CN115630316A - Ultrashort-term wind speed prediction method based on improved long-term and short-term memory network - Google Patents

Abstract

The invention belongs to the field of electrical technology, and discloses an ultra-short-term wind speed prediction method based on an improved long-term and short-term memory network. The prediction method includes: calculation of delay time and embedded dimension of wind speed sequence by coordinate delay method; analysis of chaotic characteristics of wind speed sequence; Information method, constructing wind speed sequence and reconstructing phase space; wind speed sequence standardization processing; long short-term memory network training; based on the above steps, carry out ultra-short-term wind speed prediction. In terms of ultra-short-term wind speed prediction, the embodiment of the present invention determines the chaotic characteristics of the wind speed sequence through Lyapunov index analysis, constructs the phase space of the wind speed sequence reconstruction by introducing the mutual information method, and carries out the analysis of the characteristics of the wind speed sequence to realize the high-dimensional wind speed sequence The extraction of features, the introduction of long short-term memory network in the regression prediction, realized the accurate prediction of wind speed.

Description

Ultra-short-term wind speed prediction method based on improved long-short-term memory network

technical field

The invention relates to the field of electrical technology, in particular to an ultra-short-term wind speed prediction method based on an improved long-short-term memory network.

Background technique

Vigorously developing and utilizing clean new energy such as wind and light is a solid force supporting my country's energy transformation and development, and an important measure to gradually improve my country's energy structure and achieve high-quality green development. The grid connection of renewable energy also contributes to the security and stability of the power system. Running presents serious challenges. As a time series data forecasting method, high-precision wind speed prediction can provide reference for distribution network day-ahead and intra-day scheduling and generation of microgrid control instructions by predicting wind speed points or wind speed sequence data in the next control cycle, thereby improving power distribution. System operation economy and stability.

Existing research mainly focuses on the application of neural network algorithms, regression prediction algorithms, and scenarios such as interval prediction and spatial correlation analysis. However, few existing studies have analyzed the inherent characteristics of time series. In the process of actually training the prediction model, it relies on a large amount of wind speed or power data as the basis, and has not fully explored the internal characteristics of sequence changes. lower. Therefore, how to introduce multi-dimensional data analysis methods on the basis of the above research, and use the analysis results as input for model training to fit the high-dimensional characteristics of wind speed, so as to build a wind speed prediction method that adapts to wider scenarios and has higher prediction accuracy requires more research. A lot of exploration and research.

The domestic journal "Acta Solaris Sinica" in Volume 42, Issue 9, titled "Ultra-short-term Wind Speed Prediction Based on Sequence-to-Sequence and Attention Mechanism" discloses an ultra-short-term wind speed prediction method based on sequence-to-sequence network and attention mechanism. The 1-dimensional convolutional neural network and the gated recurrent unit are used to encode the wind speed sequence data, and then the attention mechanism and the gated recurrent unit are used to dynamically decode the semantic vector to obtain the predicted value. The model considers the adaptability of wind speed sequence and specific neural network, and improves the accuracy of ultra-short-term wind speed prediction. However, the above control methods do not analyze and study the inherent characteristics of the wind speed sequence, and cannot consider the original multi-mechanical characteristics of the wind speed in the prediction process, and the ultra-short-term wind speed prediction accuracy is low.

Contents of the invention

In order to solve the above-mentioned deficiencies in the prior art, the present invention provides an ultra-short-term wind speed prediction method and device, a computer-readable storage medium, and computer equipment based on an improved long-short-term memory network.

An ultra-short-term wind speed prediction method based on an improved long-short-term memory network, including:

(1) Calculation of delay time and embedding dimension of wind speed series by coordinate delay method;

(2) Analysis of chaotic characteristics of wind speed sequence;

(3) Based on the mutual information method, the reconstruction phase space of the wind speed sequence is constructed;

(4) standardization of wind speed sequence;

(5) Long-term short-term memory network training;

(6) Based on the above steps (1)-(5), carry out ultra-short-term wind speed prediction.

Preferably, for the coordinate delay method described in step (1), for a one-dimensional chaotic sequence { _xi |i=1,2,...,N}, the general expression of its reconstructed phase space is:

Among them, x _i is the chaotic sequence element, m is the embedding dimension, τ is the delay time, n is the length of the phase space, where n=N-(m-1)τ, N is the length of the chaotic sequence.

