Data prediction method and system of echo state network based on memristor

Abstract

The invention relates to the field of computers, and provides a memristor-based echo state network data prediction method and system. The method comprises the following steps: based on an obtained historical wind speed data set of a wind turbine, adopting an improved echo state network to obtain wind speed data; wherein the improved echo state network comprises the following steps: calculating a reserve pool connection matrix according to an initialized reserve pool state and a model parameter of a discrete memristor; and according to the reservoir connection matrix, aiming at the input data of the historical wind speed data set of the wind turbine at each moment, updating the reservoir state, and solving the wind speed data. According to the invention, accurate prediction of wind speed data of the wind turbine can be realized.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular, to a data prediction method and system for an echo state network based on memristors. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] The echo state network (ESN) has significant differences from traditional RNNs in model construction and learning algorithms. It uses pre-generated rather than trained hidden layer weight values and projects the input recursively into a high-dimensional, non-linear representation. However, the reservoir matrix in ESN is randomly initialized, which may lead to poor performance of the model in predicting complex non-linear systems. For wind turbine wind speed prediction, this means that the model may not be able to accurately capture the complex dynamic relationships in wind turbine wind speed data.

[0004] Although some scholars have proposed improvement strategies, such as performing singular value decomposition on the randomly generated reservoir matrix to obtain a new reservoir matrix or increasing the sparsity of the reservoir matrix, etc., there are problems such as redundant reservoir nodes, which reduce the prediction effect of the model on wind turbine wind speed data. Summary of the Invention

[0005] To solve the technical problems existing in the above background technique, the present invention provides a data prediction method and system for an echo state network based on memristors. The present invention uses a discrete memristor to replace the original randomly initialized reservoir weight matrix of the echo state network, which can not only capture the complex patterns in the wind turbine signal but also help the reservoir better retain important early information, thus more accurately analyzing long-term trends. However, such an update method will cause the gradual increase of redundant or irrelevant information, so dimensionality reduction is performed through compressive sensing technology to remove redundant information, improving the memory and prediction capabilities of the echo state network. Using an intelligent optimization algorithm to optimize the memristor coefficients can achieve accurate prediction of wind turbine wind speed data.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The first aspect of the present invention provides a data prediction method for an echo state network based on memristors.

[0008] A data prediction method for an echo state network based on memristors, comprising:

[0009] Based on the obtained historical wind speed dataset of the wind turbine, an improved echo state network is used to obtain wind speed data;

[0010] Among them, the improved echo state network includes: calculating a reservoir connection matrix according to the initialized reservoir state and the model parameters of the discrete memristor; updating the reservoir state for the input data at each moment of the historical wind speed data set of the wind turbine according to the reservoir connection matrix, and solving the wind speed data.

[0011] Furthermore, after each update of the reservoir state, compressive sensing processing is adopted to reduce the dimension of the reservoir state, and the compressed reservoir state is obtained.

[0012] Furthermore, the compressive sensing processing includes: transforming the reservoir state into a sparse transform domain to obtain a sparse vector; constructing an observation matrix based on the sparse vector; and recovering the observation matrix to obtain the compressed reservoir state.

[0013] Furthermore, during the training process of the improved echo state network, the artemisinin algorithm is adopted to optimize the model parameters of the discrete memristor, the normalized root mean square error in the training stage is used as the fitness function of the artemisinin algorithm, the optimal model parameters of the discrete memristor are obtained, and the reservoir state set and output weight matrix corresponding to the model parameters are calculated according to the optimal model parameters of the discrete memristor.

[0014] Furthermore, the method for solving the wind speed data includes: solving the wind speed data by using a least squares solution according to the updated reservoir state.

