Wi-Fi-oriented intelligent performance sensing and predicting method and system

Abstract

The invention discloses a Wi-Fi-oriented intelligent performance perception and prediction method and system, and belongs to the technical field of communication. According to the method, based on collected user signal data, a high-order tensor with a user, a performance index and time as dimensions is constructed, and a low-order matrix is further constructed. And in combination with a deep reinforcement learning algorithm, decomposition prediction is performed on the high-order tensor and the low-order matrix, and a tensor prediction model and a matrix prediction model are constructed. And adjusting the fusion weight through error feedback, and carrying out weighted fusion on the prediction results of the two models to generate a final prediction result. According to the method, the nonlinear and sudden change of the Wi-Fi signal can be effectively captured, the accuracy and generalization ability of network performance perception and prediction are improved, an optimization basis is provided for frequency band switching, load balancing, bandwidth scheduling and the like, and the overall Wi-Fi service quality and user experience are improved.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and more specifically, relates to a Wi-Fi-oriented intelligent performance perception and prediction method and system. Background Art

[0002] With the advancement of wireless network technology and the widespread use of smart devices, Wi-Fi has become an indispensable part of our daily lives. However, Wi-Fi signals are affected by various factors, such as physical obstacles, electronic interference, and user density, resulting in fluctuating performance. Therefore, sensing and predicting future Wi-Fi signal trends can help network administrators more efficiently identify and resolve potential Wi-Fi service issues, effectively improving the overall Wi-Fi user experience. This technology has attracted widespread attention from both academia and industry, and has been applied to smart homes, smart cities, the Industrial Internet of Things, and other fields.

[0003] Currently, research on Wi-Fi signal perception and prediction focuses on propagation model design, link budget analysis, and field data correction. Propagation model design considers the impact of factors such as distance, frequency, antenna height, and the surrounding environment on signal strength. Link budget analysis allows researchers to calculate the maximum path loss between communication systems based on given signal data. Field data correction relies on actual signal strength data obtained from field measurements to adjust the theoretical model to improve its accuracy. However, most of these methods are based on traditional communication theory and mathematical statistics, resulting in low efficiency and practicality in practical applications. Therefore, there is an urgent need to develop more efficient methods and systems to more accurately perceive and predict Wi-Fi signal changes.

[0004] Against this backdrop, artificial intelligence (AI) technology, leveraging its strengths in complex pattern recognition, big data processing, and adaptive learning, has brought new breakthroughs to Wi-Fi signal perception and prediction. However, existing approaches mostly focus on using deep learning methods such as long- and short-term time series prediction and recurrent neural networks for Wi-Fi signal prediction. While these methods demonstrate strong capabilities in processing time series data, they still have limitations when dealing with the sudden changes and nonlinear characteristics of Wi-Fi signals. Furthermore, these methods struggle to effectively exploit deep correlations between data, resulting in low Wi-Fi signal perception and prediction accuracy. Summary of the Invention

[0005] To address the shortcomings of related technologies, the present invention aims to provide an intelligent Wi-Fi performance perception and prediction method and system. This approach aims to address the limitations of existing methods in processing sudden changes in Wi-Fi signals, nonlinear characteristics, and cross-dimensional indicator correlations, which lead to low Wi-Fi performance perception and prediction accuracy.

[0006] To achieve the above objectives, in a first aspect, the present invention provides an intelligent Wi-Fi performance perception and prediction method, comprising: S100: Collect raw signal data from Wi-Fi users, obtain corresponding eigenvalues ​​based on the network performance indicators of each user, and construct high-order tensor data in the user dimension based on the eigenvalues; wherein the eigenvalues ​​are in matrix form, with rows of the matrix representing different network performance indicators and columns representing different sampling time slots; S200, based on the high-order tensor data, normalize each network performance indicator, calculate the cross-time slot weight of each network performance indicator, and integrate the multi-dimensional network performance indicators consisting of multiple network performance indicators corresponding to multiple sampling time slots into one-dimensional standardized data to form a low-order matrix; S300: Inputting the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and inputting the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix; wherein the tensor prediction model and the matrix prediction model are trained based on a deep reinforcement learning algorithm; S400. Obtain a high-order prediction error based on the high-order prediction tensor and the high-order tensor data, obtain a low-order prediction error based on the low-order prediction matrix and the low-order matrix data, dynamically adjust the fusion weight based on the high-order prediction error and the low-order prediction error, and fuse the prediction results of the tensor prediction model and the matrix prediction model by weighted averaging according to the adjusted fusion weight to obtain a fused prediction result.

[0007] Optionally, step S100 includes: S110, capturing signal interaction information between user devices and access points in a Wi-Fi network environment, and collecting original signal data of Wi-Fi users at different network usage periods; S120. For each independent Wi-Fi user, convert different network performance indicator data in the original signal data into a matrix, with the network performance indicator as the row and the sampling time slot as the column, to serve as the eigenvalue matrix of the network performance indicator of the corresponding user; S130, treat each user's eigenvalue matrix as an independent substructure, stack and combine them on the user dimension to form high-order tensor data .

[0008] Optionally, step S200 includes: S210, the high-order tensor data The elements of each eigenvalue matrix in are normalized, and the normalized result is Expressed as:

[0009] in, Represents the column vector in the eigenvalue matrix, the column vector is , Represents a column vector No. elements, is a column vector The maximum value in ; S220, constructing a vector based on the normalized value of each network performance indicator in all sampling time slots , calculate the weight of each network performance indicator on all sampling time slots ;in, Indicates the total number of network performance indicators, represents the total number of sampling time slots, Indicates the The network performance index is Normalized value over sampling time slots; S230, integrate multi-dimensional network performance indicators into one-dimensional standardized data and construct a low-order matrix ; Among them, the low-order matrix Corresponding elements The expression is: , Indicates the The weight of each network performance indicator, Indicates the The network performance index is Normalized value over sampling time slots.

