Time series data segmentation method and device in wireless network field and electronic equipment

Abstract

The invention relates to the technical field of IT support, and provides a time series data segmentation method and device and electronic equipment in the field of wireless networks, and the method comprises the steps: obtaining to-be-processed time series data in a wireless network and external context data synchronized in time, carrying out the feature fusion at a corresponding time point, and generating a multi-dimensional feature vector sequence; determining a plurality of candidate segmentation points and corresponding local complexity indexes, and dividing the multi-dimensional feature vector sequence into at least one high-complexity region and one low-complexity region; and generating a data segment with a shorter length in the high-complexity region, and generating a data segment with a longer length in the low-complexity region, thereby obtaining a group of structured data segments with a dynamic length. According to the method, external factors are introduced as input, so that the model can comprehensively consider the time sequence data and related environmental conditions, the multi-scale characteristics of the data can be better captured, dynamic segmentation is realized, and the prediction accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of IT support technology, and in particular to time-series data segmentation methods, apparatus and electronic devices in the field of wireless networks. Background Technology

[0002] In today's digital age, wireless network technology has been widely applied and popularized, carrying a large number of data transmission and interaction tasks, from smart homes to the Industrial Internet of Things. However, the wireless network environment is complex and ever-changing, affected by various factors such as signal interference, fluctuations in user behavior, and differences in device performance, resulting in highly complex and dynamic network data.

[0003] Most existing data processing methods use fixed-length data segments to process time series data. This approach does not take into account the differences in data complexity across different time periods and cannot be flexibly adjusted according to actual data changes. In areas of high data complexity, fixed-length data segments may fail to accurately capture rapid changes in the data, leading to the loss of important information; while in areas of low data complexity, excessively long data segments increase the burden of storage and processing, reducing processing efficiency.

[0004] Due to a lack of effective analysis and regional segmentation of data complexity, existing data analysis methods struggle to employ different processing strategies for data of varying complexity, failing to fully leverage the advantages of various analytical algorithms and models, thus impacting the accuracy and efficiency of data analysis. When faced with complex and ever-changing data in wireless networks, existing technologies are no longer sufficient to meet practical needs. Summary of the Invention

[0005] This application provides a method, apparatus, and electronic device for segmenting time series data in the field of wireless networks, aiming to achieve efficient processing and analysis of multivariate time series structured data in the field of wireless networks, and improve the accuracy and adaptability of time series forecasting.

[0006] This application provides a time-series data segmentation method in the field of wireless networks, comprising: acquiring time-series data to be processed in the wireless network and external context data synchronized with the time-series data in time; fusing the external context data and the time-series data at corresponding time points to generate a multi-dimensional feature vector sequence; determining multiple candidate segmentation points in the multi-dimensional feature vector sequence and their corresponding local complexity indices; dividing the multi-dimensional feature vector sequence into at least one high-complexity region and one low-complexity region according to the local complexity indices; generating shorter data segments in the high-complexity region and longer data segments in the low-complexity region, thereby obtaining a set of dynamically lengthed structured data segments.

[0007] According to the time series data segmentation method in the field of wireless networks provided in this application, multiple candidate segmentation points and corresponding local complexity indices in a multidimensional feature vector sequence are determined, including: based on the multidimensional feature vector sequence, a mutation detection algorithm and a pre-trained machine learning classifier are used to identify multiple candidate segmentation points in the sequence; and the local complexity index of each segmentation point among the multiple candidate segmentation points is calculated.

[0008] According to the time-series data segmentation method in the field of wireless networks provided in this application, the external context data includes at least one of weather information, holiday information, network configuration information, and user information.

[0009] According to the time series data segmentation method in the field of wireless networks provided in this application, the local complexity index is an indicator used to characterize the rate of change, volatility, or information entropy of a multidimensional feature vector sequence within a local time window.

[0010] According to the time series data segmentation method in the field of wireless networks provided in this application, a multidimensional feature vector sequence is divided into at least one high-complexity region and one low-complexity region based on a local complexity index, including: setting a complexity threshold; identifying regions with local complexity indices higher than the complexity threshold as high-complexity regions; and identifying regions with local complexity indices lower than or equal to the complexity threshold as low-complexity regions.

[0011] According to the time series data segmentation method in the field of wireless networks provided in this application, the external context data and time series data are fused at corresponding time points to generate a multidimensional feature vector sequence, including: mapping the feature vector of the external context data to a low-dimensional space and encoding it into a low-dimensional vector through an embedding layer; and fusing the low-dimensional vector with the time series data according to the corresponding time points to generate a multidimensional feature vector sequence.

[0012] This application also provides a time-series data segmentation device in the field of wireless networks, comprising: an acquisition module for acquiring time-series data to be processed in a wireless network, and external context data synchronized with the time-series data in time; a fusion module for performing feature fusion of the external context data and the time-series data at corresponding time points to generate a multi-dimensional feature vector sequence; an index module for determining multiple candidate segmentation points in the multi-dimensional feature vector sequence and corresponding local complexity indices; a segmentation module for dividing the multi-dimensional feature vector sequence into at least one high-complexity region and one low-complexity region according to the local complexity index; and an output module for generating shorter data segments in the high-complexity region and longer data segments in the low-complexity region, thereby obtaining a set of dynamically lengthed structured data segments.