[0015] Furthermore, the updating of the reservoir state is represented by the following formula:

[0016] x(t) = f(W in u(t) + W res x(t - 1));

[0017] W res = α + β·x(t - 1) 2 ;

[0018] Among them, u(t) represents the data affecting the wind speed of the wind turbine, W in represents the input matrix, W res represents the reservoir matrix, x(t - 1) represents the reservoir state at the previous moment, f represents the internal neuron activation function, and both α and β represent the model parameters of the discrete memristor.

[0019] The second aspect of the present invention provides a data prediction system based on a memristive echo state network.

[0020] A data prediction system based on a memristive echo state network includes:

[0021] A prediction module, configured to: based on the acquired historical wind speed data set of the wind turbine, use an improved echo state network to obtain wind speed data;

[0022] Wherein, the improved echo state network includes: calculating a reservoir connection matrix according to the initialized reservoir state and the model parameters of the discrete memristor; updating the reservoir state for the input data at each moment of the historical wind speed data set of the wind turbine according to the reservoir connection matrix, and solving the wind speed data.

[0023] The third aspect of the present invention provides a computer-readable storage medium.

[0024] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in the data prediction method based on the memristive echo state network as described in the first aspect above.

[0025] The fourth aspect of the present invention provides a computer device.

[0026] A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the data prediction method based on the memristive echo state network as described in the first aspect above.

[0027] The fifth aspect of the present invention provides a computer program product or a computer program.

[0028] The present invention provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the data prediction method based on the memristive echo state network as described in the first aspect above.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] The present invention provides a data prediction method and system based on a memristive echo state network, including: based on the acquired historical wind speed data set of a wind turbine, an improved echo state network is used to obtain wind speed data; wherein, the improved echo state network includes: calculating a reservoir connection matrix according to the initialized reservoir state and the model parameters of a discrete memristor; according to the reservoir connection matrix, for the input data at each moment of the historical wind speed data set of the wind turbine, updating the reservoir state; after the reservoir state is updated, compressive sensing processing is introduced, and an artemisinin optimization algorithm is used to optimize the parameters of the discrete memristor model to obtain an improved echo state network; based on the prediction task, the time series data is input into the improved echo state network to obtain a wind speed prediction result. The present invention can better capture the dynamic change patterns in the data, help the reservoir better retain important information in the early stage, and reducing the redundancy of reservoir nodes is the key to improving the model effect.

[0031] Based on the basic structure of the echo state network, the present invention introduces a discrete memristor model to replace the random initialization of the reservoir connection matrix. It has a memory function and nonlinear characteristics, which can increase the internal nonlinear state of the reservoir and capture more information inside the reservoir; compressive sensing processing can reduce the dimension of the reservoir state, reduce data redundancy and computational amount; by introducing an artemisinin optimization algorithm, the parameters of the discrete memristor model are optimized to improve the prediction accuracy for wind turbine data. The improved echo state network proposed by the present invention can improve the prediction accuracy of the wind turbine wind speed and reduce prediction errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0033] Figure 1 is a flowchart of the data prediction method based on the memristive echo state network shown in the present invention;

[0034] Figure 2 is a schematic diagram of the network structure of the echo state network shown in the present invention;

[0035] Figure 3 is a schematic diagram of the network structure of the prediction model shown in the present invention;

[0036] Figure 4(a) is a prediction diagram of the wind turbine wind speed data shown in the present invention;

[0037] Figure 4(b) is an error curve diagram of the wind turbine wind speed data shown in the present invention;

[0038] Figure 5(a) is a prediction diagram of the sunspot number shown in the present invention;

[0039] Figure 5(b) is the error curve graph of the sunspot number shown in the present invention;

[0040] Figure 6(a) is the prediction graph of the Logistic system shown in the present invention;

[0041] Figure 6(b) is the error curve graph of the Logistic system shown in the present invention;

[0042] Figure 7(a) is the prediction graph of the x-axis of the Lorenz system shown in the present invention;

[0043] Figure 7(b) is the error curve graph of the x-axis of the Lorenz system shown in the present invention;