Preferably, the calculation of the wind speed sequence delay time of the coordinate delay method described in step (1) includes:

Remember the time series x(t), then it is x(t+τ) under the delay time τ, in the case of known sequence x(t), the mutual information between x(t+τ) and I(x(t +τ), x(t)) can be expressed as:

I(x(t+τ),x(t))＝H(x(t+τ))-H(x(t+τ)|x(t))

Among them, i is the number of elements in the time series x(t), j is the number of elements in the time series x(t+τ), P _x(t) is the probability of x(t), P _{x(t),x (t+τ)} is the joint probability density of x(t) and x(t+τ), and H is the intermediate variable function.

Preferably, the calculation of the wind speed sequence embedding dimension of the coordinate delay method described in step (1) includes:

Note that the phase point vector in the m-dimensional phase space is X _i , its nearest neighbor point is X _ji , and the distance between them is:

r _m,i ＝||X _i -X _j,i ||

When the dimension of the phase space increases by 1, the distance between them is updated as:

If r _m+1,i is significantly greater than r _m,i , it can be judged to be a false nearest neighbor point;

Define the embedded dimension evolution function E ₁ (m):

If the time series is a sequence of a definite system, then E ₁ (m) will tend to be stable after m reaches a certain value, and the corresponding m when the function is stable is selected as the appropriate embedding dimension;

Define the embedding dimension auxiliary function E ₂ (m):

The greater the randomness of the sequence, the smaller the fluctuation of E ₂ (m) around 1; when the fluctuation of E ₂ (m) is large, the sequence can be considered as a chaotic time series.

Preferably, the chaotic characteristic analysis of the wind speed sequence described in step (2) includes:

The wolf method of Lyapunov index is used to calculate, and the steps are as follows:

1) Based on the reconstruction method of the coordinate delay method described in step (1), it is determined that the embedding dimension of the phase space is m, and the delay time is τ, then the phase space constructed by a chaotic time series with a length of N can be expressed as:

X＝{X _i |i∈1,2,...,n}

X _i ＝{ _xi , _xi+τ ,..., _xi+(m-1)τ }

where n=N-(m-1)τ;

2) Select X ₀ as the initial point, record the distance between it and its nearest neighbor as L ₀ , set the distance threshold as ε, and when the time advances to a certain moment t, when L ₀ is greater than ε, record the maximum value of X _t Adjacent point X _t ′ and the distance value L ₀ ′ at this time, and search another nearest neighbor point of X _t ′ according to the principle of minimum included angle. When the distance between two points is less than ε, record the distance value L _i at this time;

3) Repeat step 2) until all points in the phase space are traversed, and the total number of iterations is recorded as κ, then the Lyapunov exponent of the system is:

Preferably, based on the mutual information method described in step (3), constructing a wind speed sequence to reconstruct the phase space includes:

Assuming that the wind speed sequence at a certain observation point in a certain period of time is V={v _i ,i=1,2,...,N}, where N is the number of data points, the known wind speed sequence is reconstructed into the following phase space:

Among them, v _i is the wind speed sequence element, m is the embedding dimension, τ is the delay time, N=n+(m-1)τ.

Preferably, the wind speed sequence standardization process described in step (4) includes:

The wind speed series were normalized using the following formula:

为风速序列均值，σ(V)为风速序列方差。in,

is the mean value of the wind speed series, and σ(V) is the variance of the wind speed series.

Preferably, the long-short-term memory network training described in step (5) includes:

作为神经网络输入，第二个时间点的实际标准化风速，即

作为输出，对网络进行训练，并依时间向前滚动迭代，直至训练集最后一个点作为输出，参与网络训练。Select the first 90% of the known wind speed sequence as the neural network training set, and the last 10% as the verification set. The training set and the verification set sequence are respectively expanded by m and τ as the phase space, and the first point in the phase space of the training set is taken as Phase space information, namely

As input to the neural network, the actual normalized wind speed at the second time point, namely

As an output, the network is trained, and iterates forward according to time until the last point of the training set is used as an output to participate in network training.