[0010] Optionally, step S300 includes: S310, using tensor decomposition based on deep reinforcement learning algorithm to decompose the high-order tensor data Decompose and construct the model parameters of the initial tensor prediction model to obtain high-order tensor data Minimizing the prediction error between the prediction tensor and the prediction tensor of the initial tensor prediction model is the optimization goal, and the decomposition method of the tensor decomposition is continuously adjusted through strategy iteration to update the model parameters of the initial tensor prediction model; the initial tensor prediction model is trained according to the model parameters of the updated initial tensor prediction model until the network converges or reaches a preset number of iterations, thereby obtaining a tensor prediction model; S320, using matrix decomposition based on deep reinforcement learning algorithm to decompose the low-order matrix, construct model parameters of the initial matrix prediction model, and use the low-order matrix Minimizing the error between the prediction matrix and the initial matrix prediction model is used as the optimization goal, the decomposition method of the matrix decomposition is continuously adjusted through strategy iteration, and the model parameters of the initial matrix prediction model are updated; the initial matrix prediction model is trained according to the model parameters of the updated initial matrix prediction model until the network converges or reaches a preset number of iterations, thereby obtaining a matrix prediction model; S330. Input the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and input the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix.

[0011] Optionally, step S310 includes: S311, initialize the experience replay pool, randomly initialize the main network, and set the parameters of the main network Copy to the target network; input the high-order tensor data into the initialized tensor prediction model to obtain a high-order prediction state set ,make Get Status ; Determine the set of optional actions ,in, Indicates at time Implement the actions selected by tensor decomposition; S312, by probability Randomly select an action to The probability of selecting the optimal action based on the main network's valuation ;in, ; S313, optimal action obtained according to S312 , and get the corresponding reward value , reach the next state , the vector Put them into the experience replay pool respectively; if the experience replay pool is full, go to step S314; if the experience replay pool is not full, then , and proceed to step S312; S314. Randomly extract a batch of vectors from the experience replay pool, and take out the vectors one by one to input into the main network. Find the action corresponding to the maximum Q value in the main network and combine the above vectors to input into the target network to obtain the target network Q value under the new mode. Calculate the loss function by the obtained main network Q value and the target network Q value, and use the gradient descent method to update the parameters of the tensor prediction model. probability; S315, repeat step S314, and set the parameter The target network of the tensor prediction model is copied until the network converges or reaches the preset number of iterations to obtain the optimal tensor decomposition strategy and tensor prediction model.

[0012] Optionally, step S320 includes: S321, initialize the experience replay pool, randomly initialize the main network, and set the parameters of the main network Copy to the target network; input the low-order matrix into the initialization matrix prediction model to obtain a low-order prediction state set ,make Get Status ; Determine the set of optional actions ,in, Indicates at time Implement the actions selected by matrix decomposition; S322, by probability Randomly select an action to The probability of selecting the optimal action based on the main network's valuation ;in, ; S323, the optimal action obtained according to S322 , and get the corresponding reward value , reach the next state , the vector Put them into the experience replay pool respectively; if the experience replay pool is full, go to step S324; if the experience replay pool is not full, then , and proceed to step S322; S324. Randomly extract a batch of vectors from the experience replay pool, and take out the vectors one by one to input into the main network. Find the action corresponding to the maximum Q value in the main network and combine the above vectors to input into the target network to obtain the target network Q value under the new mode. Calculate the loss function by the obtained main network Q value and the target network Q value, and use the gradient descent method to update the parameters of the matrix prediction model. ; S325, repeat step S324, and set the parameter The target network is copied to the matrix prediction model until the network converges or reaches the preset number of iterations, and the optimal matrix decomposition strategy and matrix prediction model are obtained.

[0013] Optionally, step S400 includes: S410: performing a restoration operation on the prediction result of the low-order prediction matrix according to the original dimension information index of the low-order matrix data and the cross-time slot weight to obtain a restored tensor consistent with the high-order tensor structure; S420: Determine the initial fusion weight of the matrix prediction model and the tensor prediction model based on the errors between the matrix prediction model and the tensor prediction model. The initial fusion weight is expressed as:

[0014] in, and The matrix prediction model and the tensor prediction model are respectively The error, is a smoothing coefficient used to avoid division by zero and smooth the error; S430: Normalize the initial fusion weight to obtain a dynamic fusion weight coefficient:

[0015] in, is the influence coefficient, which determines the relative influence between the two model errors; when When , the fusion weight of the model with smaller error in the tensor prediction model and the matrix prediction model is set to be greater than that of the model with larger error; when When the fusion weight of the model with larger error in the tensor prediction model and the matrix prediction model is set to be larger than that of the model with smaller error; the fusion weight coefficient satisfies ; S440, performing weighted fusion on the high-order prediction tensor and the restored tensor obtained from the low-order matrix to obtain a fused prediction result as a prediction output tensor; the weighted fusion formula is as follows:

[0016] in, For the tensor prediction model at time The prediction tensor of Based on the time The restored tensor obtained by the low-order matrix prediction result, The fusion prediction results.

[0017] In a second aspect, the present invention provides an intelligent Wi-Fi performance perception and prediction system, comprising: A collection module, configured to collect raw signal data from Wi-Fi users, obtain corresponding eigenvalues ​​based on each user's network performance indicator, and construct high-order tensor data in the user dimension based on the eigenvalues; wherein the eigenvalues ​​are in matrix form, with rows representing different network performance indicators and columns representing different sampling time slots; a processing module for normalizing each network performance indicator based on the high-order tensor data, calculating the cross-time slot weight of each network performance indicator, and integrating the multi-dimensional network performance indicator consisting of multiple network performance indicators corresponding to multiple sampling time slots into one-dimensional standardized data and forming a low-order matrix; A prediction module, configured to input the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and input the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix; wherein the tensor prediction model and the matrix prediction model are trained based on a deep reinforcement learning algorithm; An integration module is used to obtain a high-order prediction error based on a high-order prediction tensor and the high-order tensor data, obtain a low-order prediction error based on a low-order prediction matrix and the low-order matrix data, dynamically adjust the fusion weight based on the high-order prediction error and the low-order prediction error, and fuse the prediction results of the tensor prediction model and the matrix prediction model by weighted averaging according to the adjusted fusion weight to obtain a fused prediction result.

