CN115330085A - Wind speed prediction method based on deep neural network and without future information leakage - Google Patents

Wind speed prediction method based on deep neural network and without future information leakage Download PDF

Info

Publication number
CN115330085A
CN115330085A CN202211136465.0A CN202211136465A CN115330085A CN 115330085 A CN115330085 A CN 115330085A CN 202211136465 A CN202211136465 A CN 202211136465A CN 115330085 A CN115330085 A CN 115330085A
Authority
CN
China
Prior art keywords
wind speed
prediction
data
sequence data
prediction model
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
CN202211136465.0A
Other languages
Chinese (zh)
Inventor
闫渤文
申瑞芳
李珂
舒臻孺
王振国
杨庆山
周绪红
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.)
Chongqing University
Original Assignee
Chongqing University
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 Chongqing University filed Critical Chongqing University
Priority to CN202211136465.0A priority Critical patent/CN115330085A/en
Publication of CN115330085A publication Critical patent/CN115330085A/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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

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

Abstract

The invention discloses a wind speed prediction method based on a deep neural network and without future information leakage, which comprises the following steps: step 1: respectively screening and preprocessing the wind speed sequence data; screening the effective component Tc according to the sub-modal energy proportion and taking the effective component Tc as an output value of a prediction model; preprocessing the wind speed sequence data by adopting a real-time rolling decomposition strategy to obtain an input value of a prediction model without information leakage; constructing a data set with input values and output values; and 2, step: determining a prediction step size: determining the future time steps needing to be predicted according to the actual prediction demand; and step 3: constructing a prediction model: constructing a bidirectional long-short term memory network combined with an attention mechanism as a prediction model, and training and testing the prediction model by using a data set; and 4, step 4: and (3) wind speed prediction: and predicting the wind speed by using the prediction model obtained by training to obtain a wind speed prediction result. The method avoids the risk of data leakage and improves the practicability.