Preferably, based on the above-mentioned steps (1)-(5) described in step (6), the ultra-short-term wind speed prediction is carried out, including:

For the trained neural network, the phase space information of the first N points in the verification set is still used as input, and the actual standardized wind speed of the N+1th point is used as output, and the output value is denormalized and compared with the actual wind speed to obtain a prediction Accuracy, prediction accuracy is measured by root mean square error RMSE, and its calculation formula is:

Among them, v ^pre _k is the predicted wind speed, and v ^real _k is the actual wind speed.

An ultra-short-term wind speed prediction device based on an improved long-short-term memory network, including:

Calculation module, calculating the wind speed sequence delay time and embedding dimension of the coordinate delay method;

The analysis module analyzes the chaotic characteristics of the wind speed sequence;

The building block, based on the mutual information method, constructs the wind speed sequence and reconstructs the phase space;

Standardization processing module, wind speed sequence standardization processing;

Training module, long short-term memory network training;

The forecasting module predicts ultra-short-term wind speed.

A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above method steps are implemented.

A computer device includes a memory and a processor, the memory stores a computer program, and it is characterized in that the above-mentioned method steps are implemented when the processor executes the computer program.

The positive beneficial effect of the present invention:

The present invention proposes an ultra-short-term wind speed prediction method based on the improved long-term short-term memory network LSTM, analyzes the chaotic characteristics of the wind speed sequence and the phase space reconstruction method, and aims at the current situation that the traditional LSTM network prediction method cannot fully extract and utilize the characteristics of the wind speed sequence, Based on the randomness and chaotic characteristics of the wind speed sequence, a wind speed prediction model based on LSTM and phase space reconstruction is constructed, and the measured wind speed is used to verify the chaotic characteristics of the wind speed sequence and the effectiveness of the proposed prediction method. Through Lyapunov exponent analysis , determined the chaotic characteristics of the wind speed sequence, introduced the mutual information method to construct the wind speed sequence and reconstructed the phase space, carried out the analysis of the wind speed sequence characteristics, realized the extraction of high-dimensional features of the wind speed sequence, introduced the long-term short-term memory network in the regression prediction, and realized the wind speed The exact prediction of , as detailed below:

1) The wind speed sequence phase space reconstruction method of the present invention can effectively extract the characteristics of the wind speed sequence and restore its original dynamic phase space, providing effective data support for the subsequent prediction process;

2) The wind speed prediction method based on LSTM and phase space reconstruction of the present invention can memorize the inherent characteristics of the wind speed sequence through the training of a large amount of data, so that the high-dimensional data characteristics of the wind speed sequence can be used more efficiently in the prediction process, making the method more traditional. The error of the prediction result of the LSTM method is greatly reduced;

3) Simply increasing the input dimension of the neural network can not effectively improve the prediction accuracy. In order to improve the learning efficiency and prediction accuracy of the neural network, the present invention needs to decompose and reconstruct the inherent characteristics of the wind speed sequence to extract its characteristic parameters and improve the prediction accuracy. high.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description serve to explain the principles of the invention.

Figure 1 is a schematic diagram of wind speed sequence;

Fig. 2 is a flow chart of the prediction method of the present invention;

FIG. 3 is a diagram showing an iterative process of delay time calculation according to an embodiment of the present invention;

Fig. 4 is a schematic diagram showing iterative evolution of embedding dimensions according to an embodiment of the present invention;

Fig. 5 is the ultra-short-term wind speed prediction timing waveform diagram under the prediction method of the present invention;

Fig. 6 is the ultra-short-term wind speed prediction error analysis diagram under the prediction method of the present invention;

Figure 7 is a timing waveform diagram of ultra-short-term wind speed prediction using the Case1 method;

Figure 8 is an analysis diagram of ultra-short-term wind speed prediction error using the Case1 method;

Figure 9 is a timing waveform diagram of ultra-short-term wind speed prediction using the Case2 method;

Figure 10 is an analysis diagram of ultra-short-term wind speed prediction error using the Case2 method;

Fig. 11 is a schematic diagram of a computer device according to an embodiment of the present invention.

