CN116488151A - Short-term wind power prediction method based on condition generation countermeasure network - Google Patents

Short-term wind power prediction method based on condition generation countermeasure network Download PDF

Info

Publication number
CN116488151A
CN116488151A CN202310387839.4A CN202310387839A CN116488151A CN 116488151 A CN116488151 A CN 116488151A CN 202310387839 A CN202310387839 A CN 202310387839A CN 116488151 A CN116488151 A CN 116488151A
Authority
CN
China
Prior art keywords
data
wind power
short
lstm
cgan
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
CN202310387839.4A
Other languages
Chinese (zh)
Inventor
陈汝斯
刘海光
杨旋
李大虎
蔡德福
周悦
周鲲鹏
孙冠群
王尔玺
王文娜
张良一
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.)
State Grid Corp of China SGCC
State Grid Hubei Electric Power Co Ltd
Electric Power Research Institute of State Grid Hubei Electric Power Co Ltd
Original Assignee
State Grid Corp of China SGCC
State Grid Hubei Electric Power Co Ltd
Electric Power Research Institute of State Grid Hubei 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 State Grid Corp of China SGCC, State Grid Hubei Electric Power Co Ltd, Electric Power Research Institute of State Grid Hubei Electric Power Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN202310387839.4A priority Critical patent/CN116488151A/en
Publication of CN116488151A publication Critical patent/CN116488151A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • 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
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/09—Supervised learning
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
    • G06Q10/06375—Prediction of business process outcome or impact based on a proposed change
    • 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
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/381—Dispersed generators
    • 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
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/46—Controlling of the sharing of output between the generators, converters, or transformers
    • H02J3/466—Scheduling the operation of the generators, e.g. connecting or disconnecting generators to meet a given 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]
    • 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
    • H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
    • H02J2300/20—The dispersed energy generation being of renewable origin
    • H02J2300/28—The renewable source being wind energy
    • 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
    • H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
    • H02J2300/40—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation wherein a plurality of decentralised, dispersed or local energy generation technologies are operated simultaneously
    • 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)
  • Human Resources & Organizations (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Power Engineering (AREA)
  • Artificial Intelligence (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • Game Theory and Decision Science (AREA)
  • Water Supply & Treatment (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Wind Motors (AREA)
  • Control Of Eletrric Generators (AREA)

Abstract

The invention relates to a short-term wind power prediction method based on a generation countermeasure network, which comprises the following steps: extracting a meteorological variable fluctuation time sequence and a historical wind power time sequence generating feature matrix from a short-term meteorological scale to serve as condition data, and establishing a short-term wind power prediction model based on a CNN-LSTM-CGAN improved condition generating countermeasure network; inputting the condition data and the random noise data into a generator together to obtain the generated prediction data; inputting the generated prediction data and the condition data into a discriminator together, and discriminating with the real power data; introducing a characteristic loss function, measuring the characteristic value deviation of the generated predicted data and the real data, and updating the optimized network parameter weight; if the maximum iteration number is reached, the iteration is terminated, and network parameters are output; and predicting by using the trained network model to obtain a final wind power prediction result. Compared with the prior art, the invention has the advantages of improving generalization of the pretension test model, and the like.