Description

Wind speed prediction method based on deep neural network and without future information leakage
Technical Field
The invention belongs to the technical field of wind speed prediction, and particularly relates to a wind speed prediction method based on a deep neural network and without future information leakage.
Background
Accurate wind speed prediction is of great importance in the aspects of improvement of weather forecast precision, early warning of meteorological disasters, development of clean wind power and the like. In the field of wind power, accurate short-term wind speed prediction can enable operators to reasonably regulate and control the overall operation of a wind power plant, maintain stable energy supply and reduce the impact on a power grid. With the gradual development of domestic onshore wind power plants to low wind speed areas, more and more wind power plants are located in mountain areas. Due to the influence of factors such as terrain change, wake effect is easy to generate in mountainous regions, and complicated flow phenomena such as flow separation and reattachment, shear flow and the like occur. Therefore, in order to ensure safe and stable operation of the wind power plant in the mountainous area, a method for accurately predicting the short-term wind speed needs to be developed urgently.
Currently, the commonly used prediction methods are classified into the following categories: physical methods, statistical methods, artificial intelligence methods, and hybrid methods. The physical method consumes a large amount of computing resources and time, performs well in medium and long-term prediction, but is difficult to meet the timeliness requirement of short-term prediction. Due to the inherent linear assumption of the statistical method, reasonable modeling and prediction of the nonlinear wind speed sequence are difficult to perform. The artificial intelligence method only uses a shallow neural network and lacks the capability of extracting deep nonlinear features of a wind speed sequence.
With the rapid development of computer capability, it becomes feasible to use a deep neural network to predict wind speed. Although short-term wind speed prediction models based on deep learning achieve good prediction results, it is undeniable that a single neural network model may be difficult to adapt to different wind speed sequences. Moreover, a large amount of noise information exists in the actually measured wind speed sequence, which seriously affects the accuracy of wind speed prediction. Therefore, hybrid predictive models incorporating data preprocessing methods have been widely developed, with modal decomposition methods being most widely used. The data preprocessing operation can obviously reduce the randomness and high-frequency noise of the wind speed, improve the predictability of the wind speed and the performance of a prediction model, and greatly improve the accuracy of wind speed prediction. However, in the hybrid prediction model, all wind speed sequence data are decomposed during data preprocessing, and a training set and a test set are divided on decomposed sub-modes to establish a prediction model. This means that test set data that should not be known is considered known, inevitably resulting in the leakage of future information. When new data is added into the original wind speed sequence data, decomposition needs to be carried out again, and then the new data is substituted into the sub-modal prediction model for prediction. However, even if only part of the new data is added at the end of the original data, the decomposed sub-modalities will produce significant changes at the end of the sequence. This part of the end data is the most critical information in timing prediction. This means that a prediction model built on a sub-modality of the original data may be difficult to apply to the new data, resulting in a lack of utility for such hybrid prediction models.
Disclosure of Invention
In view of this, the present invention provides a wind speed prediction method based on a deep neural network and without future information leakage, and the method avoids the risk of data leakage and improves the practicability by adopting a post-evaluation manner.
In order to achieve the purpose, the invention provides the following technical scheme:
a wind speed prediction method based on a deep neural network and without future information leakage comprises the following steps:
step 1: respectively carrying out data screening and data preprocessing on the wind speed sequence data; screening the effective component Tc according to the sub-mode energy ratio and taking the effective component Tc as an output value of a prediction model; preprocessing the wind speed sequence data by adopting a real-time rolling decomposition strategy to obtain an input value of a prediction model without information leakage; constructing a data set with input values and output values;
step 2: determining a prediction step size: determining the future time step number needing to be predicted according to the actual prediction demand;