[0044] Figure 8(a) is the prediction graph of the y-axis of the Lorenz system shown in the present invention;

[0045] Figure 8(b) is the error curve graph of the y-axis of the Lorenz system shown in the present invention;

[0046] Figure 9(a) is the prediction graph of the z-axis of the Lorenz system shown in the present invention;

[0047] Figure 9(b) is the error curve graph of the z-axis of the Lorenz system shown in the present invention. Detailed implementation manners

[0048] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0049] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0050] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0051] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of methods and systems according to various embodiments of the present disclosure. It should be noted that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the flowchart and / or block diagram, as well as the combination of blocks in the flowchart and / or block diagram, may be implemented using a dedicated hardware-based system for performing the specified functions or operations, or may be implemented using a combination of dedicated hardware and computer instructions.

[0052] Embodiment 1

[0053] As Figure 1 shown, this embodiment provides a data prediction method for a memristor-based echo state network. This embodiment takes the application of this method to a server as an example. It can be understood that this method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, web servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and this application does not make any restrictions here. In this embodiment, the method includes the following steps:

[0054] Step 1: Obtain a prediction task, divide the historical wind speed data set of the wind turbine into a training set and a test set; input the divided data into a prediction model, and the prediction model is an improved echo state network. Specifically: use a discrete memristor to replace the original randomly initialized reservoir weight matrix, and adopt compressive sensing technology for dimensionality reduction.

[0055] Step 2: Initialize the input weight W in , the reservoir state x, the learning rate, the spectral radius, and the model parameters α and β of the discrete memristor;

[0056] As Figure 2As shown, the traditional echo state network includes an input layer, a reservoir, and an output layer. Assume that the echo state network has k input units, l internal processing units in the reservoir, and n output units;

[0057] At time n, the values of the input unit u(t), the internal state x(t), and the output unit y(t) are respectively:

[0058] u(t) = [u1(t), u2(t), …, u k (t)] T ;

[0059] x(t) = [x1(t), x2(t), …, x l (t)] T ;

[0060] y(t) = [y1(t), y2(t), …, y n (t)] T ;

[0061] Use the historical wind speed dataset of the wind turbine as the input u(t) at each moment;

[0062] The reservoir updates the corresponding state, and the state update equation is:

[0063] x(t) = f(W in ·u(t) + W res x(t - 1));

[0064] Among them, x(t) is the node state of the reservoir at the current moment, x(t - 1) is the node state of the reservoir at the previous moment, f represents the activation function of the internal neurons in the reservoir, and f(·) generally takes the tanh function or the sigmod function;

[0065] The output equation of the echo state network is:

[0066]

[0067] Among them, f out (·) represents the output function. Generally, the output layer is linear. In the wind turbine dataset, when inputting the data u(t) at a certain moment, it is the predicted value for the next step of the dataset. When calculating the error, calculate the true value and the predicted value for the next step, that is:

[0068]

[0069] Among them, err(t) is the error value at this moment, Figure 2 The connection weight matrix W in ∈Rl×k Denotes the connection weight matrix \(W\) from the input layer to the reservoir, and the connection from the reservoir state to the reservoir state res ∈R l×l Denotes the connection between neurons in the reservoir itself, and the connection weight matrix \(W\) from the reservoir state to the output out ∈R N×l Denotes the connection weight from the reservoir to the output layer. Among them, \(W\) in and \(W\) res are randomly initialized, and only \(W\) out is obtained through training;

[0070] The training process of the echo state network is mainly divided into two stages: the sampling training stage and the weight calculation stage.

