Sequence prediction method and system based on compressed sensing pooling echo state network
By introducing a pooling layer and a compressed sensing layer into the echo state network, the state of the reserve pool nodes is optimized, which solves the problems of high computational cost and performance degradation caused by redundant nodes and achieves more efficient prediction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-08
- Publication Date
- 2026-04-10
AI Technical Summary
The existing echo state network model has redundant nodes, which leads to high computational cost and performance degradation. Furthermore, existing compressed sensing methods may cause data feature loss when applied to ESN.
In the echo state network, a pooling layer and a compressed sensing layer are introduced. The pooling algorithm is used to optimize the state of the reserve pool nodes, and the compressed sensing is used for sparse transformation and random subsampling to reduce redundant nodes and improve node activity performance.
This effectively reduces redundant nodes, improves the computational efficiency and accuracy of the model, and enhances the model's predictive performance.
Smart Images
Figure CN115169740B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of network information prediction, and particularly relates to a sequence prediction method and system of a pooling echo state network based on compressed sensing. BACKGROUND
[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.
[0003] Today, with the rapid development of machine learning and deep learning technologies and applications, echo state network models (ESN) are increasingly widely used in various industries, such as weather prediction, stock prediction, natural language processing, wind prediction, and chaotic sequence prediction; compared with traditional neural networks, echo state networks use a reservoir pool mode instead of a large number of hidden layers included in the original neural network model, and the prediction process of ESN is more focused on the change of data dimension, and unlike the form of back propagation to update parameters, most parameters of ESN are randomly initialized within a limited range, and other parameters are determined in the form of least squares solution, and the parameter values will not change after being determined during the training or prediction process. The dimension of data will also be increased to high dimension after entering the reservoir pool from the input layer, and then decreased to the specified dimension according to the task requirements, and the prediction process is completed.
[0004] The reservoir pool mainly includes a group of reservoir pool nodes with a graph structure and an activation function, and under the action of the nodes and the activation function, the entire echo state network will have strong nonlinear fitting capability. Since the prediction principle of the network is based on the division of node values at different times on the time axis, and time series are extremely related to time, the echo state network has strong prediction performance for various forms of time series data prediction. In addition, the reservoir pool involves a large number of node relationships, and the model may contain redundant nodes, which will affect the performance of the model. Moreover, the calculation amount of the training process and the prediction process of the model is huge, and directly reducing the number of nodes of the reservoir pool will also lead to the decline of the performance of the model, so finding a suitable optimization strategy and improving the calculation efficiency is the key to further optimization of the echo state network.
[0005] The current methods for reducing the number of reservoir pool nodes include compressed sensing and neural network Dropout layer, but the improper way of sparse representation and random sampling of compressed sensing applied to ESN will cause the loss of data features during sparse representation and sampling, and directly reducing the number of reservoir pool nodes will also lead to the deterioration of the prediction effect of the model. Therefore, finding a better update method for the reservoir pool nodes and the representation form of the nodes is the key to improving the performance of the model. SUMMARY
[0006] In order to overcome the above-mentioned deficiencies of the prior art, solve the problems of how to more effectively reduce the node redundancy in the echo state network model, improve the model accuracy, computational efficiency and the like, the present application provides a sequence prediction method and system of a pool compressed sensing echo state network (PCSESN) based on compressed sensing, adds a pooling algorithm and compressed sensing to the pool echo state network model, effectively reduces the redundant nodes, improves the node activity performance, improves the model performance of the ESN, so that the reservoir pool can be active while reducing the amount of calculation, and improves the accuracy and running efficiency of model calculation.
[0007] In order to achieve the above-mentioned purpose, one or more embodiments of the present application provide the following technical solutions:
[0008] The present application provides a sequence prediction method of a pool compressed sensing echo state network based on compressed sensing in a first aspect of the present application;
[0009] The sequence prediction method of the pool compressed sensing echo state network based on compressed sensing comprises:
[0010] Obtaining sequence data required by a prediction task;
[0011] Constructing a pool compressed sensing echo state network model based on compressed sensing, and training the network model;
[0012] Inputting the obtained sequence data into the trained pool compressed sensing echo state network model to obtain a prediction result;
[0013] Wherein, a pooling layer and a compressed sensing layer are added to the reservoir pool of the pool compressed sensing echo state network model, the pooling layer is used to readjust the weight of the node state of the reservoir pool, and the compressed sensing layer is used to perform sparse transformation and random subsampling on the node.