and step 3: constructing a prediction model: constructing a bidirectional long-short term memory network combined with an attention mechanism as a prediction model, and training and testing the prediction model by using a data set;
and 4, step 4: and (3) wind speed prediction: and predicting the wind speed by using the prediction model obtained by training to obtain a wind speed prediction result.
Further, in the first step, the method for screening out the effective component Tc from the wind speed sequence data comprises the following steps:
s1: dividing the wind speed sequence data into a training set, a verification set and a test set, and establishing three data groups by using the divided data groups, wherein the first data group comprises the training set, the second data group comprises the verification set, and the third data group comprises the test set;
s2: modal decomposition is carried out on the wind speed sequence data in the three data groups respectively;
s3: based on the energy principle, the index is judged, and a proper submodule is selected to obtain the effective component Tc capable of representing most energy of the wind speed sequence data i I =1,2 or 3, respectively, representing the effective fractions screened from the wind speed sequence data of the three data sets;
s4: sequentially adding Tc as effective component i And splicing to obtain a target output value Tc of the prediction model.
Further, in step S3, when the sum of the current m sub-modal energies is greater than or equal to 99% of the energy of the original wind speed sequence, tc is i The ER calculation mode of the submode is the sum of the first m submodes:
Figure BDA0003852310430000021
wherein ER (k) represents the energy fraction of the kth sub-mode;
Figure BDA0003852310430000022
representing a kth sub-modality sequence; x represents wind speed sequence data; n represents the length of the wind speed sequence data X.
Further, in the first step, the method for preprocessing the wind speed sequence data by adopting the real-time rolling decomposition strategy comprises the following steps:
(1) Determining the window length L: analyzing the frequency spectrum characteristics of the wind speed sequence data X, searching a frequency peak point to obtain a corresponding period, and selecting the period number as the window length required by RTRD;
(2) Determining a sliding step s;
(3) Reconstructing a wind speed signal: reconstructing the wind speed sequence data X into a two-dimensional matrix X according to the window length L and the sliding step s R The matrix shape is (N-L + s, L), wherein N is the length of the wind speed sequence data X;
(4) And executing RTRD: progressive decomposition of X R Obtaining a decomposition result without information leakage
Figure BDA0003852310430000031
Further, the prediction model comprises an input layer, two Bi-LSTM networks, an attention layer and a full connection layer which are sequentially arranged.
Further, the principle of the LSTM network is as follows:
f t ＝σ(W xf x t +W hf h t-1 +b f )
g t ＝tanh(W xg x t +W hg h t-1 +b g )
i t ＝σ(W xi x t +W hi h t-1 +b i )
c t ＝f t ⊙c t-1 +i t ⊙g t
o t ＝σ(W xo x t +W ho h t-1 +b o )
y t ＝h t ＝o t ⊙tanh(c t )
wherein x is t As input vector, c t In a long-term state, h t In a short-term state, y t Is an output vector; w is a group of xf ,W xg ,W xi ,W xo Are respectively and x t Weight matrix of connections, W hf ,W hg ,W hi ,W ho Are respectively h t-1 Weight matrix of connections, b f ,b g ,b i ,b o Four layers of bias terms respectively; f. of t For forgetting the controller of the door, by f t Determination c t-1 Which portions of (a) should be deleted; g t Is the intermediate output of the LSTM network, which acts to analyze x t And h t-1 ；i t For the controller of the input gate, judge g t Which important parts of c should be added to t ；o t C is judged to be read for the controller of the output gate t And applied to h t And y t Performing the following steps; σ denotes the activation function.
Further, the Attention layer adopts a Luong-Attention model, and the calculation method comprises the following steps:
the score calculation method is as follows:
Figure BDA0003852310430000032
attention weight α ts The calculation method comprises the following steps:
Figure BDA0003852310430000033
wherein,
Figure BDA0003852310430000034
and h t All hidden states and the last hidden state output by the encoder are respectively; w represents a weight matrix;
context vector
Figure BDA0003852310430000035
The calculation method is：
Figure BDA0003852310430000036
Wherein alpha is ts It is the weight of the attention that is being paid,
Figure BDA0003852310430000041
is a fully hidden state;
attention vector a t The calculation method comprises the following steps:
Figure BDA0003852310430000042
wherein,
Figure BDA0003852310430000043
is a context vector; h is a total of t Is the last hidden state; w is a group of c Representing the model parameters to be learned.
Further, attention vector a t And inputting the wind speed prediction result into a full connection layer to obtain a wind speed prediction result.