[0013] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a time-series data segmentation method as described in any of the above-described wireless network fields.

[0014] This application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a time-series data segmentation method in the field of wireless networks as described above.

[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements a time-series data segmentation method as described above in the field of wireless networks.

[0016] This application provides a method, apparatus, and electronic device for time-series data segmentation in the field of wireless networks. The method includes: acquiring time-series data to be processed in a wireless network, and external context data synchronized with the time-series data in time; fusing features of the external context data and the time-series data at corresponding time points to generate a multi-dimensional feature vector sequence; determining multiple candidate segmentation points in the multi-dimensional feature vector sequence and their corresponding local complexity indices; dividing the multi-dimensional feature vector sequence into at least one high-complexity region and one low-complexity region based on the local complexity indices; generating shorter data segments in the high-complexity region and longer data segments in the low-complexity region, thereby obtaining a set of dynamically lengthed structured data segments. Through this approach, this application introduces external factors as part of the model input, enabling the model to comprehensively consider the time-series data and its related environmental conditions, better capture the multi-scale characteristics of the data, achieve dynamic segmentation, and thus improve prediction accuracy. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating a time-series data segmentation method in the field of wireless networks provided in this application embodiment.

[0019] Figure 2 This is the second flowchart illustrating the time-series data segmentation method in the field of wireless networks provided in this application embodiment.

[0020] Figure 3This is a schematic diagram of the structure of a time-series data segmentation device in the field of wireless networks provided in an embodiment of this application.

[0021] Figure 4 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0024] This application provides a method for segmenting time-series data in the field of wireless networks, which is an application of the Transformer model in Long-Term Time Series Forecasting (LTSF).

[0025] In related technologies, time series data is generally divided into fixed-length patches, with each patch serving as an input token. However, this fixed segmentation method has limitations; when processing data with multiple time scale characteristics, it cannot effectively capture all time scale information.

[0026] Related technologies such as PatchTST use Transformer models based on fixed-length segments for long-term time series prediction. However, most of them divide the time series into segments of fixed length or a set of predetermined lengths. This static segmentation method requires expert knowledge to select the optimal segment length and faces challenges in extracting time features and dependencies. Most models are limited by fixed segment lengths and cannot fully capture the diverse characteristics of the data.

[0027] Time series prediction technologies mainly rely on traditional machine learning methods and deep learning models based on the Transformer architecture. While they perform well in many fields, they have significant shortcomings when faced with complex and diverse real-world scenarios. (1) Fixed segment length limitation: Most existing time series forecasting models use fixed-length segments to process input data. While this method simplifies the problem to some extent, it ignores the inherent multi-scale characteristics of time series data. Data from different time periods may have different rates of change and complexities, and fixed-length segments cannot effectively capture these differences.

[0028] (2) Lack of context awareness: Time series data are often affected by external factors, such as holiday effects and weather changes. However, many existing models do not fully consider these external variables and rely solely on the historical data of the time series itself for prediction, which may lead to inaccurate prediction results.

[0029] (3) Single-scale attention mechanism: Although Transformer models improve their ability to handle long-distance dependencies through the self-attention mechanism, most implementations focus on a single time scale. This means that the model may ignore local changes on short time scales, which may be crucial for certain types of predictions.

[0030] (4) Limited Prediction Accuracy: Due to the reasons mentioned above, existing time series prediction models often encounter the problem of decreased accuracy when processing long-series data. Especially when predicting data at more distant future points in time, errors may accumulate, leading to a large deviation between the final prediction results and the actual situation.

[0031] (5) Difficulty in adapting to change: Time series data in the real world are often non-stationary, and the distribution of the data may change over time. Existing models often need to be retrained or fine-tuned when faced with this situation, which increases maintenance costs.

[0032] Based on this, this application proposes a novel time series prediction method that is more flexible, context-aware, and multi-scale adaptive. The performance of the model can be improved through the following aspects: (1) Dynamic segmentation: An algorithm that can automatically adjust the segment length according to the real-time changes of time series data. This method can not only better capture the multi-scale characteristics of the data, but also reduce the information loss caused by fixed-length segments.

[0033] (2) Enhance context awareness: Introduce external factors as part of the model input, so that the model can comprehensively consider time series data and its related environmental conditions, thereby improving the accuracy of prediction.

[0034] (3) Intelligent segmentation mechanism: A segmentation mechanism that can process information at different time scales at the same time, so that the model can pay attention to the trend changes over a long period of time, and will not ignore important details in a short period of time.

[0035] (4) Optimize prediction accuracy: It can improve the accuracy of time series prediction, especially in the prediction of long series data, and reduce the cumulative effect of error.

[0036] Please see Figure 1 , Figure 1 This is one of the flowcharts illustrating a time-series data segmentation method in the field of wireless networks provided in this application embodiment. In this embodiment, the time-series data segmentation method in the field of wireless networks may include steps S110 to S150, each step being as follows: S110: Acquire time-series data to be processed in the wireless network, as well as external context data that is time-synchronized with the time-series data.