[0071] Sampling stage: First, arbitrarily select the initial state of the network. Usually, the initial state of the network is selected as 0, that is, \(x(0)=0\). Through the input connection weight matrix \(W\) in Add the training samples \(u(n)\) (\(n = 1,2,\cdots,M\)) to the reservoir, and calculate and collect the state variables \(x(t)\) and the output \(y(t)\) in turn according to the corresponding equations. Assume that the system state is collected starting from a certain moment \(m\). At each moment, the vector \((x_1(i),x_2(i),\cdots,x\) l (i)) (\(i = m,m + 1,\cdots,M\)) forms a row, and each row forms a matrix \(X(M - m + 1,l)\). Similarly, the corresponding sample data \(y(n)\) is also collected and forms a column vector \(Y(M - m + 1,1)\) for subsequent calculation of the output connection weight matrix;

[0072] Weight calculation stage: Calculate the output connection weight matrix \(W\) of the echo state network out according to the state matrix and output collected in the sampling stage. The state variable \(x(t)\) and the predicted output are linearly related. Use the predicted output to approximate the desired output \(y(t)\);

[0073]

[0074] Make the calculated weight matrix satisfy the minimum mean square error of the echo state network. The objective equation is:

[0075]

[0076] Furthermore, it can be reduced to the form:

[0077] \(W\) out =(X -1 Y) T ;

[0078] To solve this linear regression problem, the method of ridge regression can be adopted as follows:

[0079] W out =(X T X + λI) -1 X T Y;

[0080] where X = [x(m) T , x(x + 1) T , …, x(M) T T , Y = [y(m) T , y(m + 1) T , …, y(M) T T , M is the number of training samples. T represents the matrix transpose operation. The role of λ is to impose a "noise term" on X T X. If λ is properly selected, it can effectively balance the error term and the complexity of the model, and improve the performance of the echo state network.

[0081] Based on the basic structure of the echo state network, due to the memory ability and nonlinear characteristics of the discrete memristor, it can enhance the nonlinear state inside the reservoir, capture more information inside the reservoir. At the same time, the compressive sensing technology is used to reduce the dimension of the reservoir state, reduce data redundancy and computational amount. Finally, the parameters of the discrete memristor model are adaptively optimized by the artemisinin optimization algorithm, so that it has a better prediction effect on the wind speed of the wind turbine.

[0082] Step 3: The network structure of the improved echo state network is as Figure 3 shown. Use the improved echo state network to predict the historical wind speed data of the wind turbine to obtain the wind speed prediction result;

[0083] In the training stage, according to the initialized reservoir state and the model parameters of the discrete memristor, the reservoir connection matrix is obtained. The discrete memristor can be described as:

[0084]

[0085] In the formula, i n is the discrete input current signal, V(n) is the discrete output voltage signal, q n is the internal charge of the memristor, M(q n ) is the memristance value, where the M(·) function will adopt the square function, and α and β are the model parameters of the discrete memristor. Let k = 1;

[0086] Use the discrete memristor to calculate the reservoir matrix. The formula is as follows:

[0087] W res ​​= α + β·x(t - 1) 2 ;

[0088] For each moment input u(t) of the data set, the reservoir updates the corresponding state, and the state update equation is:

[0089] x(t) = f(W in ·u(t) + W res x(t - 1));

[0090] where u(n) ∈ R k×1 is the input data composed of k input neurons; k is 1, and x(n) ∈ R l×1 represents the vector composed of the activated neurons in the reservoir, and W in ∈ R l×k is the input matrix with the initial value range between [-0.5, 0.5], and W res ∈ R l×l is the reservoir matrix, and f represents the internal neuron activation function, usually the tanh function or the sigmod function.