Detailed ways

The following description and the accompanying drawings sufficiently illustrate specific embodiments herein to enable those skilled in the art to practice them. Portions and features of some embodiments may be included in or substituted for those of other embodiments. The scope of embodiments of the present invention includes the full scope of the claims, and all available equivalents of the claims. In the present invention, the terms "first", "second", etc. are only used to distinguish one element from another element, without requiring or implying any actual relationship or order between these elements. In fact the first element can also be called the second element and vice versa. Furthermore, the terms "comprising", "comprising" or any other variation thereof are intended to cover a non-exclusive inclusion such that a structure, means or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed elements, or also elements inherent in the structure, device or equipment. Without further limitations, an element defined by the phrase "comprising a" does not preclude the presence of additional identical elements in the structure, device or equipment comprising said element. Each embodiment of the present invention is described in a progressive manner, each embodiment focuses on the differences from other embodiments, and the same and similar parts of the various embodiments can be referred to each other.

The terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", The orientation or positional relationship indicated by "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying the referred device or positional relationship. Elements must have certain orientations, be constructed and operate in certain orientations, and therefore should not be construed as limitations on the invention. In the description of the present invention, unless otherwise specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a mechanical connection or an electrical connection, and it can also be the internal communication of two components , may be directly connected, or may be indirectly connected through an intermediary. Those skilled in the art can understand the specific meanings of the above terms according to specific situations.

In the present invention, unless otherwise specified, the term "plurality" means two or more.

In the present invention, the character "/" indicates that the preceding and following objects are an "or" relationship. For example, A/B means: A or B.

In the present invention, the term "and/or" is an associative relationship describing objects, indicating that there may be three relationships. For example, A and/or B means: A or B, or, A and B, these three relationships.

Figure 1 is a schematic diagram of the wind speed sequence. In this embodiment, the present invention provides an ultra-short-term wind speed prediction method based on an improved long-short-term memory network, referring to Fig. 2, including:

(1) Calculation of delay time and embedding dimension of wind speed series by coordinate delay method;

(2) Analysis of chaotic characteristics of wind speed sequence;

(3) Based on the mutual information method, the reconstruction phase space of the wind speed sequence is constructed;

(4) standardization of wind speed sequence;

(5) Long-term short-term memory network training;

(6) Based on the above steps (1)-(5), carry out ultra-short-term wind speed prediction.

The method of this embodiment designs a phase space reconstruction method based on the coordinate delay method to restore the high-dimensional dynamic characteristics of the wind speed sequence; designs an ultra-short-term wind speed prediction method based on a long-term short-term memory network to achieve accurate prediction of ultra-short-term wind speed.

In one embodiment, the coordinate delay method described in step (1), for a one-dimensional chaotic sequence { _xi |i=1,2,...,N}, the general expression of its reconstructed phase space is :

Among them, x _i is the chaotic sequence element, m is the embedding dimension, τ is the delay time, n is the length of the phase space, where n=N-(m-1)τ, N is the length of the chaotic sequence.

Further, the calculation of the wind speed sequence delay time of the coordinate delay method described in step (1) includes:

Remember the time series x(t), then it is x(t+τ) under the delay time τ, in the case of known sequence x(t), the mutual information between x(t+τ) and I(x(t +τ), x(t)) can be expressed as:

I(x(t+τ),x(t))＝H(x(t+τ))-H(x(t+τ)|x(t))

Among them, i is the number of elements in the time series x(t), j is the number of elements in the time series x(t+τ), P _x(t) is the probability of x(t), P _{x(t),x (t+τ)} is the joint probability density of x(t) and x(t+τ), and H is the intermediate variable function.

Furthermore, in the phase space reconstruction, the expected effect is that the correlation between different state point vectors is as small as possible, so the first minimum value of the mutual information function can be selected as the delay time setting value.

Further, the calculation of the embedded dimension of the wind speed sequence of the coordinate delay method described in step (1) includes:

Note that the phase point vector in the m-dimensional phase space is X _i , its nearest neighbor point is X _ji , and the distance between them is:

r _m,i =||X _i -X _j,i || (2)

When the dimension of the phase space increases by 1, the distance between them is updated as:

If r _m+1,i is significantly greater than r _m,i , it can be judged to be a false nearest neighbor point;

Further, define the embedded dimension evolution function E ₁ (m):

If the time series is a sequence of a definite system, then E ₁ (m) will tend to be stable after m reaches a certain value, and the corresponding m when the function is stable is selected as the appropriate embedding dimension.