[0014] Further, the pool compressed sensing echo state network model is based on a classical echo state network model and consists of three parts: an input layer, a reservoir pool and an output layer.
[0015] The reservoir pool is improved and divided into three layers: a normalization layer, a pooling layer and a compressed sensing layer.
[0016] Further, the normalization layer performs normalization processing on the input data, so that the input data value is mapped to between 0 and 1.
[0017] Further, the pooling layer optimizes and updates the nodes of the reservoir pool, adjusts the state of the inactive or redundant nodes in the reservoir pool, and obtains a more optimal node state.
[0018] Further, the connection parameter value between the output layer and the reservoir layer is determined in the form of a least square solution.
[0019] Further, the training process of the echo state network model mainly includes two stages: a sampling stage and a weight calculation stage.
[0020] Further, the sampling stage is used to construct a system state matrix and a sample data vector.
[0021] The weight calculation stage calculates the output connection weight matrix of the echo state network model.
[0022] The second aspect of the present application provides a sequence prediction system of a compressed sensing based pooling echo state network.
[0023] The sequence prediction system of the compressed sensing based pooling echo state network comprises a data acquisition module, a model construction module and a result prediction module.
[0024] The data acquisition module is configured to acquire sequence data required by a prediction task.
[0025] The model construction module is configured to construct a compressed sensing based pooling echo state network model and train the network model.
[0026] The result prediction module is configured to input the acquired sequence data into the trained pooling echo state network model to obtain a prediction result.
[0027] In the reservoir pool of the pooling echo state network model, a pooling layer and a compressed sensing layer are added, the pooling layer is used to readjust the weight of the reservoir pool node state, and the compressed sensing layer is used to perform sparse transformation and random subsampling on the node.
[0028] The third aspect of the present application provides a computer readable storage medium, which stores a program, and the program is executed by a processor to realize the steps of the sequence prediction method of the compressed sensing based pooling echo state network according to the first aspect of the present application.
[0029] The fourth aspect of the present application provides an electronic device, which comprises a memory, a processor and a program stored in the memory and executable on the processor, and the processor executes the program to realize the steps of the sequence prediction method of the compressed sensing based pooling echo state network according to the first aspect of the present application.
[0030] The above one or more technical solutions have the following beneficial effects:
[0031] The application adds a normalization processing layer in the reserve pool of the network, proposes a new node value updating algorithm of the reserve pool, a pooling algorithm, and uses a new sparse representation and random sampling method of compressive sensing to extract node information, puts the sequence data into the model based on the predicted task, and obtains a prediction result, that is, based on the mechanism of compressive sensing and the pooling algorithm, a mechanism capable of effectively reducing redundant nodes and improving node activity performance is proposed.
[0032] The pooling algorithm proposed in the application does not reduce the number of nodes, but adds noise values on the basis of the traditional pooling method, adjusts the node state of the reserve pool node, makes the redundant or inactive node adjust the state, and thus the prediction mode of the reserve pool can be closer to the chaotic characteristics of the chaotic sequence to be predicted by the model, and finally the model achieves better prediction effect.
[0033] The new sparse representation and random sampling of compressive sensing can more accurately and efficiently extract the main information of the reserve pool node than the original sampling method, can avoid information leakage and loss to the greatest extent, effectively improve the model performance of the ESN, and make the reserve pool active while reducing the calculation amount, and improve the accuracy and operation efficiency of model calculation.
[0034] The advantages of the additional aspects of the application will be partially given in the following description, partially become obvious from the following description, or be known by the practice of the application. BRIEF DESCRIPTION OF DRAWINGS
[0035] The drawings accompanying the specification of the application form a part of the application and serve to provide further understanding of the application, the illustrative embodiments of the application and their description serve to explain the application without constituting improper limitations on the application.
[0036] Figure 1 The method flowchart of the first embodiment.
[0037] Figure 2 The structure schematic diagram of the classical pooling echo state network model in the first embodiment.
[0038] Figure 3 The structure schematic diagram of the improved pooling echo state network model in the second embodiment. DETAILED DESCRIPTION
[0039] The application will be further described below in combination with the drawings and embodiments.
[0040] It should be pointed out that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.
[0041] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.
