A method and system for estimating timing characteristics

Abstract

The application provides a kind of timing feature estimation method and system, it is related to wireless communication perception technical field.The method comprises: collecting channel state information timing data, utilizes the weighted moving average filter based on window function to filter processing timing data;For the timing data after filtering processing, utilize the deep learning framework combining deep separable convolution, hollow convolution and Transformer model to extract local feature, and utilize the long time dependence between local feature in multi-head self-attention mechanism and feedforward network in Transformer model;Utilize the adaptive fine-tuning algorithm based on bayesian optimization, five-fold cross-validation, hyperparameter tuning and few-shot learning to carry out offline learning and online application optimization to deep learning framework;Timing feature estimation is carried out based on the optimized deep learning framework, and fingerprint library is established.The application can realize efficient, accurate and adaptive processing to timing data.

Description

Technical Field

[0001] This invention relates to the field of wireless communication sensing technology, and in particular to a method and system for estimating time-series characteristics. Background Technology

[0002] Temporal features refer to key attributes extracted from data sequences that change over time, revealing patterns such as trends, periodicity, and volatility. With the rapid advancement of fields such as the Internet of Things, smart homes, and autonomous driving, not only is efficient and stable data transmission required, but also the intelligent understanding and processing of the rich temporal feature information contained in signals is needed to achieve more accurate and efficient data analysis and decision support.

[0003] Current technologies primarily rely on deep learning models such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and gated recurrent units (GRUs) to process and predict time-series data. However, these models have many parameters, high computational resource requirements, and struggle to capture dynamic changes across all dimensions when faced with complex, high-dimensional time-series data. They are also sensitive to outliers and noise in the data, exhibit poor generalization ability, and low prediction accuracy. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a method and system for estimating time series features, thereby solving the problems of noise and interference, difficulty in capturing local and long-term dependencies, and insufficient adaptability to environmental changes in existing time series data analysis.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] On the one hand, a method for estimating time-series features is provided, the method comprising the following steps:

[0007] S1. Collect Channel State Information (CSI) time-series data, and filter the time-series data using a window function-based weighted moving average filter;

[0008] S2. For the filtered time-series data, a deep learning framework combining depthwise separable convolution, dilated convolution, and Transformer model is used to extract local features, and the multi-head self-attention mechanism and feedforward network in the Transformer model are used to capture the long-term dependencies between the local features.

[0009] S3. The deep learning framework is optimized for offline learning and online application using an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning.

[0010] S4. Based on the optimized deep learning framework, estimate the temporal features of the input CSI time series data and establish a fingerprint database.

[0011] Optionally, step S1 specifically includes:

[0012] Let the input signal be x[n], where n represents the time step. For each time step n, calculate the local standard deviation s[n]:

[0013]

[0014] Where N is the window size, and W is half the window size. The mean of the signal within the window is used; the window size N is adjusted based on the calculated local standard deviation s[n] and a predefined threshold T; if s[n] > T, a smaller window size N is chosen. min If s[n] ≤ T, choose the larger window size N. max The adjustment process is represented by the following formula:

[0015]

[0016] Using the adjusted window size N', calculate the weighted average y[n] for the current time step n:

[0017]

[0018] y[n] is the value of the output signal of the filter.

[0019] Optionally, step S2 specifically includes:

[0020] Let X be the filtered time-series data with shape (N, L), where N is the batch size and L is the sequence length. The processing order of the deep learning framework is as follows:

[0021] S21. Perform batch standardization on the time series data X:

[0022] X bn =BN(X)

[0023] Wherein, BN represents the batch standardization layer;

[0024] S22, Standardize the batch X bn Input depth can separate combinations of convolution and standard convolution:

[0025] X conv =Pool(AF(BN(DSC(Pool(AF(BN(DSC(X bn ))))))))

[0026] Where DSC represents depthwise separable convolutional layer, AF represents activation function, and Pool represents pooling operation;

[0027] S23, X conv Input dilated convolution:

[0028] X dil =Pool(AF(BN(Conv) dil (X conv ))))

[0029] Among them, Conv dil This indicates a dilated convolution that expands the receptive field;

[0030] S24, X dil Input adaptation layer transformation:

[0031] X adapt =W ad ·Flatten(X dil )+b ad

[0032] Among them, W ad It is the weight matrix of the adaptation layer, b ad It is the bias term; Flatten is the operation that flattens the multidimensional convolution output into one dimension.