The invention has the beneficial effects that:
the existing single-point wind speed combined prediction model generally adopts a full-decomposition data preprocessing strategy, and the data preprocessing strategy has the risk of data leakage and can cause the combined prediction model to lack the engineering practicability, so that the accurate prediction of the effective components of the wind speed is more critical and feasible. Considering that a large amount of high-frequency noise exists in the actually measured wind speed, the high-frequency noise is an important source of randomness and volatility of the wind speed, and the low-frequency effective component of the wind speed occupies most energy of the wind speed, the effective component of the wind speed sequence is extracted by a data preprocessing method without information leakage, an expression method without information leakage in the future is established, and the premise that a reliable and practical wind field prediction result is obtained is provided.
In consideration of the common problem of data leakage of the traditional combined prediction model and the noise component of a certain degree in the actually measured wind speed data, the invention provides a wind speed prediction method based on a deep neural network and without future information leakage, and aims to provide a method for accurately predicting the effective component of a wind speed sequence by adopting a short-term wind speed prediction framework of a data preprocessing strategy without information leakage. The prediction method comprises three parts of data screening, data preprocessing based on real-time rolling decomposition and Bi-LSTM neural network (Bi-LSTM-Attention) fusing an Attention mechanism, avoids the risk of data leakage, and effectively processes noise components.
Drawings
In order to make the object, technical scheme and beneficial effect of the invention more clear, the invention provides the following drawings for explanation:
FIG. 1 is a schematic diagram of a wind speed prediction method based on a deep neural network without future information leakage according to the present invention;
FIG. 2 is an illustration of the flow of data during the model training phase, the validation phase, and the testing phase (i.e., the use phase);
FIG. 3 is a single step prediction graph for all models.
Detailed Description
The present invention is further described below in conjunction with the drawings and the embodiments so that those skilled in the art can better understand the present invention and can implement the present invention, but the embodiments are not to be construed as limiting the present invention.
As shown in FIG. 1, the wind speed prediction method based on the deep neural network and without the leakage of future information of the invention comprises the following steps.
Step 1: respectively carrying out data screening and data preprocessing on the wind speed sequence data; screening the effective component Tc according to the sub-mode energy ratio and taking the effective component Tc as an output value of a prediction model; preprocessing the wind speed sequence data by adopting a Real-Time Rolling Decomposition (RTRD) strategy to obtain an input value of a prediction model without information leakage; a data set is constructed with input values and output values.
Specifically, the method for screening the effective component Tc from the wind speed sequence data comprises the following steps:
s1: the wind speed sequence data is divided into a training set, a validation set and a test set, in the embodiment, the wind speed sequence data is divided into the training set (1-0.6N), the validation set (0.6N-0.8N) and the test set (0.8N-N) according to the following ratio of 6.
S2: and performing modal decomposition on the wind speed sequence data in the three data sets respectively. And decomposing the wind speed data in the range of the first data group (1-0.6N), the second data group (0.6N-0.8N) and the third data group (0.8N-N) by adopting a modal decomposition method, wherein the VMD and SSA algorithms need to specify the decomposition layer number, and the EMD and ICEEMDAN algorithms can adaptively determine the decomposition layer number.
S3: based on the energy principle, the index is judged, and a proper submodule is selected to obtain the effective component Tc capable of representing most energy of the wind speed sequence data i I =1,2 or 3, respectively, representing the effective fractions screened from the wind speed sequence data of the three data sets. Specifically, the number of sub-modes (the sub-modes are arranged in the order from low frequency to high frequency) required for screening the effective component Tc according to the size of the Energy Ratio (ER) of the sub-modes is as follows: when the sum of the energy of the m current sub-modes is greater than or equal to 99% of the energy of the original wind speed sequence, tc can be considered i I.e. the sum of the first m sub-modes. The ER calculation for the sub-modality is:
Figure BDA0003852310430000051