[0018] In a third aspect, the present invention provides an electronic device comprising a memory and a processor; the memory is used to store computer-executable instructions, and the processor is connected to the memory via a bus; When the Wi-Fi-oriented intelligent performance perception and prediction system is running, the processor executes the computer-executable instructions stored in the memory, so that the Wi-Fi-oriented intelligent performance perception and prediction system performs the Wi-Fi-oriented intelligent performance perception and prediction method according to any one of the first aspects.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium, comprising computer-executable instructions. When the computer-executable instructions are executed on a computer, the computer executes the Wi-Fi-oriented intelligent performance perception and prediction method according to any one of the first aspects.

[0020] Compared with the prior art, the above technical solutions conceived by the present invention can achieve the following beneficial effects: 1. The present invention provides an intelligent performance perception and prediction method for Wi-Fi. By constructing a three-dimensional high-order tensor and a normalized low-order matrix with users, performance indicators and time series as dimensions, the spatiotemporal characteristics and cross-user behavior differences of Wi-Fi signals are fully preserved. Secondly, a prediction strategy guided by deep reinforcement learning is used to replace the static model training mechanism, which can adapt to the rapid fluctuations of channel conditions. Combining tensor decomposition and matrix decomposition technology, the decomposition method is continuously adjusted through policy iteration to learn the optimal decomposition strategy, so that the model can deeply explore the structural correlation between performance indicators through an interactive policy optimization process during training, thereby improving the accuracy and generalization ability of prediction. It solves the technical problem that there are still limitations in dealing with sudden changes and nonlinear characteristics of Wi-Fi signals, resulting in low accuracy in perception and prediction of Wi-Fi signals, effectively improving the prediction accuracy of network performance and the efficiency of resource allocation, thereby better meeting users' strict requirements for network service quality.

[0021] 2. This invention provides an intelligent Wi-Fi performance perception and prediction method. This method employs a dynamic adjustment mechanism to flexibly adjust the weights of various factors in prediction and decision-making based on diverse network scenarios and user needs. By dynamically adjusting fusion weights based on prediction error feedback, high-order prediction tensors are fused with low-order prediction matrices to achieve efficient and accurate prediction of Wi-Fi network performance indicators and service quality. This enables the system to more accurately perceive and predict performance and make optimal decisions in complex and changing network environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of an intelligent Wi-Fi performance perception and prediction method provided by an embodiment of the present invention.

[0023] Figure 2 This figure shows the architecture of a high-order tensor prediction model for an intelligent Wi-Fi performance perception and prediction method provided by an embodiment of the present invention.

[0024] Figure 3 This figure shows the architecture of a low-order matrix prediction model for an intelligent Wi-Fi performance perception and prediction method provided by an embodiment of the present invention.

[0025] Figure 4 A block diagram of an intelligent Wi-Fi performance perception and prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0027] The contents involved in the above embodiment are described below in conjunction with a preferred embodiment.

[0028] Example 1 like Figure 1 As shown, the present invention provides an intelligent Wi-Fi performance perception and prediction method, including: S100: Collect raw signal data from Wi-Fi users, obtain corresponding eigenvalues ​​based on the network performance indicators of each user, and construct high-order tensor data in the user dimension based on the eigenvalues; wherein the eigenvalues ​​are in matrix form, with rows of the matrix representing different network performance indicators and columns representing different sampling time slots; S200, based on the high-order tensor data, normalize each network performance indicator, calculate the cross-time slot weight of each network performance indicator, and integrate the multi-dimensional network performance indicators consisting of multiple network performance indicators corresponding to multiple sampling time slots into one-dimensional standardized data to form a low-order matrix; S300: Inputting the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and inputting the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix; wherein the tensor prediction model and the matrix prediction model are trained based on a deep reinforcement learning algorithm; S400. Obtain a high-order prediction error based on the high-order prediction tensor and the high-order tensor data, obtain a low-order prediction error based on the low-order prediction matrix and the low-order matrix data, dynamically adjust the fusion weight based on the high-order prediction error and the low-order prediction error, and fuse the prediction results of the tensor prediction model and the matrix prediction model by weighted averaging according to the adjusted fusion weight to obtain a fused prediction result.

[0029] To address the limitations of traditional methods in handling sudden changes, nonlinear features, and mining deep correlations between data, an innovative system architecture that collaborates multidimensional modeling with intelligent strategies is proposed. In the intelligent Wi-Fi performance perception and prediction method provided in this application, a tensor prediction model and a matrix prediction model are constructed in parallel. A training mechanism based on deep reinforcement learning is introduced to autonomously learn the optimal decomposition strategy for different data. Finally, a dynamic fusion strategy based on prediction error feedback is combined to achieve collaborative prediction, thereby improving the accuracy and generalization ability of the prediction model, achieving efficient and accurate prediction of Wi-Fi network performance indicators and service quality, providing a basis for the rational allocation of Wi-Fi resources, and further improving the overall user experience.

[0030] Optionally, step S100 includes: S110, capturing signal interaction information between user devices and access points in a Wi-Fi network environment, and collecting original signal data of Wi-Fi users at different network usage periods; S120. For each independent Wi-Fi user, convert different network performance indicator data in the original signal data into a matrix, with the network performance indicator as the row and the sampling time slot as the column, to serve as the eigenvalue matrix of the network performance indicator of the corresponding user; S130, treat each user's eigenvalue matrix as an independent substructure, stack and combine them on the user dimension to form high-order tensor data .