[0091] After each state update, the compressed sensing technology is used to reduce the dimension of the reservoir state. Compressed sensing allows the signal to be effectively sampled and reconstructed at a sampling rate lower than the traditional one. The steps of compressed sensing processing are sparse representation, construction of the observation matrix, and signal recovery representation in sequence. The focus of the combination in the present invention is sparse representation and construction of the observation matrix, and the principles of these two methods are as follows;

[0092] Sparse representation: Transform the target signal into a sparse transform domain as sparse as possible. Suppose there is a signal vector x ∈ R N×1 , and its K orthogonal basis vectors ψ i , i = 1, …, m. The set of these orthogonal bases is called a complete orthogonal basis ψ T = [ψ1, ψ2, …, ψ K . At this time, the vector x can be decomposed as:

[0093]

[0094] If x has only K (K << N) non-zero coefficients α i on the orthogonal basis ψ, then ψ is called the sparse basis of the signal x, and x is K-sparse;

[0095] Construction of the observation matrix: The purpose of the observation matrix is to sample M observation values and ensure that the original signal x of length N or the sparse vector α in the sparse basis can be reconstructed from them. The observation process is to project the sparse vector using M row vectors of the M × N observation matrix to obtain M observation values, that is

[0096] where α = ψ T x;

[0097] In the present invention, Fourier basis is used for sparse representation, and a random Bernoulli measurement matrix is used;

[0098] Record the compressed reservoir state.

[0099] Step 4: Based on Step 3, use the artemisinin algorithm to optimize the parameters of the discrete memristor and optimize its discrete memristor model. Use the normalized root mean square error in the training stage as the fitness function of the artemisinin algorithm to obtain the optimal parameters α and β of the discrete memristor model.

[0100] Step 5: Based on the optimal parameters obtained in Step 4, use the optimal parameters to calculate W according to the formula out .

[0101] Finally, use the obtained coefficients to participate in the prediction and apply the prediction model to the wind turbine wind speed prediction process.

[0102] The output equation of the CSMNESN model (prediction model) is the same as that of the ESN:

[0103]

[0104] Figure 4(a) is the prediction diagram of the wind turbine wind speed data, and Figure 4(b) is the error diagram of the wind turbine wind speed data. Four comparison models are adopted. CSMNESN is the model proposed in this application. It can be seen from Figure 4(a) that the model proposed in this application has the highest prediction accuracy and fitting effect on the wind turbine wind speed data; it can be obtained from Figure 4(b) that the model proposed in this application has the smallest prediction error for the wind turbine wind speed data prediction, indicating higher prediction accuracy.

[0105] The present invention obtains a prediction task and wind turbine wind speed data; in an echo state network, a discrete memristor is introduced to replace the random initialization of the reservoir matrix, and compressive sensing is introduced during the reservoir state update process. Combining with the artemisinin algorithm to optimize the parameters of the discrete memristor model, a prediction model (CSMNESN) is obtained; based on the prediction task, the wind speed data of the wind turbine is input into the prediction model to obtain the predicted wind speed. That is, based on the basic structure of the echo state network, a discrete memristor is introduced, which can capture complex patterns in the signal and help the reservoir better retain important early information, so as to more accurately analyze long-term trends. Compressive sensing can remove redundant information, and the artemisinin algorithm is a meta-heuristic algorithm that can effectively solve the optimization dilemma and improve the prediction accuracy. Using the prediction model to predict wind turbine data improves the accuracy of wind speed prediction, and the generalization ability of this model is better.

[0106] In addition, the prediction model described in the present invention can also be used for the true data set of sunspot numbers, the chaotic data set of the Logistic system, and the chaotic data set of the Lorenz system. For the chaotic series, first perform maximum-minimum processing, and then put it into the input layer of the model for training; for the true data set, normalize the data and then put it into the model for training.

[0107] The sunspot number is a set of observational data on sunspot activities, usually including information such as the position, size, number, and intensity of sunspots. These data are of great significance for studying solar activities, predicting changes in the number of sunspots, and understanding their impact on the Earth.

[0108] Figure 5(a) is the prediction graph of the sunspot number, and Figure 5(b) is the error graph of the sunspot number prediction. Four comparison models are adopted, and CSMNESN is the model proposed by the present invention. It can be obtained from Figure 5(a) that the model proposed by the present invention has the highest prediction accuracy and fitting effect on the sunspot number time series; it can be obtained from Figure 5(b) that the model proposed by the present invention has the smallest prediction error for the sunspot number prediction, indicating higher prediction accuracy.