wherein, ER (k) represents the energy ratio of the kth sub-mode;
Figure BDA0003852310430000052
representing a kth sub-modality sequence; x represents wind speed sequence data; n represents the length of the wind speed sequence data X.
S4: sequentially adding effective components Tc i Splicing to obtain target output value Tc of the prediction model, wherein Tc is target output of the subsequently constructed modelData source, i.e. true value.
Specifically, the method for preprocessing the wind speed sequence data by adopting the real-time rolling decomposition strategy comprises the following steps:
(1) Determining the window length L: analyzing the frequency spectrum characteristics of the wind speed sequence data X, searching a frequency peak point to obtain a corresponding period, and selecting a proper period number as a window length required by RTRD;
(2) Determining a sliding step s, and setting the sliding step s to be 1 when the short-term wind speed is predicted in a step-by-step prediction mode;
(3) Reconstructing a wind speed signal: according to the window length L and the sliding step s, the wind speed sequence data X is reconstructed into a two-dimensional matrix X R The matrix shape is (N-L + s, L), wherein N is the length of the wind speed sequence data X;
(4) And executing RTRD: progressive decomposition of X R Obtaining the decomposition result without information leakage
Figure BDA0003852310430000053
Will be provided with
Figure BDA0003852310430000054
The first m submodes of each row of data are added to form an input matrix
Figure BDA0003852310430000055
Calculating out
Figure BDA0003852310430000056
The correlation coefficient with Tc; judging composition under different decomposition algorithms by using correlation coefficient criterion
Figure BDA0003852310430000061
The optimal number of sub-modes.
Step 2: determining a prediction step size: and determining the number of future time steps needing to be predicted according to the actual predicted demand.
And 3, step 3: constructing a prediction model: and constructing a bidirectional long-short term memory network (Bi-LSTM-Attention) combined with an Attention mechanism as a prediction model, and training and testing the prediction model by using a data set. Specifically, in this embodiment, the prediction model includes a layer of input layer, two layers of Bi-LSTM network, a layer of attention layer, and a layer of full connectivity layer, which are sequentially arranged.
The LSTM network of this embodiment updates the hidden state at time t as follows:
f t ＝σ(W xf x t +W hf h t-1 +b f )
g t ＝tanh(W xg x t +W hg h t-1 +b g )
i t ＝σ(W xi x t +W hi h t-1 +b i )
c t ＝f t ⊙c t-1 +i t ⊙g t
o t ＝σ(W xo x t +W ho h t-1 +b o )
y t ＝h t ＝o t ⊙tanh(c t )
wherein x is t As an input vector, c t In a long-term state, h t In a short-term state, y t Is an output vector; w xf ,W xg ,W xi ,W xo Are respectively equal to x t Weight matrix of connections, W hf ,W hg ,W hi ,W ho Are respectively h t-1 Weight matrix of connections, b f ,b g ,b i ,b o Four layers of bias terms respectively; f. of t For forgetting the controller of the door, by t Determination c t-1 Which parts of (a) should be deleted; g t Is the intermediate output of the LSTM network, which acts to analyze x t And h t-1 ；i t For the controller of the input gate, judge g t Which important parts of c should be added to t ；o t C is judged to be read for the controller of the output gate t And applied to h t And y t Performing the following steps; σ denotes the activation function.
The Attention layer of this embodiment adopts the Luong-Attention model, and its calculation method is:
the score calculation method is as follows:
Figure BDA0003852310430000062
attention weight α ts The calculation method comprises the following steps:
Figure BDA0003852310430000063
wherein,
Figure BDA0003852310430000064
and h t All hidden states and the last hidden state output by the encoder are respectively; w represents a weight matrix;
context vector
Figure BDA0003852310430000065
The calculation method comprises the following steps:
Figure BDA0003852310430000066
wherein alpha is ts It is the weight of the attention that is being paid,
Figure BDA0003852310430000071
is a fully hidden state;
attention vector a t The calculation method comprises the following steps:
Figure BDA0003852310430000072
wherein,
Figure BDA0003852310430000073
is a context vector; h is t Is the last hidden state; w is a group of c Representing the model parameters to be learned.
Finally, it will be notedForce vector a t And inputting the wind speed prediction result into a full connection layer to obtain a wind speed prediction result.
And 4, step 4: and (3) wind speed prediction: and predicting the wind speed by using the prediction model obtained by training to obtain a wind speed prediction result.