[0033] S25, X adapt Input Transformer model:

[0034] X trans =Transformer(X) adapt )

[0035] The Transformer model is used to capture long-term dependencies.

[0036] S26, X trans Input to a fully connected layer and output:

[0037] Y = W fc ·X trans +b fc

[0038] Where Y is the final predicted output, and W fc It is the weight matrix of the fully connected layer, b fc It is a bias term;

[0039] The forward propagation of the deep learning framework can be represented by the following composite function:

[0040] Y = W fc Transformer (W) ad·Flatten(Pool(AF(BN(Conv dil (Pool(AF(BN(DSC(BN(X))))))))))+b ad )+b fc

[0041] Optionally, step S3 specifically includes:

[0042] S31. Use Bayesian optimization to select the optimal combination of hyperparameters;

[0043] S32. Evaluate the performance of each optimal hyperparameter combination selected by Bayesian optimization through five-fold cross-validation;

[0044] S33. Improve the adaptability of the deep learning framework to new environments through few-shot learning.

[0045] Optionally, step S31 specifically includes:

[0046] Let the hyperparameter space be Θ. For a combination of hyperparameters θ in the hyperparameter space Θ, Bayesian optimization is used to find the optimal combination of hyperparameters θ. * Maximize or minimize the performance index f(θ):

[0047] θ * =argmax θ∈Θ f(θ)

[0048] or,

[0049] θ * =argmin θ∈Θ f(θ).

[0050] Optionally, step S32 specifically includes:

[0051] Each optimal combination of hyperparameters θ selected by Bayesian optimization was verified using five-fold cross-validation. * Performance f(θ) * To conduct an evaluation:

[0052]

[0053] Among them, f i (θ * ) represents the optimal hyperparameter combination θ at the i-th fold. * The performance of the trained model.

[0054] Optionally, step S33 specifically includes:

[0055] Let D new For a small dataset in a new environment, the deep learning framework's adaptability to the new environment is improved through few-shot learning:

[0056]

[0057] in, It is a loss function used to measure the performance of the deep learning framework in D. new Performance on the surface.

[0058] On the other hand, a time-series feature estimation system is provided for implementing the method described in any of the above embodiments, the system comprising:

[0059] The data processing module is used to collect channel state information (CSI) time-series data and to filter the time-series data using a window function-based weighted moving average filter.

[0060] The deep learning module is used to extract local features from filtered time-series data using a deep learning framework that combines depthwise separable convolution, dilated convolution, and Transformer models. It also uses the multi-head self-attention mechanism and feedforward network in the Transformer model to capture long-term dependencies between these local features.

[0061] An adaptive fine-tuning module is used to perform offline learning and online application optimization of the deep learning framework using an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning.

[0062] The estimation module is used to estimate the temporal features of the input CSI time series data based on the optimized deep learning framework and to build a fingerprint database.

[0063] On the other hand, an electronic device is provided, the electronic device comprising:

[0064] processor;

[0065] A memory storing computer-readable instructions, which, when loaded and executed by the processor, implement the steps of the estimation method described above.

[0066] On the other hand, a computer-readable storage medium is provided, wherein program code is stored therein, the program code being invoked by a processor to execute the steps of the estimation method described above.

[0067] The beneficial effects of the technical solution provided by this invention include at least the following:

[0068] (1) The estimation method proposed in this invention can solve the problems of noise and interference, difficulty in capturing time dependence and insufficient adaptability to environmental changes in existing time series data analysis, and achieve efficient, accurate and adaptive processing of time series data.