[0109] Logistic system: X(t + 1) = uX(t)*(1 - X(t)), u ∈ [0, 4], X(0) ∈ [0, 1], where the parameter u = 3.58 and the initial value of the iteration X(0) is 0.005. Iterate n times to obtain the Logistic chaotic sequence;

[0110] Figure 6(a) is the prediction graph of the Logistic system, and Figure 6(b) is the error graph of the Logistic system prediction. Four comparison models are used, and CSMNESN is the model proposed by the present invention. It can be obtained from Figure 6(a) that the model proposed by the present invention has the highest prediction accuracy and fitting effect on the Logistic system; from Figure 6(b), it can be obtained that the model proposed by the present invention has the smallest prediction error for the Logistic system prediction, indicating higher prediction accuracy.

[0111] The Lorenz chaotic system is a typical chaotic system:

[0112] Lorenz system:

[0113] where a = 10, b = 28, c = 8 / 3, and the system presents a chaotic state;

[0114] Figure 7(a), Figure 8(a), and Figure 9(a) are the prediction graphs of the x-axis, y-axis, and z-axis of the Lorenz system, and Figure 7(b), Figure 8(b), and Figure 9(b) are the error graphs of the x-axis, y-axis, and z-axis predictions of the Lorenz system. Four comparison models are used, and CSMNESN is the model proposed by the present invention. It can be obtained from Figure 7(a), Figure 8(a), and Figure 9(a) that the model proposed by the present invention has the highest prediction accuracy and fitting effect on the three axes of the Lorenz system; from Figure 7(b), Figure 8(b), and Figure 9(b), it can be obtained that the model proposed by the present invention has the smallest prediction error for the three axes of the Lorenz system prediction, indicating higher prediction accuracy.

[0115] Embodiment 2

[0116] This embodiment provides a data prediction system based on a memristive echo state network.

[0117] A data prediction system based on a memristive echo state network, comprising:

[0118] A prediction module, which is configured to: based on the acquired historical wind speed data set of the wind turbine, use an improved echo state network to obtain wind speed data;

[0119] Wherein, the improved echo state network includes: calculating a reservoir connection matrix according to the initialized reservoir state and the model parameters of the discrete memristor; updating the reservoir state according to the reservoir connection matrix for the input data at each moment of the historical wind speed data set of the wind turbine, and solving the wind speed data.

[0120] Specifically, the prediction module is specifically configured to:

[0121] Construct a memristor-based echo state network, initialize the relevant parameters before training the model for the least squares solution, and initialize the node values of the reservoir pool;

[0122] Based on the prediction task, input the historical wind speed dataset of the wind turbine into the input layer of the memristor-based echo state network to start model training. During the training process, update the reservoir pool nodes, and finally solve the wind speed data by the least squares solution;

[0123] During the training process, use the discrete memristor model to replace the randomly initialized reservoir pool matrix of the basic echo state network. The discrete memristor model uses the previously updated reservoir pool state for solution each time, and then updates the reservoir pool nodes. And add a compressive sensing layer to reduce the dimension of the reservoir pool state. This process is looped n times, and the artemisinin algorithm is used to optimize its discrete memristor model. Use the normalized root mean square error in the training stage as the fitness function of the artemisinin algorithm to obtain the optimal discrete memristor model parameters, as well as the corresponding reservoir pool state set and output weight matrix. In the prediction stage, use the optimal parameters and output weight matrix obtained in the training stage for prediction. In the training stage, the update of the reservoir pool weights includes:

[0124] W res =α + β·x(t - 1) 2 ;

[0125] The update formula of the reservoir pool state includes:

[0126] x(t)=f(W in u(t)+W res x(t - 1));

[0127] Among them, u(t) represents the historical wind speed dataset of the wind turbine, W in represents the input matrix, W res represents the reservoir pool matrix, x(t - 1) represents the reservoir pool state at the previous moment, f represents the internal neuron activation function, and both α and β represent the model parameters of the discrete memristor.