[0042] Embodiment one
[0043] The embodiment discloses a sequence prediction method of a pooling echo state network based on compressed sensing;
[0044] As Figure 1 shown, the sequence prediction method of the pooling echo state network based on compressed sensing comprises:
[0045] Step S1 acquires sequence data required by a prediction task;
[0046] The pooling echo state network is applied to jobs in various scenarios, such as stock prediction, wind strength prediction, signal trend prediction, industrial dynamic data soft measurement, time series classification, and pattern learning, so the prediction task and the sequence data, such as the stock prediction task and the stock sequence data, the chaotic sequence prediction task and the chaotic sequence data, and the like, are acquired. The stock data can be acquired through a website, and the chaotic sequence data is generated through a corresponding chaotic sequence generation formula and in combination with specific numerical values.
[0047] Step S2 constructs a pooling echo state network model based on compressed sensing, and trains the network model;
[0048] Step S3 inputs the acquired sequence data into the trained pooling echo state network model to obtain a prediction result;
[0049] The construction of the pooling echo state network model is specifically as follows:
[0050] (1) Initialization: randomly initialize input weights and graph structure connection weights inside the reservoir, and initialize initial time node values of the reservoir.
[0051] (2) Normalization: after the initialization of the nodes inside the reservoir is completed, the sequence data required by the task is put into the model, the data enters the normalization layer inside the reservoir for normalization processing, so that the data value distribution is between 0 and 1, the dimension and the data value range are eliminated, and the bad influence on the model prediction is eliminated, and the state update formula of the reservoir node is established in contact with the reservoir node.
[0052] (3) Pooling: The nodes in the reserve pool enter the pooling layer inside the reserve pool to update and optimize their state. The state of inactive or redundant nodes in the reserve pool is adjusted to obtain a better node state.
[0053] After the pooling layer is updated, the number of nodes in the reserve pool does not change. Instead, the pooling algorithm optimizes the node state of the original reserve pool, enabling the echo state network model to make better predictions.
[0054] (4) Compressed sensing: The updated and optimized reserve pool nodes enter the compressed sensing layer. The compressed sensing layer performs sparse transformation and random subsampling on the nodes. After passing through the compressed sensing layer, the number of reserve pool nodes is indirectly reduced, but the original node information can still be completely described.
[0055] (5) Output: After the reservoir node is trained, the connection parameter values between the output layer and the reservoir layer are determined. These parameters are generally determined by the least squares solution.
[0056] This invention is based on a special type of recurrent neural network—the echo state network model, such as the classic ESN network. Figure 2 As shown, the network consists of three parts: an input layer, a storage pool (hidden layer), and an output layer. It is assumed that the network has K input units, N storage pool internal processing units, and L output units.
[0057] At time t, the values of the input layer, reservoir, and output layer of the network are as follows:
[0058] u(t)=[u1(t),u2(t),······,u K (t)] T
[0059] x(t)=[x1(t),x2(t),······,x N (t)] T
[0060] y(t)=[y1(t),y2(t),·······,y L (t)] T
[0061] For each input u(t) at time step, the reservoir updates its state accordingly. The state update formula is:
[0062] x(t)=f(W in ·u(t)+Wx(t-1)(
[0063] where u(t) is the input at current time, x(t) is the state of the reservoir at current time, x(t-1) is the state of the reservoir at last time, f(·) represents the activation function of the neuron in the reservoir, which is usually the hyperbolic tangent function tanh().
[0064] The output of the ESN is given by:
[0065] y(t) = f out (W out ·x(t))
[0066] f out represents the output function, in general, the output layer is linear, i.e. f out is the identity function.
[0067] Figure 2 The input-to-state connection weight matrix W in (N x K) represents the connection between the input layer and the reservoir, the state-to-state connection weight matrix W(N x N) represents the connection between the neurons in the reservoir, and the state-to-output connection weight matrix W out (L x N) represents the connection weight between the reservoir and the output layer. Among them, W in and W are randomly generated in the initialization stage and remain unchanged once generated. While W out needs to be trained according to the input and output data of the system, because there is a linear relationship between the state variable and the output, so usually the output layer connection weight only needs to be obtained by solving the linear regression problem, that is:
[0068] W out ·x = y
[0069] The training process of the echo state network model mainly includes two stages: sampling stage and weight calculation stage.