[0069] (2) The deep learning framework (CNN-DDT) proposed in this invention, which combines depthwise separable convolution, dilated convolution and Transformer model, can effectively extract local features of time series data and capture the temporal dependencies between these local features, showing significant efficiency and accuracy in processing complex time series data.

[0070] (3) The Smart Adaptive Filter (SAF) based on window function proposed in this invention can effectively remove noise from time series data and perform smoothing, ensuring the flexibility and accuracy of signal processing.

[0071] (4) The adaptive fine-tuning algorithm proposed in this invention, based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning and few-shot learning, can significantly reduce the model’s dependence on a large amount of training data and improve the model’s generalization ability and adaptability. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1 This is a flowchart of a time series feature estimation method provided in an embodiment of the present invention;

[0074] Figure 2 This is a schematic diagram of using an intelligent adaptive filter for smooth noise reduction provided in an embodiment of the present invention;

[0075] Figure 3 This is a schematic diagram of the training loss function provided in an embodiment of the present invention;

[0076] Figure 4 This is a schematic diagram of the structure of a time-series feature estimation system provided in an embodiment of the present invention;

[0077] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0079] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0080] This invention provides a method for estimating time-series features, which can be implemented by an electronic device, such as a terminal or a server. Figure 1 As shown, the processing flow of this method may include the following steps:

[0081] S1. Collect Channel State Information (CSI) time-series data and filter the time-series data using a window function-based weighted moving average filter.

[0082] In this embodiment of the invention, a window function-based weighted moving average filter—the "Smart Adaptive Filter" (SAF)—is designed. This filter can effectively remove high-frequency noise and smooth signals. Before signal processing, a grid search algorithm is designed to determine the optimal filter parameters (minimum window size, maximum window size, and threshold for adjusting the window size) based on the maximum signal-to-noise ratio. The advantage of this method is that it provides a more flexible processing method than traditional fixed-window-size moving average filters, and can better adapt to different signal characteristics.

[0083] SAF seeks a balance between reducing high-frequency noise and preserving important signal characteristics. Let the input signal be x[n], where n represents the time step. For each time step n, the local standard deviation s[n] is calculated:

[0084]

[0085] Local standard deviation represents the degree of signal variation within a local window centered on the current point. Here, N is the window size, and W is half the window size. It is the mean of the signal within the window; the window size N is adjusted based on the calculated local standard deviation s[n] and the predefined threshold T; if s[n] > T, it indicates that the local change at the current point is large, and a smaller window size N is selected. min If s[n]≤T, it means that the local change at the current point is small, and a larger window size N is selected. max The adjustment process is represented by the following formula:

[0086]

[0087] Using the adjusted window size N', calculate the weighted average y[n] for the current time step n:

[0088]

[0089] This formula represents the mean of the signal values ​​within the window. The window size N determines the average range. The final output signal is obtained by assigning a weighted average value y[n] to each time step n, where y[n] is the value of the filter's output signal.

[0090] like Figure 2 As shown, the filter design process smooths the input signal by dynamically adjusting the window size based on the signal's local variability. The filter's sensitivity to local signal characteristics allows it to maintain signal detail in highly variable regions while reducing noise in stable regions. This design performs well when processing non-stationary signals and effectively removes noise.

[0091] S2. For the filtered time-series data, a deep learning framework combining depthwise separable convolution, dilated convolution, and Transformer model is used to extract local features, and the multi-head self-attention mechanism and feedforward network in the Transformer model are used to capture the long-term dependencies between the local features.

[0092] This invention proposes a deep learning framework (or deep learning model) that combines depthwise separable convolution, dilated convolution, and the Transformer model, named CNN-DDT. This framework extracts local features from time-series data through depthwise separable convolution and dilated convolution, and utilizes the multi-head self-attention mechanism and feedforward network operations in the Transformer model to capture long-term dependencies between these local features. This framework is suitable for processing complex time-series data, such as time-series analysis in fields like communications, the Internet of Things, and natural language processing.