Fig. 2 illustrates how the data operates during the training, validation, and testing phases of the neural network. In fig. 2 (a), when constructing the input of the model, the modal decomposition follows a real-time rolling decomposition strategy, which is performed on a blue data pane. This strategy ensures that the input of the model always uses information before the predicted value, i.e. the confidentiality of future information is ensured, which means that no information is leaked. The predicted value mainly refers to the effective component of the predicted wind speed, and is represented by Tc, and the meaning of the true value is followed. In FIG. 2 (b), during the model training process, a true value Tc is obtained because the data has been collected over the full time axis 1 Can be obtained by modal decomposition when training the neural network. Thus, the error can be directly calculated and used to update the parameters of the neural network. In FIG. 2 (b), during the validation and testing process, the parameters of the neural network have been determined, and the error Tc between the predicted result and the true value 2 Or Tc 3 In a delayed manner, rather than in real time. This means that errors can be calculated after a period of operation and they are only used to prove the accuracy of the prediction. In a word, a real-time rolling decomposition strategy is always adhered to in the training process or the verification and test process, and information leakage is avoided as the primary principle of the algorithm.
Experimental verification
The wind speed prediction method based on the deep neural network and without future information leakage is verified by combining specific examples.
Selecting 10-min interval wind speed data X of a certain wind measuring tower in Yunnan China at the height of 70m for 91 days continuously as an input signal, and selecting 13104 groups with the date of 2010.4.1-2010.6.30.
1. Prediction result error evaluation criterion
In this embodiment, three common performance indexes are selected as the criteria for measuring the error of the prediction model, including MAPE, MAE, and RMSE, which are defined as follows.
Figure BDA0003852310430000074
Figure BDA0003852310430000075
Figure BDA0003852310430000076
Wherein: x is the number of n Representing a target output value (also known as true value),
Figure BDA0003852310430000077
representing the predicted value. The smaller the values of MAPE, MAE and RMSE, the higher the prediction accuracy.
In addition, the relative lifting ratio P of the three error indexes of RMSE, MAPE and MAE is defined metric So as to further quantitatively evaluate the performance of different models. P metric Is defined as follows
Figure BDA0003852310430000081
Wherein, metric represents three indexes of MAPE, MAE and RMSE, E a ，E b Representing the prediction error of model a and model b.
2. Comparison of a predictive model with different models
The deep learning platform selected in this embodiment is TensorFlow 2.3, python 3.7 version based on GPU, and a Bi-LSTM-Attention model is constructed. An Adam optimizer is adopted in the training process, the learning rate lr is set to be 0.001, a loss function is selected to be Mean Squared Error, and the iteration number is fixed to be 100 epochs. The parameters of each layer of the model are determined by manual selection and random search.
Table 1 shows Persistence model (PR), BP model, GRU, LSTM and Bi-LSTM-Attention developed in the present inventionA comparison was made. Tables 1 and 2 show the RMSE, P for the single-step prediction and the 1-6-step prediction for different prediction models MSE 、MAPE、P MAPE MAE and P MAE And (4) error results.
TABLE 1 prediction Performance of each model based on VMD method-Single step prediction
Figure BDA0003852310430000082
TABLE 2 model prediction Performance based on VMD method-1-6 step prediction
Figure BDA0003852310430000083
It can be seen from the comparison of all the neural network models, the method can accurately predict the effective component Tc occupying most energy of the wind speed sequence, the accuracy of single-step and 1-6-step prediction of the model is superior to that of a PR model, and the Bi-LSTM-orientation prediction model provided by the invention obtains the best prediction accuracy, which fully proves the effectiveness of the RTRD-Bi-LSTM-orientation prediction method provided by the invention. Specifically, under single-step prediction, the precision improvement ratio P of the Bi-LSTM-Attention prediction module relative to the PR model metric Between 23.79% and 32.77%; under the prediction of steps 1-6, the precision improvement ratio P of the Bi-LSTM-Attention prediction module relative to the PR model metric 9.90 to 18.09 percent. And when the prediction step size is increased, P of all the neural network models metric All are reduced, which shows that the performance of the neural network model is reduced in multi-step wind speed prediction.
The above-mentioned embodiments are merely preferred embodiments for fully illustrating the present invention, and the scope of the present invention is not limited thereto. The equivalent substitution or change made by the technical personnel in the technical field on the basis of the invention is all within the protection scope of the invention. The protection scope of the invention is subject to the claims.