[0070] Sampling stage: first, the initial state of the network is randomly selected, usually the initial state of the network is 0, i.e. x(0) = 0; through the input connection weight matrix W in The training sample u(n)(n = 1, 2, ···, M) is added to the reservoir, and the calculation and collection of the state variable x(t) and the output y(t) are completed in turn according to the corresponding equation. It is assumed that the system state is collected from a certain time m, and the vector (x1(i), x2(i), ···, x N (i))(i = m, m + 1, ···, M) is constructed as a row matrix X(M-m+1, N), and the corresponding sample data y(n) is also collected and constructed as a column vector Y(M-m+1, 1), which is used for subsequent calculation of the output connection weight matrix.
[0071] Weight calculation stage: according to the state matrix and output collected in the sampling stage, the output connection weight matrix W of the echo state network model is calculated out The state variable x(t) and the predicted output are linearly managed, and the predicted output approximates the expected output y(t):
[0072]
[0073] The calculated weight matrix satisfies the minimum mean square error of the echo state network model, and the target equation is:
[0074]
[0075] which can be further summarized as:
[0076] W out =(X -1 Y) T
[0077] In order to solve this linear regression problem, the ridge regression method can be used, as follows:
[0078] W out =(X T X+C -1 I) -1 X T Y
[0079] Where, X=[x(m) T ,x(m+1) T ,···,x(M) T ] T ,Y=[y(m) T ,y(m+1) T ,···,yMT]T,p is the starting time of sampling, M is the number of training samples; T represents the matrix transpose operation; The role of C is to impose a regularization term coefficient on X T X, if C is properly selected, the error term and the complexity of the model can be effectively balanced, and the performance of the echo state network can be improved.
[0080] The prediction task inputs the sequence data into the pooling echo state network model to obtain a prediction result, specifically:
[0081] First, the prediction task and the time series data are obtained, and the training set and the test set are divided according to the data size.
[0082] Then the pooling echo state network model is established, including: normalization layer, pooling layer and compressed sensing layer, an echo state network model which can effectively optimize node state information and reduce the number of nodes is established; the input connection weight matrix and the reserve pool internal connection weight matrix are initialized. The network is trained, and the output weight matrix is solved by using the least square method.
[0083] Finally, based on the prediction task, the time series data is input into the pooling echo state network model to obtain the prediction result.
[0084] For example, the prediction task V and the time or chaotic sequence data P are obtained, the pooling echo state network model based on compressed sensing is constructed, the input connection weight matrix and the reserve pool internal connection weight matrix are randomly initialized, and it is ensured that the spectral radius (the absolute value of the largest eigenvalue of the matrix) of the reserve pool internal connection weight matrix is less than 1, the appropriate pooling layer parameters are selected, and the pooling echo state network model A is obtained; as shown in the following figure, the reserve pool of the pooling echo state network model based on compressed sensing is different from that of the classical echo state network model, the reserve pool of the echo state network model based on compressed sensing is divided into three layers, which are normalization layer, pooling layer and compressed sensing layer. The normalization layer performs normalization processing on the input data, so that the value of the input data is mapped to 0 to 1, and the influence of the dimension and the data value range on the model prediction is eliminated; the pooling layer optimizes and updates the nodes of the original reserve pool through the pooling algorithm, adjusts the state of the inactive or redundant nodes in the reserve pool, and obtains better node state; the compressed sensing layer uses the compressed sensing technology to compress and feature the original reserve pool nodes, extracts the information that can describe the state of the original reserve pool nodes through sparse transformation and random subsampling, reduces the computational complexity of the model, and improves the prediction accuracy of the model. Figure 3
[0085] In the training process, the input connection weight and the reserve pool internal connection weight of the ESN remain unchanged, and only the output connection weight is trained; finally, based on the prediction task V, the time series data P is input into the pooling echo state network model A to obtain the prediction result S, the prediction error is calculated, and the prediction is completed.
[0086] The input data is normalized by the normalization layer, so that the data is mapped to 0 to 1, including:
[0087] The input data is normalized, and the formula is as follows:
[0088]
[0089] Where, DATA_MAX and DATA_MIN are the maximum and minimum values of the data set respectively.
[0090] In the above formula, u represents the input layer data, y represents the label value corresponding to u, and after normalization processing, the values of u and y are mapped to the value range of 0 to 1. After normalization processing, the influence of the dimension and value range of the input data on the model prediction is eliminated, which improves the training speed of the model and improves the prediction accuracy of the model.
[0091] The node state is updated and optimized by the pooling layer, and the state of the inactive or redundant node is adjusted to obtain better node information. The original reserve pool nodes are sequentially grabbed according to the set number of nodes cross and the interval step of each grabbing, and if the number of nodes is insufficient when grabbing, the node with a state of 0 is supplemented.