[0031] Relying on network monitoring equipment and software systems to collect data, comprehensively capture the signal interaction information between user devices and access points in the Wi-Fi network environment.

[0032] In terms of time, we employ long-term continuous monitoring, sampling at second-level intervals to ensure coverage of different periods of network usage and comprehensively capture data characteristics under varying network loads. In terms of users, we identify and classify network access devices, encompassing a wide range of common Wi-Fi terminals, such as smartphones, tablets, laptops, and smart IoT devices. This allows us to obtain diverse user data samples, enhancing their representativeness and universality.

[0033] After data collection is complete, the collected network performance indicator data for each individual user is converted into a matrix format using a "metric as row, time as column" structure. Specifically, the rows of the matrix are organized with different network performance indicators. For example, signal strength data is listed in the first row, transmission rate data in the second row, latency data in the third row, packet loss rate data in the fourth row, and so on. This ensures that each network performance indicator has a unique row in the matrix. The columns of the matrix are determined by different sampling time slots. Based on a pre-defined fixed-time sampling strategy, the sampled data, arranged in chronological order, is sequentially populated into the matrix columns.

[0034] After constructing the eigenvalue matrix of each user, the matrices of multiple users are integrated to form a high-order tensor In the high-order tensor, each user's matrix is ​​treated as an independent substructure and combined in a new dimension, thereby achieving efficient organization and management of large-scale Wi-Fi user data.

[0035] Optionally, step S200 includes: S210, the high-order tensor data The elements of each eigenvalue matrix in are normalized, and the normalized result is Expressed as:

[0036] in, Represents the column vector in the eigenvalue matrix, the column vector is , Represents a column vector No. elements, is a column vector The maximum value in ;

[0037] S220, constructing a vector based on the normalized value of each network performance indicator in all sampling time slots , calculate the weight of each network performance indicator on all sampling time slots ;in, Indicates the total number of network performance indicators, represents the total number of sampling time slots, Indicates the The network performance index is Normalized value over sampling time slots; S230, integrate multi-dimensional network performance indicators into one-dimensional standardized data and construct a low-order matrix ; Among them, the low-order matrix The corresponding element expression is: , Indicates the The weight of each network performance indicator, Indicates the The network performance index is Normalized value over sampling time slots.

[0038] After acquiring high-order tensor data, the network performance metrics are normalized and, using a weighted scoring mechanism, a comprehensive performance score is extracted for each sampling time slot. Each sampling time slot corresponds to multiple performance metrics (such as signal strength, rate, and latency). The performance metric data for multiple time slots forms a multidimensional vector. The normalized average of each performance metric across all sampling time slots is calculated as the cross-slot weight for that metric. The normalized values ​​of each metric are weighted and summed for each time slot according to the corresponding weight, resulting in a one-dimensional score sequence for each user. Stacking all user score sequences along the user dimension forms a low-order matrix with users as rows and time as columns, simplifying the data structure and facilitating subsequent analysis.

[0039] By normalizing the elements of the above eigenvalue matrix, the data under the network performance indicator dimension is mapped to the interval This ensures that data of different scales and dimensions are comparable, avoiding the impact of large differences in data magnitude on the accuracy of subsequent analysis. The above normalization operation is performed on the column vectors of all network performance indicator dimensions in the high-order tensor, bringing the entire data to a unified standard scale.

[0040] After completing data normalization, the influence weight of each network performance indicator is determined according to its average performance in the entire time series, thereby generating a low-order matrix. The multi-dimensional network performance indicators are integrated into one-dimensional standardized data to construct a low-order matrix. , making the data structure more compact and able to more efficiently reflect the comprehensive characteristics of network performance.

[0041] Optionally, step S300 includes: S310, using tensor decomposition based on deep reinforcement learning algorithm to decompose the high-order tensor data Decompose and construct the model parameters of the initial tensor prediction model to obtain high-order tensor data Minimizing the prediction error between the prediction tensor and the prediction tensor of the initial tensor prediction model is the optimization goal, and the decomposition method of the tensor decomposition is continuously adjusted through strategy iteration to update the model parameters of the initial tensor prediction model; the initial tensor prediction model is trained according to the model parameters of the updated initial tensor prediction model until the network converges or reaches a preset number of iterations, thereby obtaining a tensor prediction model; S320, using matrix decomposition based on deep reinforcement learning algorithm to decompose the low-order matrix, construct model parameters of the initial matrix prediction model, and use the low-order matrix Minimizing the error between the prediction matrix and the initial matrix prediction model is used as the optimization goal, the decomposition method of the matrix decomposition is continuously adjusted through strategy iteration, and the model parameters of the initial matrix prediction model are updated; the initial matrix prediction model is trained according to the model parameters of the updated initial matrix prediction model until the network converges or reaches a preset number of iterations, thereby obtaining a matrix prediction model; S330. Input the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and input the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix.

[0042] This solution utilizes the Deep Q-Network (DQN) framework, combining deep learning neural networks with reinforcement learning (Q-learning). It seeks the optimal decomposition strategy for accurate prediction by maximizing cumulative rewards. This method requires determining a basic triplet: state, action, and reward. Using the DQN algorithm from deep reinforcement learning, combined with tensor and matrix decomposition techniques, the optimization objective is to minimize the error between the original tensor or matrix and its corresponding prediction. Through policy iteration, the decomposition method is continuously adjusted to learn the optimal decomposition strategy. Since both high-order tensor and low-order matrix data are implemented using deep reinforcement learning algorithms, the only difference between the two methods is the triplet configuration. CP-WOPT, CP-NLS, and Tucker decomposition rules are used for high-order tensors, while Adagrad, RMSProp, and Adam decomposition rules are used for low-order matrices.