[0093] Specifically, let X be the filtered time-series data with shape (N, L), where N is the batch size and L is the sequence length. The processing order of the deep learning framework is as follows:

[0094] S21. Perform batch normalization on the time series data X:

[0095] X bn =BN(X)

[0096] Wherein, BN represents the batch standardization layer;

[0097] S22, Standardize the batch X bn Combination of depthwise separable convolution and standard convolution:

[0098] X conv =Pool(AF(BN(DSC(Pool(AF(BN(DSC(X bn ))))))))

[0099] Where DSC represents depthwise separable convolutional layer, AF represents activation function, and Pool represents pooling operation;

[0100] S23, X conv Input dilated convolution:

[0101] X dil =Pool(AF(BN(Conv) dil (X conv ))))

[0102] Among them, Conv dil This indicates a dilated convolution that expands the receptive field;

[0103] S24, X dil Input adapter layer transformation:

[0104] X adapt =W ad ·Flatten(X dil )+b ad

[0105] Among them, W ad It is the weight matrix of the adaptation layer, b ad It is the bias term; Flatten is the operation that flattens the multidimensional convolution output into one dimension.

[0106] S25, X adapt Input Transformer model (TransformerModel):

[0107] X trans =Transformer(X) adapt )

[0108] The Transformer model is used to capture long-term dependencies.

[0109] S26, X trans Input to a fully connected layer for output:

[0110] Y = W fc ·X trans +b fc

[0111] Where Y is the final predicted output, and W fc It is the weight matrix of the fully connected layer, b fc It is a bias term;

[0112] The forward propagation of the entire deep learning framework (CNN-DDT framework) can be represented by the following composite function:

[0113] Y = W fc Transformer (W) ad ·Flatten(Pool(AF(BN(Conv dil

[0114] (Pool(AF(BN(DSC(BN(X))))))))))+b ad )+b fc

[0115] S3. The deep learning framework is optimized for offline learning and online application using an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning.

[0116] This invention proposes an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning. Its main function is to significantly reduce the dependence on large amounts of training data while maintaining model performance, thereby improving the model's adaptability and generalization ability.

[0117] Step S3 specifically includes:

[0118] S31. Use Bayesian optimization to select the optimal combination of hyperparameters.

[0119] Bayesian optimization is a strategy for global optimization, particularly suitable for optimizing computationally expensive functions. In algorithms, Bayesian optimization is used to select the optimal combination of hyperparameters.

[0120] Let the hyperparameter space be Θ. For a combination of hyperparameters θ in the hyperparameter space Θ, Bayesian optimization is used to find the optimal combination of hyperparameters θ. * Maximize or minimize the performance index f(θ):

[0121] θ * =argmax θ∈Θ f(θ)

[0122] or,

[0123] θ * =argmin θ∈Θ f(θ).

[0124] The Bayesian optimization process involves constructing a probabilistic model of f(θ) and using this model to predict the performance of each hyperparameter combination while assessing uncertainty. Based on this information, the algorithm determines the next set of hyperparameter combinations to balance exploration (finding new territory) and exploitation (refining within known good regions).

[0125] S32. Evaluate the performance of each optimal hyperparameter combination selected by Bayesian optimization through five-fold cross-validation.

[0126] Cross-validation is a method for evaluating the generalization ability of a model. Five-fold cross-validation means dividing the dataset into five equal parts, using four parts alternately as training data, and the remaining part as validation data. Five-fold cross-validation is used to evaluate each optimal combination of hyperparameters θ selected by Bayesian optimization. * Performance f(θ) * To conduct an evaluation:

[0127]

[0128] Among them, f i (θ * ) represents the optimal hyperparameter combination θ at the i-th fold. * The performance of the trained model.