[0128] Use compressive sensing technology to reduce the dimension of its reservoir pool state, collect the compressed reservoir pool states into the set X, and find out the corresponding output weight matrix, that is, the wind speed data.

[0129] Example 3

[0130] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the data prediction method of the memristor-based echo state network as described in the above Example 1.

[0131] Example 4

[0132] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the data prediction method of the memristor-based echo state network as described in the above-mentioned Embodiment 1.

[0133] Embodiment 5

[0134] This embodiment provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the data prediction method of the memristor-based echo state network as described in the above-mentioned Embodiment 1.

[0135] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories and optical memories, etc.) containing computer-usable program codes.

[0136] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0137] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the functions specified in one process or multiple processes and / or blocks Figure 1 one process or multiple processes and / or blocks Figure 1 steps for the functions specified in one block or multiple blocks.

[0139] 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 completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0140] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data prediction method based on a memristor echo state network, characterized in that: include: Based on the historical wind speed data set of the wind turbine, the wind speed data is obtained by using an improved echo state network; The improved echo state network includes: calculating a reserve pool connection matrix according to the initialized reserve pool state and the model parameters of the discrete memristor; updating the reserve pool state according to the reserve pool connection matrix and solving the wind speed data for the input data of the historical wind speed data set of the wind turbine at each moment.

2. The data prediction method based on the memristor echo state network according to claim 1, characterized in that: After each update of the reserve pool state, compressed sensing processing is used to reduce the dimension of the reserve pool state to obtain the compressed reserve pool state.

3. The data prediction method based on the memristor echo state network according to claim 2, characterized in that: The compressed sensing process includes: transforming the reserve pool state into a sparse transform domain to obtain a sparse vector; constructing a measurement matrix based on the sparse vector; and restoring the measurement matrix to obtain a compressed reserve pool state.

4. The data prediction method based on memristor echo state network according to claim 1, characterized in that: In the improved echo state network training process, the artemisinin algorithm is used to optimize the model parameters of the discrete memristor, and the normalized root mean square error in the training phase is used as the fitness function of the artemisinin algorithm to obtain the optimal discrete memristor model parameters. According to the optimal discrete memristor model parameters, the reserve pool state set and output weight matrix corresponding to the model parameters are calculated.

5. The data prediction method based on memristor echo state network according to claim 1, characterized in that: The method for solving the wind speed data includes: solving the wind speed data using a least squares solution according to an updated reserve tank state.

6. The data prediction method based on memristor echo state network according to claim 1, characterized in that: The update of the reserve pool state is expressed by the following formula: x(t)=f(W in u(t)+W res x(t-1)); W res =α+β·x(t-1) 2 ; Where u(t) represents the data affecting the wind speed of the wind turbine, W in represents the input matrix, W res represents the reservoir matrix, x(t-1) represents the reservoir state at the previous moment, f represents the internal neuron activation function, and α and β represent the model parameters of the discrete memristor.

7. A data prediction system based on a memristor echo state network, characterized in that: include: A prediction module is configured to: obtain wind speed data by using an improved echo state network based on an acquired historical wind speed data set of a wind turbine; The improved echo state network includes: calculating a reserve pool connection matrix according to the initialized reserve pool state and the model parameters of the discrete memristor; updating the reserve pool state according to the reserve pool connection matrix and solving the wind speed data for the input data of the historical wind speed data set of the wind turbine at each moment.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the data prediction method based on the echo state network based on memristor are implemented as described in any one of claims 1 to 6.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the data prediction method based on the echo state network based on memristor are implemented as described in any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the steps in the data prediction method based on a memristor echo state network according to any one of claims 1 to 6.

Images