[0043] Step S310 includes: S311, initialize the experience replay pool, randomly initialize the main network, and copy the parameters of the main network to the target network; input the high-order tensor data into the initialized tensor prediction model to obtain a high-order prediction state set ,make Get Status ; Determine the set of optional actions ,in, Indicates at time Implement the actions selected by tensor decomposition; S312, by probability Randomly select an action to The probability of selecting the optimal action based on the main network's valuation ;in, ; S313, optimal action obtained according to S312 , and get the corresponding reward value , reach the next state , the vector Put them into the experience replay pool respectively; if the experience replay pool is full, go to step S314; if the experience replay pool is not full, then , and proceed to step S312; S314. Randomly extract a batch of vectors from the experience replay pool, and take out the vectors one by one to input into the main network. Find the action corresponding to the maximum Q value in the main network and combine the above vectors to input into the target network to obtain the target network Q value under the new mode. Calculate the loss function by the obtained main network Q value and the target network Q value, and use the gradient descent method to update the parameters of the tensor prediction model. ; S315, repeat step S314, and set the parameter The target network of the tensor prediction model is copied until the network converges or reaches the preset number of iterations to obtain the optimal tensor decomposition strategy and tensor prediction model.

[0044] Specifically, for high-order tensor data : state: State set at a moment ; Where, is the original tensor, is the prediction tensor, 、 and For one-dimensional coefficient tensors of different sizes, different tensor decomposition rules are used to decompose In the hour 、 and Randomly generated.

[0045] action: Moment Action Set ; Where, Indicates at time The actions selected by tensor decomposition are implemented according to specific rules. During the tensor decomposition process, a tensor can be represented as a combination of multiple factor matrices and weights. The tensor decomposition rules referred to are CP-WOPT, CP-NLS and Tucker.

[0046] Specifically,

[0047] in, represents the outer product, 、 and is the dimension factor matrix corresponding to each tensor and The general term for the column vector of 、 、 wait, For order.

[0048] Factor Matrix and Updates as follows:

[0049] in, and represent Khatri-Rao product and mode-n product respectively, Represents the tensor to be decomposed.

[0050]

[0051] Factor Matrix and Updates as follows:

[0052] in, is the loss function (usually the sum of squared residuals), is the learning rate.

[0053]

[0054] in, is the core tensor, And the factor matrix is ​​updated as follows

[0055] in, represents the tensor to be decomposed, represents the Moore-Penrose generalized inverse.

[0056] award: Rewards of the moment In the current state Next action The goal is to minimize and The global model is optimized with the prediction error between them as the goal. Expressed as:

[0057] Right now The weighted cumulative residual between and .

[0058] Step S320 includes: S321, initialize the experience replay pool, randomly initialize the main network, and set the parameters of the main network Copy to the target network; input the low-order matrix into the initialization matrix prediction model to obtain a high-order prediction state set ,make Get Status ; Determine the set of optional actions ,in, Indicates at time Implement the actions selected by matrix decomposition; S322, by probability Randomly select an action to The probability of selecting the optimal action based on the main network's valuation ;in, ; S323, the optimal action obtained according to S322 , and get the corresponding reward value , reach the next state , the vector Put them into the experience replay pool respectively; if the experience replay pool is full, go to step S324; if the experience replay pool is not full, then , and proceed to step S322; S324. Randomly extract a batch of vectors from the experience replay pool, and take out the vectors one by one to input into the main network. Find the action corresponding to the maximum Q value in the main network and combine the above vectors to input into the target network to obtain the target network Q value under the new mode. Calculate the loss function by the obtained main network Q value and the target network Q value, and use the gradient descent method to update the parameters of the matrix prediction model. ; S325, repeat step S324, and set the parameter The target network is copied to the matrix prediction model until the network converges or reaches the preset number of iterations, and the optimal matrix decomposition strategy and matrix prediction model are obtained.

[0059] Specifically, for low-order matrix data : state: State set at a moment ; Where, is the original matrix, is the prediction matrix, Different matrix decomposition rules are used for coefficient matrices of different sizes. get, hour Randomly generated. For decomposition rules. It can be further expressed as:

[0060] in, Represents an optional matrix decomposition rule, including Adagrad, RMSProp and Adam.

[0061] Specifically,

[0062]

[0063]

[0064] in, is a matrix Elements, is a matrix Elements, and They are matrices and The total offset of 、 and There are three constants, 、 、 and Represents elements respectively 、 and Bias-corrected first-order moment estimates, 、 、 and is the second-order moment estimate after corresponding bias correction, 、 、 and are the next decomposition results, represents the objective function.

[0065] action: Moment Action Set ; Where, Indicates at time Perform matrix decomposition actions, following the rules of matrix decomposition .

[0066] award: Rewards of the moment In the current state Next action The set of immediate rewards obtained after and The global model is optimized with the prediction error between them as the goal.

[0067] award Expressed as:

[0068] in, is the weight matrix, is the coefficient, and The number of rows, columns and potential feature dimensions respectively, is a matrix The average of all non-zero elements in .

[0069] When there are enough vectors in the experience replay pool, a batch of vectors can be randomly drawn from the experience replay pool. For example, and Input the main neural network to get the Q estimate , Represents the weight of the main neural network. Find the action corresponding to the maximum Q value in the current Q network:

[0070] Will as well as Input target Q network, plus Get the target Q value:

[0071] is the discount factor, Represents the weight of the target Q network, and the loss function is obtained by subtracting the Q value of the main network from the Q value of the target network:

[0072] Using loss function Loss Update the main network weights by gradient descent The weight of the target Q network The update depends on the training results of the main network. In order to reduce the impact of data fluctuations on the stability of the model, T , the weight of the main network Weights copied to the target network , which keeps the target network fixed for a period of time, reducing model volatility. This process is repeated repeatedly until the loss function converges to a sufficiently small range. At this point, the main neural network has converged, and deploying the optimal strategy based on the Q estimate can yield the optimal strategy.