Furthermore, in practical applications, due to the limitation of the amount of data, it is difficult to distinguish the steady state and the slow growth state of the E ₁ (m) function of the finite sequence, that is, it is difficult to distinguish the chaotic sequence that reaches the steady state from the random sequence that tends to saturation. sequence, so define the embedding dimension auxiliary function E ₂ (m):

The greater the randomness of the sequence, the smaller the fluctuation of E ₂ (m) around 1; when the fluctuation of E ₂ (m) is large, the sequence can be considered as a chaotic time series.

In one embodiment, the chaotic characteristic analysis of the wind speed sequence described in step (2) includes:

The convergence and divergence of the system and the sensitivity to initial conditions are reflected by the Lyapunov index of the system. The specific steps of the wolf method for calculating the Lyapunov index are as follows:

1) Based on the above reconstruction method, the embedding dimension of the phase space is determined to be m, and the delay time is τ, then the phase space constructed from the chaotic time series with length N can be expressed as:

where n=N-(m-1)τ;

2) Select X ₀ as the initial point, record the distance between it and its nearest neighbor as L ₀ , set the distance threshold as ε, and when the time advances to a certain moment t, when L ₀ is greater than ε, record the maximum value of X _t Adjacent point X _t ′ and the distance value L ₀ ′ at this time, and search another nearest neighbor point of X _t ′ according to the principle of minimum included angle. When the distance between two points is less than ε, record the distance value L _i at this time;

3) Repeat step 2) until all points in the phase space are traversed, and the total number of iterations is recorded as κ, then the Lyapunov exponent of the system is:

In one embodiment, in step (3), based on the mutual information method, the reconstruction phase space of the wind speed sequence is constructed, including:

Assume that the wind speed sequence of a certain observation point within a certain period of time is V={v _i , i=1,2,...,N}, where N is the number of data points. Then according to the aforementioned theory, the known wind speed sequence can be reconstructed into the following phase space:

Among them, v _i is the wind speed sequence element, m is the embedding dimension, τ is the delay time, N=n+(m-1)τ.

In one embodiment, in step (4), the wind speed sequence is standardized, in order to prevent the divergence of neural network training results, the wind speed sequence needs to be standardized with the following formula:

为风速序列均值，σ(V)为风速序列方差。in,

is the mean value of the wind speed series, and σ(V) is the variance of the wind speed series.

In one embodiment, the long short-term memory network is trained in step (5), including:

In the long-short-term memory network, the forgetting gate realizes the selective forgetting of the information transmitted by the previous unit by controlling the reception probability of the information contained in the output S _t-1 of the previous unit. The activation function selects the Sigmoid function, and the forgetting The gate can be expressed as:

f _t ＝Sigmoid(w _fS S _t-1 +w _fx x _t +δ _f ) (7)

Among them, w _fS , w _fx are the weights of the output of the previous unit passing through the forget gate and the actual input of this unit, and f is the calculation iteration parameter of the forget gate.

Furthermore, the input gate realizes the selective reception of the input to the unit memory state of the unit by controlling the degree of reception of the information contained in the input x _t of the unit, and at the same time, it can also perform the function of changing the memory state of the unit. The input gate contains two ports, which respectively use Sigmoid and tanh as activation functions:

Among them, w _iS , w _ix are the weights of the output of the previous unit through the input gate and the actual input of this unit, i is the bias parameter for calculating the input gate, and w _iS , w _ix are the output of the previous unit that updates the memory state of the unit With the actual input weight of this unit, C calculates the bias parameter for state update.

Furthermore, the input gate also has the function of updating the memory state of the unit, in which the inheritance of the long-term memory state of the unit is realized by multiplying the forgetting gate and the long-term memory state, and the inheritance of the short-term memory state actually input by the unit is realized by the two input gates. Port information is multiplied to achieve, where * represents element-wise multiplication:

Further, the output gate sets channels for the output considering long-term memory and only short-term memory, where the output corresponding to short-term memory is:

o _t ＝Sigmoid(w _oS S _t-1 +w _ox x _t +δ _o ) (10)

The output corresponding to the long-term memory is:

S _t ＝o _t *tanh(C _t ) (11)

Furthermore, for the neural network composed of the above units, there are four sets of parameters that need to be trained. Same as the general RNN, the BPTT algorithm is still used to solve the model. The main steps are as follows:

其中

计算公式为：1) Initialize the training parameters, calculate the output of each gate unit and the expected output of the current sequence index

in

The calculation formula is:

Among them, V and c are the weights and biases of the regression layer, which are regarded as auxiliary parameters for calculation in the unit, and do not participate in the forward propagation process.