[0092] The state of each group of grabbed nodes is enhanced using the pooling algorithm. The method adds a certain range of noise value to the average pooling:
[0093] xa=mean(xc)*r
[0094] Where xa is the node after pooling, xc is a group of node information grabbed, mean() is an average value function, and r is a parameter for optimizing the node state, and the value is 10 times the learning rate.
[0095] The reserve pool nodes are compressed and characterized by the compression sensing layer. Through sparse transformation and random subsampling, information describing the state of the original reserve pool nodes is extracted to improve the prediction performance of the ESN model, including:
[0096] 1) Signal sparse representation. The node information after the pooling layer is sparsely represented by discrete Fourier transform. The discrete Fourier transform formula is:
[0097]
[0098] Where, f n is a finite-length sequence of length M, and N is the number of data points.
[0099] 2) Design the observation matrix. The compressed signal is observed using a random Gaussian measurement matrix.
[0100]
[0101] Where, Φ is a Gaussian random measurement matrix, RN is the number of compressed reserve pool nodes, N is the number of original reserve pool nodes before compression, cross is the number of nodes grabbed by the pooling operation each time, and the randn(x, y) function generates an x y matrix composed of random arrays.
[0102] Further comprising training the pooling echo state network model, comprising:
[0103] Obtaining a sample set, dividing the sample data set into a training set and a test set, initializing all parameter information in the pooling echo state network model: initializing the number of reservoir nodes N and the number of reservoir nodes RN after passing through the compressed sensing layer, initializing the reservoir self-connection weight W according to N, spectral radius and sparsity parameter, input weight W in Randomly initializing, initializing the pooling parameters cross and step.
[0104] The data in the training set is input into the reservoir through the input layer, and the state information of the nodes after passing through the compressed sensing layer and the corresponding labels are recorded.
[0105] The output weight matrix W is calculated by the ridge regression method according to the recorded state information of the nodes after passing through the compressed sensing layer and the corresponding labels. out .
[0106] Finally, the trained pooling echo state network model is obtained for actual time series prediction process, and the prediction effect of the prediction model in various time series is shown in Table 1:
[0107] Table 1. Model effect of pooling echo state network based on compressed sensing in time series prediction.
[0108]
[0109] Among them, the data amount of chaotic sequence and chaotic system is 10000000, the closing price of 200 stocks is intercepted, and there are 828799 data.
[0110] In summary, by obtaining a prediction task and time series data; constructing a pooling echo state network model based on compressed sensing; through the normalization layer, the input data is normalized to eliminate the adverse effects of dimension and data value range on model prediction; through the pooling layer, the state of the original reservoir pool is adjusted, and the original reservoir pool node is updated to obtain a better node state; through the compressed sensing layer, the original reservoir pool node is compressed and represented, reducing the number of original reservoir pool nodes and improving the training speed of the model. Based on the prediction task, the time series data is input into the network model to obtain a prediction result.
[0111] Embodiment two
[0112] The embodiment discloses a sequence prediction system of a pooling echo state network based on compressed sensing;
[0113] The sequence prediction system based on the compressed sensing pooling echo state network comprises a data acquisition module, a model construction module and a result prediction module.
[0114] The data acquisition module is configured to acquire sequence data required by a prediction task.
[0115] The model construction module is configured to construct a compressed sensing pooling echo state network model and train the network model.
[0116] The result prediction module is configured to input the acquired sequence data into the trained pooling echo state network model to obtain a prediction result.
[0117] In the reservoir pool of the pooling echo state network model, a pooling layer and a compressed sensing layer are added, the pooling layer is used to readjust the weight of the reservoir pool node state, and the compressed sensing layer is used to perform sparse transformation and random subsampling on the node.
[0118] Embodiment three
[0119] The purpose of this embodiment is to provide a computer-readable storage medium.
[0120] The computer-readable storage medium stores a computer program, and the program is executed by a processor to implement the steps in the sequence prediction method based on the compressed sensing pooling echo state network according to Embodiment 1 of the present disclosure.
[0121] Embodiment four
[0122] The purpose of this embodiment is to provide an electronic device.
[0123] The electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, and the processor executes the program to implement the steps in the sequence prediction method based on the compressed sensing pooling echo state network according to Embodiment 1 of the present disclosure.