[0073] like Figure 2 and Figure 3 As shown, the specific steps for high-order prediction tensors and low-order prediction matrices are as follows: (1) Initialize the experience replay pool, randomly initialize the main Q network, and copy the parameters to the target Q network, that is, .

[0074] (2) Initialize the high-order prediction state set and low-order prediction state set , and let Get Status and . Determine the set of optional actions and .

[0075] (3) An action is randomly selected with probability The probability of selecting the optimal action based on the main network's valuation and .

[0076] (4) Execute the final selected action and , and get reward values ​​respectively and , reach the next state and , the vector and Put them into the experience replay pool of each model training respectively; if the experience replay pool is full, go to step (5); if the experience replay pool is not full, then , and go to step (3).

[0077] (5) Randomly extract a small batch of vector sets from the experience replay pool.

[0078] (6) Take out vectors one by one and , and input it into the main network, find the action corresponding to the maximum Q value in the main network and input the above vector into the target network to obtain the Q value of the target network under the new mode. The loss function is calculated by the obtained main network Q value and the target network Q value:

[0079] Update the parameters using the loss function via gradient descent .

[0080] (7) Repeat steps (5)-(6) and repeat every time length T Copy the main network weights to the target network , until the network converges.

[0081] Optionally, step S400 includes: S410: performing a restoration operation on the prediction result of the low-order prediction matrix according to the original dimension information index of the low-order matrix data and the cross-time slot weight to obtain a restored tensor consistent with the high-order tensor structure for subsequent unified fusion processing; Because the structure of low-order matrices only retains partial dimensional information (e.g., indicator dimensions are weighted and integrated), their prediction results cannot be directly fused with the data structure output by the high-order tensor prediction model. To achieve structural alignment and fusion calculations, the matrix prediction results are reverse-restored based on the cross-slot weights previously used for compression operations, reconstructing them into a three-dimensional structured tensor consistent with the structure of the high-order tensor. This allows for weighted fusion with the high-order tensor prediction results in a unified dimension, improving fusion accuracy and adaptability.

[0082] S420: Determine the initial fusion weight of the matrix prediction model and the tensor prediction model based on the errors between the matrix prediction model and the tensor prediction model. The initial fusion weight is expressed as:

[0083] in, and The matrix prediction model and the tensor prediction model are respectively The error, is a smoothing coefficient used to avoid division by zero and smooth the error; S430: Normalize the initial fusion weight to obtain a dynamic fusion weight coefficient:

[0084] in, is the influence coefficient, which determines the relative influence between the two model errors; when When , the fusion weight of the model with smaller error in the tensor prediction model and the matrix prediction model is set to be greater than that of the model with larger error; when When the fusion weight of the model with larger error in the tensor prediction model and the matrix prediction model is set to be larger than that of the model with smaller error; the fusion weight coefficient satisfies ; S440, performing weighted fusion on the high-order prediction tensor and the restored tensor obtained from the low-order matrix to obtain a fused prediction result as a prediction output tensor; the weighted fusion formula is as follows:

[0085] in, For the tensor prediction model at time The prediction tensor of Based on the time The restored tensor obtained by the low-order matrix prediction result, The fusion prediction results.

[0086] The fusion prediction results obtained in the above scheme It is based on fully learning the evolution trend of user performance indicators in multiple historical sampling time slots, and generates a time series prediction output through a model inference process driven by deep reinforcement learning, which can accurately characterize the evolution trend of network performance indicators in the future. Essentially reflects the time The prediction results of key performance parameters such as signal strength, transmission delay, throughput, and packet loss rate in one or more subsequent time slots are then generated. Based on these prediction results, the network system can identify potential performance bottlenecks in advance and implement adaptive adjustments to dynamic resource allocation strategies, such as early frequency band switching, load balancing, and link priority adjustment, thereby effectively alleviating network congestion and avoiding signal degradation. In addition, this invention can provide a decision-making basis for scenarios such as user roaming strategies, bandwidth scheduling, and QoS assurance in high-density access environments, further improving overall network operation efficiency and the continuity, stability, and satisfaction of user experience.

[0087] In this embodiment of the present invention, by constructing a three-dimensional high-order tensor and a normalized low-order matrix based on user, performance indicator, and time series dimensions, the spatiotemporal characteristics of Wi-Fi signals and the cross-user behavioral differences are fully preserved. Secondly, a prediction strategy guided by deep reinforcement learning replaces the static model training mechanism, which can adapt to the rapid fluctuations in channel conditions. Combining tensor decomposition and matrix decomposition techniques, the decomposition method is continuously adjusted through policy iteration to learn the optimal decomposition strategy. This allows the model to deeply explore the structural correlations between performance indicators through an interactive policy optimization process during training, thereby improving prediction accuracy and generalization capabilities. This solves the technical problem of low Wi-Fi signal perception and prediction accuracy, which still exists when dealing with sudden changes and nonlinear characteristics of Wi-Fi signals. It effectively improves the prediction accuracy of network performance and the efficiency of resource allocation, thereby better meeting users' stringent requirements for network service quality.