2) Define the loss function J and the unit error term δ _t :

Then the forward pass error term can be expressed as:

in:

Then it is easy to get the error gradient of the parameter to be optimized as:

3) Utilize the gradient descent principle in the iteration to update the unit connection weight:

w _ij +λδ _i x _ij →w′ _ij (17)

Among them, λ is the learning efficiency, which is generally reduced after the number of iterations reaches the threshold, so as to avoid network non-convergence due to oscillation.

作为神经网络输入，第二个时间点的实际标准化风速，即

作为输出对网络进行训练，并依时间向前滚动迭代，直至训练集最后一个点作为输出，参与网络训练，可见输入量为一个长度为m的序列，序列点数为0.9n-1。In step (5), the first 90% of the known wind speed sequence is selected as the neural network training set, and the last 10% is the verification set. The training set and the verification set sequence are respectively expanded by m and τ as the phase space. The phase space information of the first point in , namely

As input to the neural network, the actual normalized wind speed at the second time point, namely

As the output, the network is trained, and iterates forward according to time until the last point of the training set is used as the output to participate in network training. It can be seen that the input is a sequence of length m, and the number of sequence points is 0.9n-1.

Further, based on the above steps (1)-(5) described in step (6), the ultra-short-term wind speed prediction is carried out, including:

For the trained neural network, the phase space information of the first N points in the verification set is still used as input, and the actual standardized wind speed of the N+1th point is used as output, and the output value is denormalized and compared with the actual wind speed to obtain a prediction Accuracy, prediction accuracy is measured by root mean square error RMSE, and its calculation formula is:

Among them, v ^pre _k is the predicted wind speed, and v ^real _k is the actual wind speed.

In order to verify the effectiveness of the control strategy proposed in the present invention, a phase space reconstruction and LSTM prediction model were constructed based on the Matlab2020b platform, and a numerical example was carried out for verification.

The iterative process of the mutual information function changing with the delay time is shown in Figure 3. It can be seen that the first minimum point appears in the iterative process when τ = 30, so the phase space reconstruction delay time is determined to be 30.

After determining the delay time, the iterative calculation of the embedding dimension is performed. Assuming that the maximum embedding dimension is 50, as the embedding dimension increases, the changes of the embedding dimension evolution functions E ₁ (m) and E ₂ (m) are shown in Fig. 4 . It can be seen that when the embedding dimension is at the point of m=9, E ₁ (m) turns into a stable state, while the value of E ₂ (m) still fluctuates around 1, and then it can be determined that this wind speed sequence is not a random sequence, and is relatively The embedding dimension for spatial reconstruction is 10.

Based on the above delay time and embedding dimension, the Lyapunov exponent calculated by the wolf method is 0.000265>0, then the original wind speed sequence has chaotic characteristics, and the chaotic sequence method can be used to analyze the wind speed sequence.

Based on the wind speed prediction method based on the improved LSTM proposed by the present invention, the prediction model is constructed. The delay time of the wind speed phase space is 30, and the embedding dimension is 10. The LSTM network includes four layers of sequence input layer, LSTM layer, fully connected layer and regression layer. The LSTM layer contains 200 hidden units. Specify the number of network training iterations to be 150, the gradient threshold to be 1, and the initial learning rate to be 0.005. After 75 generations, the learning rate begins to decrease, and the rate of decrease is 0.2. Due to the superiority of the Adam algorithm in dealing with the LSTM regression problem, the Adam algorithm is used here to solve it. The predicted wind speed curve and the actual wind speed curve are shown in Figure 5, and the error distribution and root mean square error are shown in Figure 6.

It can be seen that the fit between the predicted wind speed curve and the actual wind speed curve is good, and the root mean square error is 0.39484. The proposed wind speed prediction method based on LSTM and phase space reconstruction has high accuracy.