[0124] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A method for sequence prediction based on a pooling echo state network of compressed sensing, characterized in that, The method comprises the following steps: obtaining sequence data required by a prediction task; constructing a pooling echo state network model based on compressed sensing, and training the network model; inputting the obtained sequence data into the trained pooling echo state network model to obtain a prediction result; wherein, in the reservoir pool of the pooling echo state network model, a pooling layer and a compressed sensing layer are added, the pooling layer is used to readjust the weight of the state of the reservoir pool node, and the compressed sensing layer is used to perform sparse transformation and random subsampling on the node; the pooling layer optimizes and updates the node of the reservoir pool, adjusts the state of the inactive or redundant node in the reservoir pool, and obtains a more optimal node state, including sequentially grabbing the original reservoir pool node according to the set number of original reservoir pool nodes cross and the interval step of each grabbing, and supplementing a node with a state of 0 if the number of original reservoir pool nodes is insufficient when grabbing in sequence; a noise value is added to each group of grabbed nodes based on average pooling using a pooling algorithm; after the update of the pooling layer, the number of reservoir pool nodes does not change. 2.The sequence prediction method of the compressed-sensing-based pooling echo state network according to claim 1, wherein, The pooling echo state network model is based on a classical echo state network model and consists of three parts: an input layer, a reservoir pool and an output layer. The reservoir pool is improved and divided into three layers: a normalization layer, a pooling layer and a compressed sensing layer. 3.The sequence prediction method of the compressed-sensing-based pooling echo state network according to claim 2, wherein, The normalization layer normalizes the input data so that the input data value is mapped to between 0 and 1. 4.The sequence prediction method of the compressed-sensing-based pooling echo state network according to claim 2, wherein, The connection parameter value between the output layer and the reservoir pool is determined in the form of least squares solution. 5.The sequence prediction method of the compressed-sensing-based pooling echo state network according to claim 1, wherein, The training process of the echo state network model mainly includes two stages: a sampling stage and a weight calculation stage. 6.The sequence prediction method of the compressed-sensing-based pooling echo state network according to claim 5, wherein, The sampling stage is used to construct a system state matrix and a sample data vector. The weight calculation stage calculates the output connection weight matrix of the echo state network model.
7. A system for sequence prediction based on a compressed sensing based pooling echo state network characterized by: The method comprises a data acquisition module, a model construction module and a result prediction module. The data acquisition module is configured to obtain sequence data required by a prediction task. The model construction module is configured to construct a pooling echo state network model based on compressed sensing, and train the network model. The result prediction module is configured to input the obtained sequence data into the trained pooling echo state network model to obtain a prediction result. wherein, in the reservoir pool of the pooling echo state network model, a pooling layer and a compressed sensing layer are added, the pooling layer is used to readjust the weight of the state of the reservoir pool node, and the compressed sensing layer is used to perform sparse transformation and random subsampling on the node; the pooling layer optimizes and updates the node of the reservoir pool, adjusts the state of the inactive or redundant node in the reservoir pool, and obtains a more optimal node state, including sequentially grabbing the original reservoir pool node according to the set number of original reservoir pool nodes cross and the interval step of each grabbing, and supplementing a node with a state of 0 if the number of original reservoir pool nodes is insufficient when grabbing in sequence; a noise value is added to each group of grabbed nodes based on average pooling using a pooling algorithm; after the update of the pooling layer, the number of reservoir pool nodes does not change. The pooling echo state network model is based on a classical echo state network model and consists of three parts: an input layer, a reservoir pool and an output layer. The reservoir pool is improved and divided into three layers: a normalization layer, a pooling layer and a compressed sensing layer. The normalization layer normalizes the input data so that the input data value is mapped to between 0 and 1. The connection parameter value between the output layer and the reservoir pool is determined in the form of least squares solution. The training process of the echo state network model mainly includes two stages: a sampling stage and a weight calculation stage. The sampling stage is used to construct a system state matrix and a sample data vector. The weight calculation stage calculates the output connection weight matrix of the echo state network model.
8. A computer-readable storage medium having stored thereon a program, characterized in that, The program, when executed by the processor, implements the steps in the sequence prediction method of the compressive sensing based pooling echo state network as claimed in any one of claims 1-6.
9. An electronic device comprising a memory, a processor, and a program stored in the memory and capable of running on the processor, characterized by The processor, when executing the program, implements the steps in the sequence prediction method of the compressive sensing based pooling echo state network as claimed in any one of claims 1-6.