[0129] S33. Improve the adaptability of the deep learning framework to new environments through few-shot learning.

[0130] The goal of few-shot learning is to enable a model to learn from a limited number of data samples. During model fine-tuning, few-shot learning techniques are used to improve the model's ability to quickly adapt to new environments.

[0131] Let Dnew For a small dataset in a new environment, the deep learning framework's adaptability to the new environment is improved through few-shot learning:

[0132]

[0133] in, It is a loss function used to measure the performance of the deep learning framework in D. new Performance on the surface.

[0134] Specifically, firstly, Bayesian optimization is used to guide the hyperparameter search process. This is achieved by establishing a probabilistic model between hyperparameters and model performance, and using this model to predict the performance and uncertainty of unexplored hyperparameter combinations, thus balancing the need to explore new hyperparameter spaces with the utilization of known good hyperparameters. Secondly, to accurately evaluate the performance of different hyperparameter combinations, five-fold cross-validation is employed. The dataset is divided into five equal parts, with four parts used as training data and the remaining part as validation data, repeated five times to ensure the evaluation quality of the hyperparameter optimization process. The subsequent hyperparameter tuning aims to find the hyperparameter combination that maximizes model performance. The selection of hyperparameter combinations is based on the performance results of previous iterations and the guidance of the Bayesian optimization algorithm. Finally, few-shot learning is utilized in the model fine-tuning stage. By optimizing the objective function designed to adapt to a small number of sample datasets in new environments, the model's ability to quickly adapt to new tasks and its generalization ability are improved. This significantly reduces the dependence on large amounts of training data while maintaining model performance, thus enhancing the model's adaptability and generalization ability.

[0135] S4. Based on the optimized deep learning framework, the temporal features of the input CSI time series data are estimated, and a fingerprint database is established, which can be used for applications such as intelligent positioning.

[0136] The method of the present invention will now be verified through a specific example.

[0137] In an indoor setting, a fingerprint database was built using CSI time-series data to achieve intelligent positioning. A receiver and a transmitter were deployed to cover the entire area and collect signal data. The experiment used two computers equipped with network cards as transceivers, each with three antennas for transmission and reception, utilizing a Wi-Fi module to acquire CSI information. During the fingerprint database establishment phase, a two-dimensional plane coordinate system was established, and each fingerprint collection point was represented by its coordinates.

[0138] according to Figure 3As can be seen, the proposed method (CNN-DTT) performs best overall, with the fastest loss reduction and remaining at a low level, indicating that the method can effectively learn data features and has a stronger ability to learn complex temporal data patterns, requiring fewer iterations compared to LSTM and RNN.

[0139] To verify the effectiveness of the proposed adaptive fine-tuning algorithm when the environment changes, the laboratory environment was changed from LOS to NLOS. The average positioning errors before and after using the algorithm were 5.23m and 0.17m, respectively, and the model prediction accuracy was improved by more than 95%.

[0140] Accordingly, embodiments of the present invention also provide an estimation system for time-series features, such as... Figure 4 As shown, the system includes:

[0141] Data processing module 201 is used to collect channel state information (CSI) time series data and to filter the time series data using a window function-based weighted moving average filter.

[0142] The deep learning module 202 is used to extract local features from filtered time-series data using a deep learning framework that combines depthwise separable convolution, dilated convolution, and Transformer models, and to capture long-term dependencies between the local features using the multi-head self-attention mechanism and feedforward network in the Transformer model.

[0143] The adaptive fine-tuning module 203 is used to perform offline learning and online application optimization of the deep learning framework using an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning and few-shot learning.

[0144] The estimation module 204 is used to estimate the temporal features of the input CSI time series data based on the optimized deep learning framework and to establish a fingerprint database.

[0145] For ease of explanation, Figure 4 Only the main components of the system are shown. The system of this embodiment can be used to perform... Figure 1 The technical solutions of the method embodiments shown are similar in principle and in effect, and will not be described again here.

