction method and prediction terminal based on wavelet decomposition
CN115275991A (en) Active power distribution network operation situation prediction method based on IEMD-TA-LSTM model
CN115860177A (en) Photovoltaic power generation power prediction method based on combined machine learning model and application thereof
Castellanos et al. Average hourly wind speed forecasting with ANFIS
CN114742298A (en) Ultra-short-term power prediction method for wind power generation
Quang et al. Applying Artificial Intelligence in Forecasting the Output of Industrial Solar Power Plant in Vietnam
Souabi et al. Data-driven prediction models of photovoltaic energy for smart grid applications
Mosaico et al. A hybrid methodology for the day-ahead PV forecasting exploiting a Clear Sky Model or Artificial Neural Networks
Xu et al. Short-term wind speed prediction based on GRU
Phan et al. Application of a new Transformer-based model and XGBoost to improve one-day-ahead solar power forecasts
Sharika et al. Long-term solar irradiance forecasting approaches-a comparative study
CN113111592A (en) Short-term wind power prediction method based on EMD-LSTM
Prema et al. LSTM based Deep Learning model for accurate wind speed prediction
Sun et al. Short-term photovoltaic power prediction modeling based on AdaBoost algorithm and Elman
Zuo Integrated forecasting models based on LSTM and TCN for short-term electricity load forecasting
Bantupalli et al. Wind Speed forecasting using empirical mode decomposition with ANN and ARIMA models
CN115660893A (en) Transformer substation bus load prediction method based on load characteristics
Ye et al. A comparative study of ensemble support vector regression methods for short-term load forecasting

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