Description

Short-term wind power prediction method based on condition generation countermeasure network
Technical Field
The invention relates to the technical field of accurate short-term wind power prediction, in particular to a short-term wind power prediction method based on a condition generation countermeasure network.
Background
Under the development trend of constructing a novel power system taking new energy as a main body, the wind power grid-connected proportion is continuously increased. Because wind power has stronger volatility and uncertainty, great challenges are brought to planning, scheduling and running of a power grid. When the wind power is low in occupied area, the real-time balance between power generation and power consumption can be well realized by only adjusting the load of the conventional power supply output tracking change, the accurate short-term wind power prediction technology can predict the wind power fluctuation time sequence in the short time in the future, a basis is provided for the power grid to formulate a reasonable scheduling plan, the air discarding quantity is reduced, and the method has important significance for guaranteeing the safe and stable operation of the power grid and improving the economic benefit.
After the wind power plant is integrated into a power grid on a large scale, the conventional power supply and wind power are required to meet load demands together, and under the condition that wind power output is unknown, the conventional power supply is required to reserve a large amount of rotation reserve capacity to cope with unknown wind power fluctuation and load demands, so that wind power consumption space is greatly occupied, and the influence on safe and stable operation of a power system is brought. And because wind speed and wind direction change cause wind turbine generator system generation to form great randomness, volatility and instability, a lot of difficulties are added to the quick, effective and safe dispatching of the power grid. How to fully learn the hidden deep relation in the nonlinear power data, improve the precision of the prediction result, make a reasonable scheduling plan, and coordinate and solve the main difficulty of the prediction of the power grid frequency modulation. The prior art mainly adopts the following methods: according to the short-term wind power prediction method based on the BP neural network, the BP neural network can approach any nonlinear mapping with arbitrary precision, learn and adapt to unknown information, and certain fault tolerance is achieved, but the BP neural network is easy to fall into a local minimum value, and a global optimal solution cannot be obtained; the ultra-short-term wind power prediction model based on the LSTM can give consideration to the data time sequence and the nonlinear relation, the accuracy of a prediction result under the normal condition is ideal, but the deep relation of the data sample with higher complexity is difficult to mine; the traditional generation countermeasure network adopts a fully connected neural network as a generator and a discriminator, is only suitable for a relatively simple data set, and the learning process of GAN is too free, so that the training process and the learning result are unstable.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a short-term wind power prediction method based on a condition generation countermeasure network.
The aim of the invention can be achieved by the following technical scheme:
a short-term wind power prediction method based on a condition generation countermeasure network comprises the following steps:
1) Selecting a historical wind power data time sequence and a historical meteorological data time sequence of a wind power plant power station, and cleaning the data;
2) Screening wind power prediction influence factors and judging importance degrees from the cleaned data in the step 1) by using a random forest method to obtain weights of different influence factors, removing factors with lower influence on prediction, generating a feature matrix as condition data, and establishing and training a short-term wind power prediction model based on CNN-LSTM-CGAN improved conditions to generate an countermeasure network;
3) Initializing a weight of a CNN-LSTM-CGAN improved condition generation countermeasure network, and setting the maximum iteration times;
4) Utilizing a CNN construction generator to input the condition data and the random noise data obtained in the step 2) into four layers of one-dimensional CNN networks in the generator at the same time to generate power data under the constraint of the condition data;
5) Constructing a discriminator by utilizing LSTM, introducing the power data obtained in the step 4) into an LSTM network together with real data and condition data to perform countermeasure training, and obtaining a preliminary prediction result close to real power data distribution through a full connection layer under the constraint of the condition data;
6) Introducing a characteristic loss function, calculating a CNN-LSTM-CGAN improved condition to generate a prediction error of the countermeasure network based on the preliminary prediction result obtained in the step 5), and optimizing parameter weights;
7) Judging the iteration times of the step 6), if the iteration times reach the maximum iteration times set in the step 3), ending the iteration, outputting CNN-LSTM-CGAN improved condition to generate parameters of the countermeasure network, otherwise, enabling the current iteration times to be +1, and executing the step 4);
8) And predicting by using the trained short-term wind power prediction model, and finally obtaining a wind power prediction result.
Further, in the step 2), the importance degree of the weather data of the wind speed, the temperature, the air pressure and the humidity with different heights of 10m, 30m and 70m on the influence of the historical wind power fluctuation is calculated by using a random forest method, the feature selection is performed, and a feature matrix is obtained as the condition data, and the specific steps are as follows:
21 Randomly extracting K new self-service book sets from the original training data set by applying a bootstrap method, and constructing K classification regression trees, wherein K OOB sets are formed by the books which are not extracted each time;
22 Randomly decimating m at each node of each tree ny Each feature is used as a random feature subset, the information quantity of each feature in the feature subset is calculated, and the information quantity is m ny Selecting one of the features with the most classification capability to perform node splitting;
23 Score VIM with Gini index j (Gini) The calculation formula for calculating the feature importance degree, gini index is:
wherein: k represents a total of K categories; p (P) mk Representing the proportion of the class k in the node m;
feature X j The significance of the node m, namely the Gini index change amount calculation formula before and after the node m branches, is as follows:
wherein: GI (GI) l 、GI r Respectively representing Gini indexes of two new nodes after branching.
Further, step 3) initializes weights of CNN-LSTM-CGAN improved condition generation countermeasure network, and the specific steps include:
31 Gaussian initializing the convolution layer, sampling from the gaussian distribution with the mean value of 0 and the variance of 1 as an initial weight;
32 A zero_state function is called to realize LSTM composite network initialization.
Further, the objective function of the short-term wind power prediction model based on the CNN-LSTM-CGAN improved condition generation countermeasure network is as follows:
L G ＝-E z,y (D(G(z|y)|y))
L D ＝-E x,y (D(x|y))+E z,y (D(G(z|y)|y))
min G max D L CGAN ＝E x,y (lnD(x|y))
+E z,y (ln(1-D(G(z|y)|y)))
wherein: g represents a generator, D represents a neural network, L G Loss function of generator, L D Is the loss function of the neural network, x is the real power data, y is the condition data, z is the noise data, L CGAN Is a loss function of CNN-LSTM-CGAN, E x,y (. Cndot.) represents the expected value for the x, y distribution, E z,y (. Cndot.) represents the expected value for the z, y distribution; g (-) and D (-) are data output by the generator and the neural network, respectively.
Further, in step 6), the method further includes improving the feature loss function, and specifically includes the steps of:
61 Combining with the traditional loss functions L1 and L2 norms, the generating result of the generator is more similar to the real data distribution, the generating result obtained by using the L1 norms as the loss functions by the CNN-LSTM-CGAN is more accurate, and the redefined L1 norms are as follows:
L L1 ＝E x,y,z (||x-G(z|y)||)
wherein: x is real power data, y is conditional data, z is noise data，E x,y,z (. Cndot.) represents the expected value for the x, y, z distribution, G (z|y) is the generated power data;
62 Utilizing the characteristic value deviation between the power data and the real data generated by the hidden layer measurement generator of the LSTM neural network to further improve the accuracy of the CNN-LSTM-CGAN for predicting the ultra-short-term wind power and the characteristic loss function L per The method comprises the following steps:
L per (G,x)＝E x,y,z (λ i P i (G,x))
wherein: p (P) i (G, x) the generator generated prediction data and real power data are in the hidden layer h i Wherein h is the average deviation value of i Hiding a layer for an ith layer of the neural network; h is a i (x) Is h i True power data values at the hidden layer; h is a i (G) To be in hidden layer h i A predicted data value generated by the generator; c (C) i 、H i And W is i Respectively is a hidden layer h i Channel number, height and width values; lambda (lambda) i Is a weight parameter;
63 To L1 norm loss function L and characteristic loss function L per After combination, the total loss function of the short-term wind power prediction model is as follows:
L(G,D)＝min G max D L CGAN +λ L1 L L1 +λ per L per (G,x)
wherein: lambda (lambda) L1 And lambda (lambda) per Calculating a difference value between the preliminary predicted value and the true value by using the loss function as a weight parameter of the loss function;
64 After the difference value is obtained, the model is reversely transmitted to update each parameter to reduce the loss between the real value and the predicted value, so that the predicted value generated by the model is close to the real value, and the learning purpose is achieved;
65 The whole CNN-LSTM-CGAN short-term wind power prediction model is continuously subjected to iterative optimization until the maximum iterative times are reached, so that the model training optimal parameters are obtained.
Furthermore, in the short-term wind power prediction model training process, a generator and neural network alternate training mode is adopted, the neural network is updated once and then the generator is updated, the generator and the neural network are optimized by using an Adam optimizer, and a random gradient descent algorithm is adopted to accelerate model convergence.
Compared with the prior art, the short-term wind power prediction method based on the condition generation countermeasure network provided by the invention has the following beneficial effects:
1. according to the method, historical power data and meteorological data are taken as condition variables to be input into a generator together with noise, prediction data and real data are input into a neural network, and a characteristic loss function is introduced through CNN-LSTM-CGAN game training, so that the prediction accuracy of short-term wind power is further improved;
2. the invention provides a random forest method for describing the relevance between influencing factors and wind power, introduces a convolutional neural network to process nonlinear data with complex characteristic quantity, adopts a step convolutional network to sample the network in an autonomous space, and builds a generator model by 4 layers of CNNs, so that model training is easy to achieve balance, the difference between the generated data and expected values is small, and the model convergence speed is high.