[0146] In summary, the time series feature estimation method and system proposed in this invention have the following advantages:

[0147] (1) This invention proposes an intelligent estimation method for time series features. By combining a weighted moving average filter based on window function with a deep learning framework of depthwise separable convolution, dilated convolution and Transformer model, as well as an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning and few-shot learning, this method solves the problems of noise and interference, difficulty in capturing time dependencies and insufficient adaptability to environmental changes in existing time series data analysis, thereby achieving efficient, accurate and adaptive processing of time series data.

[0148] (2) This invention proposes a deep learning framework named CNN-DTT, which combines depthwise separable convolution, dilated convolution, and Transformer models to optimize feature extraction and temporal dependency capture capabilities for time-series data. It also utilizes a multi-head self-attention mechanism and a feedforward network to achieve efficient processing and analysis of complex sequence data.

[0149] (3) This invention proposes a weighted moving average filter based on a window function, named Smart Adaptive Filter (SAF). First, the optimal filtering parameters are adaptively selected based on the local changes of the signal through a grid search algorithm. Then, the window size is dynamically adjusted according to the local standard deviation to achieve effective noise removal and signal smoothing.

[0150] (4) This invention proposes an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning, which significantly reduces the need for large training sets while optimizing model performance. This algorithm achieves a balance between the requirements of new and old parameters while ensuring evaluation quality through Bayesian optimization-guided hyperparameter search, combined with parameter-performance probability model prediction and five-fold cross-validation. Furthermore, it utilizes few-shot learning to enhance the model's adaptability and generalization ability to new tasks, significantly reducing its dependence on large amounts of data.

[0151] In an exemplary embodiment, the present invention also provides an electronic device, the electronic device comprising:

[0152] processor;

[0153] A memory storing computer-readable instructions, which, when loaded and executed by the processor, implement the steps of the estimation method described above.

[0154] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 5As shown, the electronic device 300 may include a processor 3001 and a memory 3002. Optionally, the electronic device 300 may also include a transceiver 3003. The processor 3001, memory 3002, and transceiver 3003 may be connected via a communication bus. The memory 3002 stores computer-readable instructions, which, when executed by the processor 3001, implement the steps of the estimation method described above.

[0155] In a specific implementation, as one example, the processor 3001 may include one or more CPUs, for example... Figure 5 CPU0 and CPU1 are shown in the diagram.

[0156] In a specific implementation, as one example, the electronic device 300 may also include multiple processors, for example... Figure 5 The processors 3001 and 3004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, "processor" can refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0157] The memory 3002 is used to store the software program that executes the present invention, and is controlled by the processor 3001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0158] The transceiver 3003 is used to communicate with network devices or with terminal devices.

[0159] Optionally, the transceiver 3003 may include a receiver and a transmitter. The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0160] Optionally, the transceiver 3003 can be integrated with the processor 3001 or exist independently and be coupled to the processor 3001 through the interface circuit of the electronic device 300. This embodiment of the invention does not specifically limit this.

[0161] It should be noted that, Figure 5 The structure of the electronic device 300 shown is not intended to limit the electronic device. Actual electronic devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects of the electronic device 300 can be understood by referring to the technical effects of the above-described method embodiments, and will not be repeated here.