[0088] Example 2 like Figure 4 As shown, the present invention provides an intelligent performance perception and prediction system for Wi-Fi, including: A collection module, configured to collect raw signal data from Wi-Fi users, obtain corresponding eigenvalues ​​based on each user's network performance indicator, and construct high-order tensor data in the user dimension based on the eigenvalues; wherein the eigenvalues ​​are in matrix form, with rows representing different network performance indicators and columns representing different sampling time slots; a processing module for normalizing each network performance indicator based on the high-order tensor data, calculating the cross-time slot weight of each network performance indicator, and integrating the multi-dimensional network performance indicator consisting of multiple network performance indicators corresponding to multiple sampling time slots into one-dimensional standardized data and forming a low-order matrix; A prediction module, configured to input the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and input the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix; wherein the tensor prediction model and the matrix prediction model are trained based on a deep reinforcement learning algorithm; An integration module is used to obtain a high-order prediction error based on a high-order prediction tensor and the high-order tensor data, obtain a low-order prediction error based on a low-order prediction matrix and the low-order matrix data, dynamically adjust the fusion weight based on the high-order prediction error and the low-order prediction error, and fuse the prediction results of the tensor prediction model and the matrix prediction model by weighted averaging according to the adjusted fusion weight to obtain a fused prediction result.

[0089] The Wi-Fi-oriented intelligent performance perception and prediction system provided by the embodiment of the present invention is used to perform the above Figure 1The Wi-Fi-oriented intelligent performance perception and prediction method in the embodiment shown. For details not covered in this embodiment, please refer to the aforementioned Figure 1 The intelligent performance perception and prediction method for Wi-Fi in the illustrated embodiment will not be described in detail here.

[0090] Example 3 The present invention provides an electronic device, comprising a memory and a processor; the memory is used to store computer-executable instructions, and the processor is connected to the memory via a bus; When the Wi-Fi-oriented intelligent performance perception and prediction system is running, the processor executes the computer-executable instructions stored in the memory, so that the Wi-Fi-oriented intelligent performance perception and prediction system performs the Wi-Fi-oriented intelligent performance perception and prediction method described in any one of Embodiment 1.

[0091] Example 4 The present invention provides a computer-readable storage medium, comprising computer-executable instructions. When the computer-executable instructions are executed on a computer, the computer is caused to execute the Wi-Fi-oriented intelligent performance perception and prediction method described in any one of the first embodiments.

[0092] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent performance perception and prediction method for Wi-Fi, characterized by: include: S100: Collect raw signal data from Wi-Fi users, obtain corresponding eigenvalues ​​based on the network performance indicators of each user, and construct high-order tensor data in the user dimension based on the eigenvalues; wherein the eigenvalues ​​are in matrix form, with rows of the matrix representing different network performance indicators and columns representing different sampling time slots; S200, based on the high-order tensor data, normalize each network performance indicator, calculate the cross-time slot weight of each network performance indicator, and integrate the multi-dimensional network performance indicators consisting of multiple network performance indicators corresponding to multiple sampling time slots into one-dimensional standardized data to form a low-order matrix; S300: Inputting the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and inputting the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix; wherein the tensor prediction model and the matrix prediction model are trained based on a deep reinforcement learning algorithm; S400. Obtain a high-order prediction error based on the high-order prediction tensor and the high-order tensor data, obtain a low-order prediction error based on the low-order prediction matrix and the low-order matrix data, dynamically adjust the fusion weight based on the high-order prediction error and the low-order prediction error, and fuse the prediction results of the tensor prediction model and the matrix prediction model by weighted averaging according to the adjusted fusion weight to obtain a fused prediction result.

2. The method according to claim 1, wherein Step S100 includes: S110, capturing signal interaction information between user devices and access points in a Wi-Fi network environment, and collecting original signal data of Wi-Fi users at different network usage periods; S120. For each independent Wi-Fi user, convert different network performance indicator data in the original signal data into a matrix, with the network performance indicator as the row and the sampling time slot as the column, to serve as the eigenvalue matrix of the network performance indicator of the corresponding user; S130, treat each user's eigenvalue matrix as an independent substructure, stack and combine them on the user dimension to form high-order tensor data .

3. The method according to claim 1, wherein Step S200 includes: S210, the high-order tensor data The elements of each eigenvalue matrix in are normalized, and the normalized result is Expressed as: ;in, Represents the column vector in the eigenvalue matrix, the column vector is , Represents a column vector No. elements, is a column vector The maximum value in ; S220, constructing a vector based on the normalized value of each network performance indicator in all sampling time slots , calculate the weight of each network performance indicator on all sampling time slots ;in, Indicates the total number of network performance indicators, represents the total number of sampling time slots, Indicates the The network performance index is Normalized value over sampling time slots; S230, integrate multi-dimensional network performance indicators into one-dimensional standardized data and construct a low-order matrix ; Among them, the low-order matrix Corresponding elements The expression is: , Indicates the The weight of each network performance indicator, Indicates the The network performance index is Normalized value over sampling time slots.

4. The method according to claim 1, wherein Step S300 includes: S310, using tensor decomposition based on deep reinforcement learning algorithm to decompose the high-order tensor data Decompose and construct the model parameters of the initial tensor prediction model to obtain high-order tensor data Minimizing the prediction error between the prediction tensor and the prediction tensor of the initial tensor prediction model is the optimization goal, and the decomposition method of the tensor decomposition is continuously adjusted through strategy iteration to update the model parameters of the initial tensor prediction model; the initial tensor prediction model is trained according to the model parameters of the updated initial tensor prediction model until the network converges or reaches a preset number of iterations, thereby obtaining a tensor prediction model; S320, using matrix decomposition based on deep reinforcement learning algorithm to decompose the low-order matrix, construct model parameters of the initial matrix prediction model, and use the low-order matrix Minimizing the error between the prediction matrix and the initial matrix prediction model is used as the optimization goal, the decomposition method of the matrix decomposition is continuously adjusted through strategy iteration, and the model parameters of the initial matrix prediction model are updated; the initial matrix prediction model is trained according to the model parameters of the updated initial matrix prediction model until the network converges or reaches a preset number of iterations, thereby obtaining a matrix prediction model; S330. Input the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and input the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix.