Drawings
FIG. 1 is a flow chart of a short-term wind power prediction method based on a condition generation countermeasure network according to an embodiment of the present invention;
fig. 2 is a CGAN block diagram of an embodiment of the present invention;
FIG. 3 is a short-term wind power prediction model based on CGAN according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a random forest feature selection method according to an embodiment of the present invention.
Detailed Description
The invention will now be described in detail with reference to the drawings and specific examples. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
The invention relates to a short-term wind power prediction method based on a condition generation countermeasure network, which considers the macroscopic fluctuation characteristic of a short-term wind power output power time sequence to highlight the change characteristic of the short-term wind power time sequence and establishes a CNN-LSTM-CGAN-based short-term wind power prediction model.
The main principle of the short-term wind power prediction model based on the improved CNN-LSTM-CGAN established by the invention is as follows:
in the aspect of feature processing, because wind power data presents complex nonlinearity, and input features for prediction, including wind speeds at different heights, temperatures, weather fluctuation variables and the like are relatively independent feature time sequences, the traditional generator is only suitable for processing a simple data set and free to learn, and therefore internal connection between each feature time sequence and wind power data is difficult to describe. In practical training, when the GAN processes nonlinear data with complex feature quantity, not only is the model training difficult to reach balance, but also the problems of large difference between the generated data and the expected value, slow convergence speed of the model and the like may occur. Aiming at the problems, the method firstly adopts a random forest algorithm to screen wind power influence factors and judge the importance degree. The random forest algorithm eliminates factors with small influence on wind power, simplifies the model to a certain extent, and improves the prediction speed and precision. The generator is constructed using a Convolutional Neural Network (CNN), which not only has a strong capability of multi-hidden feature extraction, but also can share a convolutional kernel, thus being capable of easily handling the problem of high-dimensional data.
In the aspect of the prediction method, the short-term wind power prediction method mainly comprises a physical method, a traditional time series method, a statistical method and a combined prediction method. The traditional time sequence method can establish a model which can relatively meet the prediction precision requirement by only needing a limited sample sequence, and is widely applied, wherein the most used model mainly comprises three types: autoregressive model (AR), moving average Model (MA), and autoregressive moving average model (ARMA). The method only depends on the time sequence and autocorrelation of the sequence to provide information for modeling, ignores the influence of meteorological factors such as wind speed, temperature and the like on ultra-short-term wind power prediction, and has low prediction precision. Aiming at the problems, the method takes meteorological factors such as wind speed, temperature and the like as condition data, establishes a short-term wind power prediction model based on a condition generation countermeasure network, constrains the distribution of data generated by the model, and enables a generator to learn a sample probability distribution mapping relation under the corresponding condition.
Based on the above principle, the short-term wind power prediction method based on the condition generation countermeasure network in the embodiment of the invention specifically comprises the following steps (as shown in fig. 1):
firstly, selecting a historical wind power data time sequence and a historical meteorological data time sequence of a wind power plant power station, and cleaning data;
secondly, the identifiable errors (data consistency optimization, data missing, etc.) in the historical dataset are corrected as follows:
11 Data consistency optimization:
abnormal point detection is achieved using the 3σ criterion, i.e., outlier detection according to normal distribution. Firstly, assuming that only random errors exist in a measured value, then, determining a section of standard deviation values obtained by statistical processing of the standard deviation values according to corresponding probability, and judging that the deviation values beyond the section do not form random errors but are coarse deviations, namely, judging the deviation values as abnormal values, and eliminating the values with the deviation values, wherein the principle of 3 sigma criterion is as follows:
P(|x-μ|＞3σ)≤0.003 (1)
wherein the probability of the distribution of the values of (mu-sigma, mu+sigma) is 0.6826; the probability of the (mu-2σ, mu+2σ) value distribution in is 0.9545; the probability of the (μ -3σ, μ+3σ) value distribution in this is 0.9973.
The values are considered to be almost entirely concentrated in the (μ -3σ, μ+3σ) interval, and the probability outside the interval is less than 0.003. In normal distribution sigma represents standard deviation, mu represents mean value, and abnormal value is processed in the invention as follows:
a. if the abnormal value is obviously found but the quantity is smaller, the method directly eliminates the abnormal value;
b. if the algorithm is insensitive to the abnormal value, the algorithm is not processed;
c. replacing by using the average value;
d. treated as a missing value, and processed according to the method of processing the missing value.
12 Missing value processing):
and extracting missing values in the data set, and processing according to the key degree of the attribute where the missing values are and the distribution condition of the missing values. The specific method comprises the following steps:
a. when the deletion proportion is less (< 5%), and the sequence attribute is not strongly correlated with the target predicted sequence (the Pearson correlation coefficient r of the sequence and the target predicted sequence meets the requirements of: |r| < 0.8), the sequence is selected for median filling, and the expression of the person correlation coefficient r is as follows:
wherein x is i For wind power related influencing factors, y i The method is a target prediction sequence, namely a wind power sequence;
b. when the deletion rate is high (> 95%) and the attribute importance degree is low (the Pearson correlation coefficient r of the sequence and the target predicted sequence meets the requirements that r is less than 0.3), the attribute is directly deleted;
c. when the missing value is high and the attribute importance degree is high (the sequence and the Pearson correlation coefficient r of the target predicted sequence meet the requirement of 0.8-r < 1), a hot platform interpolation method is used, namely a time sequence (matching time sequence) similar to a sample where the missing value is located is found in a non-missing data time sequence, and the missing value is interpolated by using an observed value in the time sequence.
13 Standardized processing
Considering that the larger span of the data distribution interval can influence the descending speed of the deep learning model when gradient descending solution is carried out, and model convergence is not facilitated. In addition, the order of magnitude difference existing between the multidimensional data is easy to cause errors, so that the data is mapped into a [0,1] interval by adopting min-max normalization operation, and the normalization formula is as follows:
wherein: x is X norm Is normalized data; x is current data; x is X min Is the minimum value in the data; x is X max Is the maximum value in the data.
And secondly, performing feature selection by using a random forest algorithm, firstly constructing feature vectors by using the normalized influence factor data in the first step, then taking the feature vectors as the input of the random forest method, obtaining the importance degree of each influence factor through a calculation result, selecting the most important influence factor as condition data, and then constructing a short-term wind power prediction model (shown in figure 3) based on CNN-LSTM-CGAN improved condition generation countermeasure network. The specific operation is as follows:
21 Based on the more characteristics existing in the prediction, the embodiment utilizes a random forest method (shown in fig. 4) to screen main factors influencing wind power fluctuation for characteristic selection, and the specific steps are as follows:
a) Constructing a short-term wind power prediction feature vector according to the collected influence factor data sequence;
b) The random forest method is used for characteristic selection, and comprises the following specific steps:
b1 From the original training data set, randomly extracting k new self-service sample sets in a put-back way by applying a bootstrap method, and constructing k classification regression trees, wherein each time the samples which are not extracted form k OOB;
b2 Randomly decimating m at each node of each tree y The method comprises the steps of selecting a feature with the highest classification capability from the features as a randomly generated feature subset, and selecting the feature with the highest classification capability from the features to perform node splitting by calculating the information content of each feature in the feature subset, so that decision trees have larger diversity;
b3 Scoring with Gini indexCalculating the importance degree of the feature, if the feature X j If the node belongs to the set M in the decision tree i, X j The importance at the ith tree is as follows:
b4 Let n trees in RF total, the overall score is:
b5 Normalized for all calculated importance scores as follows:
c) And obtaining a feature vector importance degree matrix. With epsilon=1 as criterion, if VIM j If ∈is not less than, the influence factor is classified as input data category, if VIM j <Epsilon, which influence factor is negligible, several influence factors exceeding 1 are formed into an input matrix as condition data.
22 The objective function of generating a short-term wind power prediction model of the countermeasure network based on the CNN-LSTM-CGAN improved condition is:
L G ＝-E z,y (D(G(z|y)|y)) (7)
L D ＝-E x,y (D(x|y))+E z,y (D(G(z|y)|y)) (8)
wherein: g represents a generator, D represents a neural network, L G Loss function of generator, L D Is the loss function of the neural network, x is the real power data, y is the condition data, z is the noise data, L CGAN Is a loss function of CNN-LSTM-CGAN, E x,y (. Cndot.) represents the expected value for the x, y distribution, E z,y Table of (-)Showing expected values for z, y distribution; g (-) and D (-) are data output by the generator and the neural network, respectively.
Initializing the weight of the CNN-LSTM-CGAN improved condition generation countermeasure network, setting the maximum iteration number K=50 and the current K 0 =1, the specific contents are:
a) The method comprises the steps of (1) carrying out Gaussian initialization on a convolution layer, sampling from Gaussian distribution with a mean value of 0 and a variance of 1, and taking the Gaussian distribution as an initial weight;
b) Calling a zero_state function (Tensorflow existing initialization function) to realize LSTM network initialization;
and fourthly, constructing a generator by using the CNN, inputting the feature matrix obtained in the second step and the random noise data into a four-layer one-dimensional CNN network in the generator, and generating power data under the constraint of conditional data (shown in figure 2).