[0162] In an exemplary embodiment, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the steps of the estimation method described above. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0163] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0164] The use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0165] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0166] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0167] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0168] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0169] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0170] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0171] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0172] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0173] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for estimating time-series features, characterized in that, Includes the following steps: S1. Collect channel state information time series data, and filter the time series data using a window function-based weighted moving average filter; Step S1 specifically includes: Let the input signal be ,in Represents a time step, for each time step Calculate the local standard deviation : ； in It refers to the window size. It is half the size of the window. It is the mean of the signal within the window; based on the calculated local standard deviation. and predefined thresholds To adjust the window size ;if Choose a smaller window size ;if Choose a larger window size The adjustment process is represented by the following formula: ； Use the adjusted window size Calculate the current time step weighted average : ； This is the value of the filter's output signal; S2. For the filtered time-series data, a deep learning framework combining depthwise separable convolution, dilated convolution, and Transformer model is used to extract local features, and the multi-head self-attention mechanism and feedforward network in the Transformer model are used to capture the long-term dependencies between the local features. Step S2 specifically includes: set up The filtered time series data has the following shape: ,in For batch size, Given the sequence length, the processing order of the deep learning framework is as follows: S21, regarding time series data Perform batch standardization: ； in, Representative batch standardization layer; S22, Standardize the batch Input depth can separate combinations of convolution and standard convolution: ； in, Represents a depthwise separable convolutional layer. Represents the activation function. Represents pooling operations; S23, will Input dilated convolution: ； in, This indicates a dilated convolution that expands the receptive field; S24, will Input adaptation layer transformation: ； in, It is the weight matrix of the adaptation layer. It is a bias term. It is an operation that flattens the output of a multidimensional convolution into a one-dimensional object; S25, will Input Transformer model: ； in, The model is used to capture long-term dependencies; S26, will Input to a fully connected layer and output: ； in This is the final output prediction result. It is the weight matrix of the fully connected layer. It is a bias term; The forward propagation of the deep learning framework is represented by the following composite function: ； S3. The deep learning framework is optimized for offline learning and online application using an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning. S4. Based on the optimized deep learning framework, estimate the time-series features of the input channel state information time-series data and establish a fingerprint database.

2. The estimation method according to claim 1, characterized in that, Step S3 specifically includes: S31. Use Bayesian optimization to select the optimal combination of hyperparameters; S32. Evaluate the performance of each optimal hyperparameter combination selected by Bayesian optimization through five-fold cross-validation; S33. Improve the adaptability of the deep learning framework to new environments through few-shot learning.

3. The estimation method according to claim 2, characterized in that, Step S31 specifically includes: Let the hyperparameter space be For hyperparameter space Hyperparameter combination in Find the optimal combination of hyperparameters using Bayesian optimization. This makes the performance indicators Maximize or minimize: ； or, 。 4. The estimation method according to claim 3, characterized in that, Step S32 specifically includes: Each optimal hyperparameter combination selected by Bayesian optimization was verified using five-fold cross-validation. performance Conduct an assessment: ； in, Indicates the first i Using the optimal hyperparameter combination The performance of the trained model.

5. The estimation method according to claim 4, characterized in that, Step S33 specifically includes: set up For a small dataset in a new environment, the deep learning framework's adaptability to the new environment is improved through few-shot learning: ； in, It is a loss function used to measure the performance of the deep learning framework. Performance on the surface.

6. A time-series feature estimation system, said system being used to implement the estimation method as described in any one of claims 1 to 5, characterized in that, The system includes: The data processing module is used to collect channel state information time-series data and to filter the time-series data using a window function-based weighted moving average filter. The deep learning module is used to extract local features from filtered time-series data using a deep learning framework that combines depthwise separable convolution, dilated convolution, and Transformer models. It also uses the multi-head self-attention mechanism and feedforward network in the Transformer model to capture long-term dependencies between these local features. An adaptive fine-tuning module is used to perform offline learning and online application optimization of the deep learning framework using an adaptive fine-tuning algorithm based on Bayesian optimization, five-fold cross-validation, hyperparameter tuning, and few-shot learning. The estimation module is used to estimate the time-series features of the input channel state information time-series data based on the optimized deep learning framework and to establish a fingerprint database.

7. An electronic device, characterized in that, The electronic device includes: processor; A memory storing computer-readable instructions that, when loaded and executed by the processor, implement the estimation method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code that can be invoked by a processor to execute the estimation method as described in any one of claims 1 to 5.

Images