5. The method according to claim 4, wherein Step S310 includes: S311, initialize the experience replay pool, randomly initialize the main network, and set the parameters of the main network Copy to the target network; input the high-order tensor data into the initialized tensor prediction model to obtain a high-order prediction state set ,make Get Status ; Determine the set of optional actions ,in, Indicates at time Implement the actions selected by tensor decomposition; S312, by probability Randomly select an action to The probability of selecting the optimal action based on the main network's valuation ;in, ; S313, optimal action obtained according to S312 , and get the corresponding reward value , reach the next state , the vector Put them into the experience replay pool respectively; if the experience replay pool is full, go to step S314; if the experience replay pool is not full, then , and proceed to step S312; S314. Randomly extract a batch of vectors from the experience replay pool, and take out the vectors one by one to input into the main network. Find the action corresponding to the maximum Q value in the main network and combine the above vectors to input into the target network to obtain the target network Q value under the new mode. Calculate the loss function by the obtained main network Q value and the target network Q value, and use the gradient descent method to update the parameters of the tensor prediction model. ; S315, repeat step S314, and set the parameter The target network of the tensor prediction model is copied until the network converges or reaches the preset number of iterations to obtain the optimal tensor decomposition strategy and tensor prediction model.

6. The method according to claim 4, wherein Step S320 includes: S321, initialize the experience replay pool, randomly initialize the main network, and set the parameters of the main network Copy to the target network; input the low-order matrix into the initialization matrix prediction model to obtain a low-order prediction state set ,make Get Status ; Determine the set of optional actions ,in, Indicates at time Implement the actions selected by matrix decomposition; S322, by probability Randomly select an action to The probability of selecting the optimal action based on the main network's valuation ;in, ; S323, the optimal action obtained according to S322 , and get the corresponding reward value , reach the next state , the vector Put them into the experience replay pool respectively; if the experience replay pool is full, go to step S324; if the experience replay pool is not full, then And proceed to step S322; S324. Randomly extract a batch of vectors from the experience replay pool, and take out the vectors one by one to input into the main network. Find the action corresponding to the maximum Q value in the main network and combine the above vectors to input into the target network to obtain the target network Q value under the new mode. Calculate the loss function by the obtained main network Q value and the target network Q value, and use the gradient descent method to update the parameters of the matrix prediction model. ; S325, repeat step S324, and set the parameter The target network is copied to the matrix prediction model until the network converges or reaches the preset number of iterations, and the optimal matrix decomposition strategy and matrix prediction model are obtained.

7. The method according to claim 1, wherein Step S400 includes: S410: performing a restoration operation on the prediction result of the low-order prediction matrix according to the original dimension information index of the low-order matrix data and the cross-time slot weight to obtain a restored tensor consistent with the high-order tensor structure; S420: Determine the initial fusion weight of the matrix prediction model and the tensor prediction model based on the errors between the matrix prediction model and the tensor prediction model. The initial fusion weight is expressed as: ;in, and are the errors of the matrix prediction model and the tensor prediction model at time t, is a smoothing coefficient used to avoid division by zero and smooth the error; S430: Normalize the initial fusion weight to obtain a dynamic fusion weight coefficient: ;in, is the influence coefficient, which determines the relative influence between the two model errors; when When , the fusion weight of the model with smaller error in the tensor prediction model and the matrix prediction model is set to be greater than that of the model with larger error; when When the fusion weight of the model with larger error in the tensor prediction model and the matrix prediction model is set to be larger than that of the model with smaller error; the fusion weight coefficient satisfies ; S440, performing weighted fusion on the high-order prediction tensor and the restored tensor obtained from the low-order matrix to obtain a fused prediction result as a prediction output tensor; the weighted fusion formula is as follows: ;in, For the tensor prediction model at time The prediction tensor of Based on the time The restored tensor obtained by the low-order matrix prediction result, The fusion prediction results.

8. An intelligent Wi-Fi performance perception and prediction system, characterized by: include: A collection module, configured to collect raw signal data from Wi-Fi users, obtain corresponding eigenvalues ​​based on each user's network performance indicator, and construct high-order tensor data in the user dimension based on the eigenvalues; wherein the eigenvalues ​​are in matrix form, with rows representing different network performance indicators and columns representing different sampling time slots; a processing module for normalizing each network performance indicator based on the high-order tensor data, calculating the cross-time slot weight of each network performance indicator, and integrating the multi-dimensional network performance indicator consisting of multiple network performance indicators corresponding to multiple sampling time slots into one-dimensional standardized data and forming a low-order matrix; A prediction module, configured to input the high-order tensor data into a pre-trained tensor prediction model to obtain a high-order prediction tensor, and input the low-order matrix data into a pre-trained matrix prediction model to obtain a low-order prediction matrix; wherein the tensor prediction model and the matrix prediction model are trained based on a deep reinforcement learning algorithm; An integration module is used to obtain a high-order prediction error based on a high-order prediction tensor and the high-order tensor data, obtain a low-order prediction error based on a low-order prediction matrix and the low-order matrix data, dynamically adjust the fusion weight based on the high-order prediction error and the low-order prediction error, and fuse the prediction results of the tensor prediction model and the matrix prediction model by weighted averaging according to the adjusted fusion weight to obtain a fused prediction result.

9. An electronic device, characterized in that: It includes a memory and a processor; the memory is used to store computer-executable instructions, and the processor is connected to the memory via a bus; When the Wi-Fi-oriented intelligent performance perception and prediction system is running, the processor executes the computer-executable instructions stored in the memory, so that the Wi-Fi-oriented intelligent performance perception and prediction system performs the Wi-Fi-oriented intelligent performance perception and prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes computer-executable instructions. When the computer-executable instructions are executed on a computer, the computer is enabled to execute the Wi-Fi-oriented intelligent performance perception and prediction method according to any one of claims 1 to 7.

Images