Claims (8)

1. A wind speed prediction method based on a deep neural network and without future information leakage is characterized by comprising the following steps: the method comprises the following steps:
step 1: respectively screening and preprocessing the wind speed sequence data; screening the effective component Tc according to the sub-mode energy ratio and taking the effective component Tc as an output value of a prediction model; preprocessing the wind speed sequence data by adopting a real-time rolling decomposition strategy to obtain an input value of a prediction model without information leakage; constructing a data set with input values and output values;
step 2: determining a prediction step size: determining the future time steps needing to be predicted according to the actual prediction demand;
and step 3: constructing a prediction model: constructing a bidirectional long-short term memory network combined with an attention mechanism as a prediction model, and training and testing the prediction model by using a data set;
and 4, step 4: and (3) wind speed prediction: and predicting the wind speed by using the prediction model obtained by training to obtain a wind speed prediction result.
2. The wind speed prediction method based on the deep neural network and without the leakage of future information as claimed in claim 1, wherein: in the first step, the method for screening out the effective component Tc from the wind speed sequence data comprises the following steps:
s1: dividing the wind speed sequence data into a training set, a verification set and a test set, and establishing three data groups by using the divided data groups, wherein the first data group comprises the training set, the second data group comprises the verification set, and the third data group comprises the test set;
s2: modal decomposition is carried out on the wind speed sequence data in the three data groups respectively;
s3: based on the energy principle judgment index, a proper submodule is selected to obtain the effective fraction Tc which can represent most of the energy of the wind speed sequence data i I =1,2 or 3, respectively, representing the effective fractions screened from the wind speed sequence data of the three data sets;
s4: sequentially adding effective components Tc i And splicing to obtain a target output value Tc of the prediction model.
3. The wind speed prediction method based on the deep neural network and without the future information leakage as claimed in claim 2, characterized in that: in step S3, when the sum of the current m sub-modal energies is greater than or equal to 99% of the energy of the original wind speed sequence, tc is i The ER calculation mode of the submode is the sum of the first m submodes:
Figure FDA0003852310420000011
wherein ER (k) represents the energy fraction of the kth sub-mode;
Figure FDA0003852310420000012
representing a kth sub-modality sequence; x represents wind speed sequence data; n represents the length of the wind speed sequence data X.
4. The wind speed prediction method based on the deep neural network and without the leakage of future information as claimed in claim 1, wherein: in the first step, the method for preprocessing the wind speed sequence data by adopting the real-time rolling decomposition strategy comprises the following steps:
(1) Determining the window length L: analyzing the frequency spectrum characteristics of the wind speed sequence data X, searching a frequency peak point to obtain a corresponding period, and selecting the period number as the window length required by RTRD;
(2) Determining a sliding step s;
(3) Reconstructing a wind speed signal: reconstructing the wind speed sequence data X into a two-dimensional matrix X according to the window length L and the sliding step s R The matrix shape is (N-L + s, L), wherein N is the length of the wind speed sequence data X;
(4) And executing RTRD: progressive decomposition of X R Obtaining the decomposition result without information leakage
Figure FDA0003852310420000021
5. The wind speed prediction method based on the deep neural network and without the leakage of future information as claimed in any one of claims 1 to 4, wherein: the prediction model comprises a layer of input layer, two layers of Bi-LSTM networks, a layer of attention layer and a layer of full connection layer which are sequentially arranged.
6. The wind speed prediction method based on the deep neural network and without the leakage of future information as claimed in claim 5, wherein: the principle of the LSTM network is as follows:
f t ＝σ(W xf x t +W hf h t-1 +b f )
g t ＝tanh(W xg x t +W hg h t-1 +b g )
i t ＝σ(W xi x t +W hi h t-1 +b i )
c t ＝f t ⊙c t-1 +i t ⊙g t
o t ＝σ(W xo x t +W ho h t-1 +b o )
y t ＝h t ＝o t ⊙tanh(c t )
wherein x is t As an input vector, c t In a long-term state, h t In a short-term state, y t Is an output vector; w is a group of xf ,W xg ,W xi ,W xo Are respectively and x t Weight matrix of connections, W hf ,W hg ,W hi ,W ho Are respectively h t-1 Weight matrix of connections, b f ,b g ,b i ,b o Four layers of bias terms respectively; f. of t For forgetting the controller of the door, by f t Determination c t-1 Which parts of (a) should be deleted; g t Is the intermediate output of the LSTM network, which acts to analyze x t And h t-1 ；i t For the controller of the input gate, judge g t Which important parts of c should be added to t ；o t For the controller of the output gate, determining that c should be read t And applied to h t And y t Performing the following steps; σ denotes laserA live function.
7. The wind speed prediction method based on the deep neural network and without the future information leakage as claimed in claim 5, characterized in that: the Attention layer adopts a Luong-Attention model, and the calculation method comprises the following steps:
the score calculation method is as follows:
Figure FDA0003852310420000022
attention weight α ts The calculation method comprises the following steps:
Figure FDA0003852310420000023
wherein,
Figure FDA0003852310420000024
and h t All hidden states and the last hidden state output by the encoder are respectively; w represents a weight matrix;
context vector
Figure FDA0003852310420000031
The calculation method comprises the following steps:
Figure FDA0003852310420000032
wherein alpha is ts It is the weight of attention that is being weighted,
Figure FDA0003852310420000033
is a fully hidden state;
attention vector a t The calculation method comprises the following steps:
Figure FDA0003852310420000034
wherein,
Figure FDA0003852310420000035
is a context vector; h is t Is the last hidden state; w c Representing the model parameters to be learned.
8. The wind speed prediction method based on the deep neural network and without the future information leakage as claimed in claim 6, characterized in that: attention vector a t And inputting the wind speed prediction result into a full connection layer to obtain a wind speed prediction result.
CN202211136465.0A 2022-09-19 2022-09-19 Wind speed prediction method based on deep neural network and without future information leakage Pending CN115330085A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211136465.0A CN115330085A (en) 2022-09-19 2022-09-19 Wind speed prediction method based on deep neural network and without future information leakage