Fifthly, constructing a discriminator by utilizing the LSTM, introducing the power data, the real data and the condition data generated in the fourth step into the LSTM network together for countertraining, obtaining a preliminary prediction result which is close to the real power data distribution through a full-connection layer under the constraint of the condition data, reading specific training loss curves and error curves of the obtained training set and verification set, observing the longitudinal distance between the training set and the verification set loss curves in the convergence process, and visually evaluating the convergence performance of the network prediction result by combining the absolute error conditions of the training set and the verification set. The following is a convergence represented by three common fitting states:
a. when the loss curve of the training set is almost not reduced, the training set is in an under fitting state and is in an unconverged state;
b. when the training set loss curve continuously descends, verifying that the training set loss curve does not descend at a certain moment, wherein the training set loss curve is in an overfitting state and is in a convergence state but not in perfect convergence;
c. and when the loss curves of the training set and the verification set have no obvious distance, the perfect fitting state is realized, and the perfect convergence is realized.
And step six, introducing a characteristic loss function, calculating the improved condition generation of the CNN-LSTM-CGAN to generate a prediction error of the countermeasure network, and optimizing the parameter weight.
Further, the invention optimizes the characteristic loss function, and comprises the following specific steps:
61 Combining with the traditional loss functions (L1 and L2 norms) to enable the generation result of the generator to be closer to the real data distribution, wherein the generation result obtained by using the L1 norms as the loss functions by the CNN-LSTM-CGAN is more accurate, and the redefined L1 norms loss functions L are as follows:
L L1 ＝E x,y,z (||x-G(z|y)||) (10)
wherein: x is real power data, y is conditional data, z is noise data, E x,y,z (. Cndot.) represents the expected value for the x, y, z distribution, G (z|y) being the generated power data.
62 Utilizing the characteristic value deviation between the power data and the real data generated by the hidden layer measurement generator of the LSTM neural network to further improve the accuracy of the CNN-LSTM-CGAN for predicting the ultra-short-term wind power and the characteristic loss function L per The method comprises the following steps:
L per (G,x)＝E x,y,z (λ i P i (G,x)) (11)
wherein: p (P) i (G, x) the generator generated prediction data and real power data are in the hidden layer h i Wherein h is the average deviation value of i Hiding a layer for an ith layer of the neural network; h is a i (x) Is h i True power data values at the hidden layer; h is a i (G) To be in hidden layer h i A predicted data value generated by the generator; c (C) i 、H i And W is i Respectively is a hidden layer h i Channel number, height and width values; lambda (lambda) i Is a weight parameter.
63 To L1 norm loss function L and characteristic loss function L per After combination, the total loss function of the short-term wind power prediction model is as follows:
L(G,D)＝min G max D L CGAN +λ L1 L L1 +λ per L per (G,x) (13)
wherein: lambda (lambda) L1 And lambda (lambda) per Calculating a difference value between the preliminary predicted value and the true value by using the loss function as a weight parameter of the loss function;
64 After the difference value is obtained, the model is reversely transmitted to update each parameter to reduce the loss between the real value and the predicted value, so that the predicted value generated by the model is close to the real value, and the learning purpose is achieved;
65 The whole CNN-LSTM-CGAN short-term wind power prediction model is continuously subjected to iterative optimization until the maximum iterative times are reached, so that the model training optimal parameters are obtained.
Step seven, judging the iteration times of the step six, if the iteration times reach the maximum iteration times (K > K), ending the iteration, outputting CNN-LSTM-CGAN improved condition to generate parameters of the countermeasure network, otherwise, enabling k=k+1, and executing the step four;
and step eight, predicting by using the trained short-term wind power prediction model to obtain a final wind power prediction result. The model training process is carried out by adopting a generator and neural network alternate training mode, the weight of the generator is optimized after the neural network is updated once, the generator and the neural network are optimized by utilizing an Adam optimizer, and the model convergence is quickened by adopting a random gradient descent algorithm.
According to the method, the fluctuation characteristics of the historical output power and the historical meteorological time sequence are considered, a model is constructed, the fluctuation time sequence is extracted, and the hidden deep relation between the non-stable fluctuation sequences is fully excavated; the improved CNN-LSTM-CGAN hybrid network prediction model comprehensively considers the correlation among wind speeds, temperatures and humidities at different heights and wind power data and the inherent characteristics of short-term wind power time sequence, so that the inherent information related to a historical characteristic sequence is fully extracted, a convolutional neural network CNN is introduced to construct an internal structure of a generator, the LSTM constructs the internal structure of a discriminator, the power data generated by the generator and real data are subjected to game countermeasure training under the constraint of conditional data, a characteristic loss function is further optimized, and the optimized parameters of the network structure are fed back to the generator and the discriminator, so that the fitting degree of the generated power data and the real power data is improved, a better prediction effect is achieved, and the generalization of the model is improved.
While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions may be made without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims (6)