In order to verify the superiority of the proposed method, the present invention sets up two comparison examples Case1 and Case2 based on the LSTM prediction method, and the input uses the actual standardized wind speed, and Case1 uses the actual wind speed of the previous point to predict the wind speed of the next point. Single input prediction In Case2, the actual wind speed of the previous 10 points is used to predict the wind speed of the next point with multiple inputs. The results are shown in Figures 7 and 8 and Figures 9 and 10 respectively. As can be seen from the above results, the root mean square error of Case1 is 0.42746, and the root mean square error of Case2 is 0.63626, then the root mean square error of the prediction method proposed by the present invention is smaller than Case1 and Case2, and it can be seen from the error distribution diagram that the error distribution of the method proposed by the present invention is It is more uniform, indicating that it has a better extraction effect on sequence features and higher prediction accuracy. At the same time, it can be seen that the prediction effect of Case2 is worse than that of Case1. It can be seen that simply increasing the input dimension of the prediction model cannot improve the prediction accuracy, but it is necessary to extract the characteristics of the wind speed sequence and reflect the characteristics into the input in real time to effectively improve. predictive effect.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure is shown in FIG. 11 . The computer device includes a processor, memory and a network interface connected by a system bus. Wherein, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs and databases. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps in the above method embodiments can be realized.

Those skilled in the art can understand that the structure shown in Figure 11 is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer equipment on which the solution of the present invention is applied. The specific computer equipment can be More or fewer components than shown in the figures may be included, or some components may be combined, or have a different arrangement of components.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the foregoing method embodiments are implemented.

In one embodiment, there is also provided a computer device, including a memory and a processor, where a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented through computer programs to instruct related hardware, and the computer programs can be stored in a non-volatile computer-readable memory In the medium, when the computer program is executed, it may include the processes of the embodiments of the above-mentioned methods. Wherein, any reference to memory, storage, database or other media used in the various embodiments provided by the present invention may include at least one of non-volatile memory and volatile memory. The non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory or optical memory, and the like. Volatile memory may include random access memory (Random Access Memory, RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

Finally, it is noted that the above embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Those of ordinary skill in the art may make other modifications or equivalent replacements to the technical solution of the present invention, as long as they do not depart from the spirit and spirit of the technical solution of the present invention. All should be included in the scope of the claims of the present invention.

Claims (10)

1. An ultra-short term wind speed prediction method based on an improved long-short term memory network is characterized by comprising the following steps:

(1) Calculating the wind speed sequence delay time and the embedding dimension by a coordinate delay method;

(2) Analyzing the chaos characteristic of the wind speed sequence;

(3) Constructing a wind speed sequence reconstruction phase space based on a mutual information method;

(4) Wind speed sequence standardization processing;

(5) Training a long-term and short-term memory network;

(6) And (5) performing ultra-short-term wind speed prediction based on the steps (1) - (5).

2. The ultra-short term wind speed prediction method based on the improved long and short term memory network as claimed in claim 1, wherein the calculation of the wind speed sequence delay time by the coordinate delay method in step (1) comprises:

time series x (t) is recorded, then it is x (t + τ) at delay time τ, and in case of known series x (t), x (t + τ) and its mutual information I (x (t + τ), x (t)) can be expressed as:

I(x(t+τ),x(t))＝H(x(t+τ))-H(x(t+τ)|x(t))

where i is the number of the elements in the time series x (t), j is the number of the elements in the time series x (t + τ), P _x(t) Probability of x (t), P _x(t),x(t+τ) Is the joint probability density of x (t) and x (t + τ), and H is an intermediate variable function.

3. The ultra-short term wind speed prediction method based on the improved long-short term memory network as claimed in claim 1, wherein the calculation of the wind speed sequence embedding dimension of the coordinate delay method in step (1) comprises:

recording the phase point vector in the m-dimensional phase space as X _i The nearest neighbor point is X _ji The distance between them is:

r _m,i ＝||X _i -X _j,i ||

when the phase space dimension increases by 1, the distance between them is updated as:

if r _m+1,i Is significantly greater than r _m,i Then it can be determined as a false nearest point;

defining an embedding dimension evolution function E ₁ (m)：

If the time series is a series of definite systems, E ₁ (m) the function tends to be stable after m reaches a certain value, and the m corresponding to the function is selected as a proper embedding dimension when the function is stable;

defining an embedding dimension auxiliary function E ₂ (m):

The greater the randomness of the sequence, E ₂ The smaller the fluctuation of the value of (m) around 1; when E is ₂ (m) when the fluctuation is large, the sequence can be regarded as a chaotic time sequence.