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211136465.0A CN115330085A (en) 2022-09-19 2022-09-19 Wind speed prediction method based on deep neural network and without future information leakage

Publications (1)

Publication Number Publication Date
CN115330085A true CN115330085A (en) 2022-11-11

Family

ID=83915084

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211136465.0A Pending CN115330085A (en) 2022-09-19 2022-09-19 Wind speed prediction method based on deep neural network and without future information leakage

Country Status (1)

Country Link
CN (1) CN115330085A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116662766A (en) * 2023-08-01 2023-08-29 浙江大学 Wind speed prediction method and device based on data two-dimensional reconstruction and electronic equipment

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116662766A (en) * 2023-08-01 2023-08-29 浙江大学 Wind speed prediction method and device based on data two-dimensional reconstruction and electronic equipment
CN116662766B (en) * 2023-08-01 2023-10-03 浙江大学 Wind speed prediction method and device based on data two-dimensional reconstruction and electronic equipment

Similar Documents

Publication Publication Date Title
Zhang et al. Short-term wind speed prediction model based on GA-ANN improved by VMD
CN109272146B (en) Flood prediction method based on deep learning model and BP neural network correction
CN113887787B (en) Flood forecast model parameter multi-objective optimization method based on long-short-term memory network and NSGA-II algorithm
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
CN109272156B (en) Ultra-short-term wind power probability prediction method
CN111079998B (en) Flow prediction method based on long and short time sequence correlation attention mechanism model
CN112232543B (en) Multi-station prediction method based on graph convolution network
CN109583588B (en) Short-term wind speed prediction method and system
CN114169251A (en) Ultra-short-term wind power prediction method
CN115239029B (en) Wind power prediction method and system considering power time sequence and meteorological dependent characteristics
CN116826737A (en) Photovoltaic power prediction method, device, storage medium and equipment
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN112183877A (en) Photovoltaic power station fault intelligent diagnosis method based on transfer learning
CN115330085A (en) Wind speed prediction method based on deep neural network and without future information leakage
CN118428754A (en) New energy power generation input power prediction method and system
CN115204362A (en) Method for predicting residual life interval of mechanical equipment
CN118040678A (en) Short-term offshore wind power combination prediction method
CN117709488B (en) Dam seepage prediction method based on RUN-XGBoost
Xu et al. Short-term wind speed prediction based on GRU
CN115796327A (en) Wind power interval prediction method based on VMD (vertical vector decomposition) and IWOA-F-GRU (empirical mode decomposition) -based models
Zhang et al. A Novel Combined Model Based on Hybrid Data Decomposition, MSWOA and ENN for Short-Term Wind Speed Forecasting
CN112561203B (en) Method and system for realizing water level early warning based on clustering and GRU
CN114372615A (en) Short-term photovoltaic power prediction method and system considering spatial correlation
CN115217152A (en) Method and device for predicting opening and closing deformation of immersed tunnel pipe joint
CN111832818B (en) LSTM network power generation multi-step prediction method based on correlation analysis

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