1. A short-term wind power prediction method based on a condition generation countermeasure network, comprising the steps of:
1) Selecting a historical wind power data time sequence and a historical meteorological data time sequence of a wind power plant power station, and cleaning the data;
2) Screening wind power prediction influence factors and judging importance degrees from the cleaned data in the step 1) by using a random forest method to obtain weights of different influence factors, removing factors with lower influence on prediction, generating a feature matrix as condition data, and establishing and training a short-term wind power prediction model based on CNN-LSTM-CGAN improved conditions to generate an countermeasure network;
3) Initializing a weight of a CNN-LSTM-CGAN improved condition generation countermeasure network, and setting the maximum iteration times;
4) Utilizing a CNN construction generator to input the condition data and the random noise data obtained in the step 2) into four layers of one-dimensional CNN networks in the generator at the same time to generate power data under the constraint of the condition data;
5) Constructing a discriminator by utilizing LSTM, introducing the power data obtained in the step 4) into an LSTM network together with real data and condition data to perform countermeasure training, and obtaining a preliminary prediction result close to real power data distribution through a full connection layer under the constraint of the condition data;
6) Introducing a characteristic loss function, calculating a CNN-LSTM-CGAN improved condition to generate a prediction error of the countermeasure network based on the preliminary prediction result obtained in the step 5), and optimizing parameter weights;
7) Judging the iteration times of the step 6), if the iteration times reach the maximum iteration times set in the step 3), ending the iteration, outputting CNN-LSTM-CGAN improved condition to generate parameters of the countermeasure network, otherwise, enabling the current iteration times to be +1, and executing the step 4);
8) And predicting by using the trained short-term wind power prediction model, and finally obtaining a wind power prediction result.
2. The short-term wind power prediction method based on condition generation countermeasure network according to claim 1, wherein in the step 2), the importance degree of wind speeds of different heights of 10m, 30m and 70m and influence of temperature, air pressure and humidity weather data on historical wind power fluctuation is calculated by using a random forest method, feature selection is performed, and a feature matrix is obtained as condition data, and the specific steps are as follows:
21 Randomly extracting K new self-service book sets from the original training data set by applying a bootstrap method, and constructing K classification regression trees, wherein K OOB sets are formed by the books which are not extracted each time;
22 Randomly decimating m at each node of each tree ny Each feature is used as a random feature subset, the information quantity of each feature in the feature subset is calculated, and the information quantity is m ny Selecting one of the features with the most classification capability to perform node splitting;
23 Score VIM with Gini index j (Gini) The calculation formula for calculating the feature importance degree, gini index is:
wherein: k represents a total of K categories; p (P) mk Representing the proportion of the class k in the node m;
feature X j The significance of the node m, namely the Gini index change amount calculation formula before and after the node m branches, is as follows:
wherein: GI (GI) l 、GI r Respectively representing Gini indexes of two new nodes after branching.
3. The short-term wind power prediction method based on a condition generation countermeasure network according to claim 1, wherein the step 3) initializes weights of the CNN-LSTM-CGAN improved condition generation countermeasure network, and the specific steps include:
31 Gaussian initializing the convolution layer, sampling from the gaussian distribution with the mean value of 0 and the variance of 1 as an initial weight;
32 A zero_state function is called to realize LSTM composite network initialization.
4. The short-term wind power prediction method based on a condition generation countermeasure network according to claim 1, wherein the objective function of the short-term wind power prediction model based on a CNN-LSTM-CGAN modified condition generation countermeasure network is:
L G ＝-E z,y (D(G(z|y)|y))
L D ＝-E x,y (D(x|y))+E z,y (D(G(z|y)|y))
min G max D L CGAN ＝E x,y (lnD(x|y))+E z,y (ln(1-D(G(z|y)|y)))
wherein: g represents a generator, D represents a neural network, L G Loss function of generator, L D Is the loss function of the neural network, x is the real power data, y is the condition data, z is the noise data, L CGAN Is a loss function of CNN-LSTM-CGAN, E x,y (. Cndot.) represents the expected value for the x, y distribution, E z,y (. Cndot.) represents the expected value for the z, y distribution; g (-) and D (-) are data output by the generator and the neural network, respectively.
5. The short-term wind power prediction method based on a condition generation countermeasure network according to claim 1, wherein in step 6), further comprising the improvement of a characteristic loss function, specifically comprising the steps of:
61 Combining with the traditional loss functions L1 and L2 norms, the generating result of the generator is more similar to the real data distribution, the generating result obtained by using the L1 norms as the loss functions by the CNN-LSTM-CGAN is more accurate, and the redefined L1 norms are as follows:
L L1 ＝E x,y,z (||x-G(z|y)||)
wherein: x is real power data, y is conditional data, z is noise data, E x,y,z (. Cndot.) represents the expected value for the x, y, z distribution, G (z|y) is the generated power data;
62 Utilizing the characteristic value deviation between the power data and the real data generated by the hidden layer measurement generator of the LSTM neural network to further improve the accuracy of the CNN-LSTM-CGAN for predicting the ultra-short-term wind power and the characteristic loss function L per The method comprises the following steps:
L per (G,x)＝E x,y,z (λ i P i (G,x))
wherein: p (P) i (G, x) the generator generated prediction data and real power data are in the hidden layer h i Wherein h is the average deviation value of i Hiding a layer for an ith layer of the neural network; h is a i (x) Is h i True power data values at the hidden layer; h is a i (G) To be in hidden layer h i A predicted data value generated by the generator; c (C) i 、H i And W is i Respectively is a hidden layer h i Channel number, height and width values; lambda (lambda) i Is a weight parameter;
63 To L1 norm loss function L and characteristic loss function L per After combination, the total loss function of the short-term wind power prediction model is as follows:
L(G,D)＝min G max D L CGAN +λ L1 L L1 +λ per L per (G,x)
wherein: lambda (lambda) L1 And lambda (lambda) per Calculating a difference value between the preliminary predicted value and the true value by using the loss function as a weight parameter of the loss function;
64 After the difference value is obtained, the model is reversely transmitted to update each parameter to reduce the loss between the real value and the predicted value, so that the predicted value generated by the model is close to the real value, and the learning purpose is achieved;
65 The whole CNN-LSTM-CGAN short-term wind power prediction model is continuously subjected to iterative optimization until the maximum iterative times are reached, so that the model training optimal parameters are obtained.
6. The short-term wind power prediction method based on the condition generation countermeasure network according to claim 1, wherein a generator and neural network alternate training mode is adopted in the short-term wind power prediction model training process, the neural network is updated once and then the generator is updated, an Adam optimizer is utilized to optimize the generator and the neural network, and a random gradient descent algorithm is adopted to accelerate model convergence.
CN202310387839.4A 2023-04-07 2023-04-07 Short-term wind power prediction method based on condition generation countermeasure network Pending CN116488151A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310387839.4A CN116488151A (en) 2023-04-07 2023-04-07 Short-term wind power prediction method based on condition generation countermeasure network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310387839.4A CN116488151A (en) 2023-04-07 2023-04-07 Short-term wind power prediction method based on condition generation countermeasure network