4. The ultra-short term wind speed prediction method based on the improved long-short term memory network as claimed in claim 1, wherein the wind speed sequence chaotic characteristic analysis in the step (2) comprises:

the calculation by adopting the wolf method of the Lyapunov index comprises the following steps:

1) Based on the reconstruction method of the coordinate delay method in step (1), if it is determined that the embedding dimension of the phase space is m and the delay time is τ, the phase space constructed by the chaos time sequence with the length of N can be expressed as:

X＝{X _i |i∈1,2,...,n}

X _i ＝{x _i ,x _i+τ ,...,x _i+(m-1)τ }

wherein N = N- (m-1) τ;

2) Selecting X ₀ As an initial point, recording the distance between the most adjacent point and the initial point as L ₀ Setting the distance threshold to epsilon, then advancing the time to a time t such that L ₀ When is greater than epsilon, record X _t Closest point X of (2) _t ' and the distance value L at this time ₀ ', and searching for X according to the principle of minimum included angle _t ' when the distance between two points is less than epsilon, the distance value at this time is recorded as L _i ；

3) Repeating the step 2) until all points in the phase space are traversed, and recording the total iteration times as kappa, wherein the systematic Lyapunov exponent is as follows:

5. the ultra-short term wind speed prediction method based on the improved long-short term memory network as claimed in claim 1, wherein the step (3) of constructing the wind speed sequence reconstruction phase space based on the mutual information method comprises:

suppose that the wind speed sequence in a certain observation point in a certain time period is V = { V = _i I =1, 2.·, N }, where N is the number of data points, the known wind speed sequence is reconstructed as the following phase space:

wherein v is _i For wind speed sequence elements, m is the embedding dimension, τ is the delay time, N = N + (m-1) τ.

6. The ultra-short term wind speed prediction method based on the improved long-short term memory network as claimed in claim 1, wherein the wind speed sequence normalization process in step (4) comprises:

the wind speed sequence is normalized using the following formula:

wherein,

is the mean value of the wind speed sequence, and sigma (V) is the variance of the wind speed sequence.

7. The ultra-short term wind speed prediction method based on the improved long-short term memory network as claimed in claim 1, wherein the training of the long-short term memory network in the step (5) comprises:

selecting the first 90% of the known wind speed sequence as a neural network training set and the last 10% as a verification set, respectively expanding the training set and the verification set sequence by m and tau as phase spaces, wherein the phase space information of the first point in the phase space of the training set is taken, namely

As input to the neural network, the actual normalized wind speed at the second point in time, i.e.

And as output, training the network, and rolling and iterating forwards according to time until the last point of the training set is used as output to participate in network training.

8. The ultra-short term wind speed prediction method based on the improved long and short term memory network as claimed in claim 1, wherein the step (6) of developing ultra-short term wind speed prediction based on the above steps (1) - (5) comprises:

for the trained neural network, before the verification centralization is still carried outThe phase space information of N points is used as input, the actual standardized wind speed of the (N + 1) th point is used as output, the output value is subjected to standardization and then is compared with the actual wind speed to obtain the prediction precision, the prediction precision is measured by a root mean square error RMSE, and the calculation formula is as follows:

wherein v is ^pre _k To predict wind speed, v ^real _k Is the actual wind speed.

9. An ultra-short term wind speed prediction device based on an improved long-short term memory network is characterized by comprising:

the computing module is used for computing the wind speed sequence delay time and the embedding dimension of the coordinate delay method;

the analysis module is used for analyzing the chaos characteristic of the wind speed sequence;

the construction module is used for constructing a wind speed sequence reconstruction phase space based on a mutual information method;

the wind speed sequence is subjected to standardization processing;

the training module is used for training the long-term and short-term memory network;

and the prediction module predicts the ultra-short term wind speed.

10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor realizes the method steps of any of claims 1-8 when executing the computer program.

Images