Publications (1)

Publication Number Publication Date
CN116488151A true CN116488151A (en) 2023-07-25

Family

ID=87213100

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310387839.4A Pending CN116488151A (en) 2023-04-07 2023-04-07 Short-term wind power prediction method based on condition generation countermeasure network

Country Status (1)

Country Link
CN (1) CN116488151A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117424232A (en) * 2023-12-19 2024-01-19 南京信息工程大学 Short-term photovoltaic power prediction method based on three-dimensional meteorological data multi-source fusion

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117424232A (en) * 2023-12-19 2024-01-19 南京信息工程大学 Short-term photovoltaic power prediction method based on three-dimensional meteorological data multi-source fusion
CN117424232B (en) * 2023-12-19 2024-03-19 南京信息工程大学 Short-term photovoltaic power prediction method based on three-dimensional meteorological data multi-source fusion

Similar Documents

Publication Publication Date Title
CN110969290B (en) Runoff probability prediction method and system based on deep learning
CN110751318B (en) Ultra-short-term power load prediction method based on IPSO-LSTM
CN106971240A (en) The short-term load forecasting method that a kind of variables choice is returned with Gaussian process
CN111563706A (en) Multivariable logistics freight volume prediction method based on LSTM network
CN110929847A (en) Converter transformer fault diagnosis method based on deep convolutional neural network
CN113107626B (en) Load prediction method of combined cycle generator set based on multivariable LSTM
CN111401599A (en) Water level prediction method based on similarity search and L STM neural network
CN113468817B (en) Ultra-short-term wind power prediction method based on IGOA (insulated gate bipolar transistor) optimized ELM (ELM)
CN110837223A (en) Combustion optimization control method and system for gas turbine
CN115860177A (en) Photovoltaic power generation power prediction method based on combined machine learning model and application thereof
CN116488151A (en) Short-term wind power prediction method based on condition generation countermeasure network
CN111210089A (en) Stock price prediction method of gated cyclic unit neural network based on Kalman filtering
CN114021483A (en) Ultra-short-term wind power prediction method based on time domain characteristics and XGboost
CN116842337A (en) Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model
CN112529684A (en) Customer credit assessment method and system based on FWA _ DBN
CN115358437A (en) Power supply load prediction method based on convolutional neural network
CN116169670A (en) Short-term non-resident load prediction method and system based on improved neural network
CN117473424A (en) Transformer fault diagnosis method, system, equipment and medium based on random forest
CN115964923A (en) Modeling method for forecasting 80-100km atmospheric wind speed in adjacent space based on VMD-PSO-LSTM
CN113095547A (en) Short-term wind power prediction method based on GRA-LSTM-ICE model
CN116680532A (en) Transformer fault online diagnosis method for processing unbalanced small sample based on NNTR-SMOTE oversampling
CN111950696A (en) Short-term power load prediction method based on dimension reduction and improved neural network
CN113496255B (en) Power distribution network mixed observation point distribution method based on deep learning and decision tree driving
CN115759343A (en) E-LSTM-based user electric quantity prediction method and device
Li et al. A short-term wind power forecasting method based on NWP wind speed fluctuation division and clustering

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