[0037] In a wireless network environment, time-series data is network-related data that changes over time, such as network traffic, signal strength, and packet loss rate. It reflects the evolution of the network's operating status over time.

[0038] External context data may include a variety of information related to network usage scenarios. Optionally, external context data may include at least one of weather information, holiday information, network configuration information, and user information, such as the user's geographical location, the type of device used, and time information.

[0039] For example, time-series data can be collected through network monitoring devices, while external context data such as the user device's positioning system and system clock can be obtained, and it can be ensured that they are consistent in timestamps.

[0040] For example, combining time-series data of user geolocation and network traffic can provide a better understanding of network usage in different regions. This helps to gain a more comprehensive understanding of the operational status of the wireless network and user behavior patterns, providing a more sufficient basis for subsequent network optimization and management.

[0041] S120: Perform feature fusion between external context data and time series data at corresponding time points to generate a multidimensional feature vector sequence.

[0042] Single time-series data or external context data contain limited information and are insufficient to comprehensively and accurately describe the true state of a wireless network. This embodiment uses feature fusion to merge data from different sources at corresponding time points, integrating the originally independent time-series data and external context data into a multi-dimensional feature vector. In this way, the data at each time point contains multifaceted information, providing a richer and more accurate reflection of the overall state of the wireless network at that moment.

[0043] For example, vector concatenation can be used to concatenate the feature vectors corresponding to the time series data and the external context data at each time point.

[0044] S130: Determine multiple candidate split points in the multidimensional feature vector sequence and their corresponding local complexity indices.

[0045] The changes in a multidimensional feature vector sequence may vary across different time periods. To analyze the data more precisely, it is necessary to identify suitable segmentation points. Candidate segmentation points are preliminary selections within the multidimensional feature vector sequence that indicate potential data segmentation.

[0046] Local complexity metrics are used to measure the degree of change of a multidimensional feature vector sequence within a certain local region, reflecting the complexity of the data. Optionally, local complexity metrics can be used to characterize the rate of change, volatility, or information entropy of a multidimensional feature vector sequence within a local time window.

[0047] S140: Based on the local complexity index, divide the multidimensional feature vector sequence into at least one high-complexity region and one low-complexity region.

[0048] Based on the local complexity index, multidimensional feature vector sequences can be divided into high-complexity regions and low-complexity regions.

[0049] High complexity regions indicate that the data changes drastically within these regions, which may mean that the wireless network is in an abnormal state, user behavior has changed significantly, or the network is experiencing external interference.

[0050] Low-complexity regions indicate relatively smooth data changes, stable network conditions, and predictable user behavior. This classification helps in employing different processing strategies for regions of varying complexity.

[0051] S150: Generate shorter data segments in high-complexity regions and longer data segments in low-complexity regions to obtain a set of dynamically lengthed structured data segments.

[0052] Regions of varying complexity have different data characteristics, and using fixed-length data segmentation methods cannot well adapt to these differences.

[0053] In high-complexity regions, where data changes rapidly, shorter data segments can capture the details of these changes more precisely, preventing information loss. In low-complexity regions, where data changes slowly, longer data segments can reduce the number of segments, lowering storage and processing overhead. By dynamically adjusting the length of data segments, data of varying complexity can be handled more effectively.

[0054] Based on this, in high-complexity regions, a shorter data segment length is set, and the data is segmented according to this length; in low-complexity regions, a longer data segment length is set, and the data is segmented, ultimately resulting in a set of dynamically lengthed structured data segments.

[0055] This application combines time-series data and external context data through feature fusion, enabling the data at each time point to contain more information. The multi-dimensional feature vector sequence can more accurately reflect the actual situation of the wireless network. Compared to single time-series data, it can explain some complex network phenomena, such as a sudden increase in network traffic during a specific time period, which may be related to the user's location and time of activity. This improves the data's ability to represent the actual situation, making the data analysis results more reliable.

[0056] Furthermore, in this embodiment, a dynamic length data segmentation method is adopted to reasonably allocate storage resources according to data complexity. In areas with low data complexity, longer data segments are used, while in areas with high data complexity, shorter data segments are used, thereby improving overall data processing efficiency.

[0057] Furthermore, this embodiment, by determining candidate segmentation points and local complexity indicators, can accurately identify high-complexity and low-complexity regions in the data. Short segments are used to capture changes and uncover anomalies in high-complexity regions, while long segments are used to reflect overall trends in low-complexity regions, helping to improve the accuracy of data analysis and promptly identify problems. Simultaneously, different analysis algorithms and models are employed for regions of different complexity, leveraging the advantages of each method to further enhance the accuracy and efficiency of data analysis.

[0058] The embodiments of this application provide a time-series data segmentation method in the field of wireless networks. This novel dynamic context-aware intelligent segmentation method integrates intelligent segmentation algorithms, context-aware coding, and dynamic segmentation, providing a valuable supplement to the optimization of the preceding input stage in the current Transformer unified model architecture. This method aims to achieve efficient processing and analysis of multivariate time-series structured data in the field of wireless networks by organically integrating the above three key technologies, laying a technical foundation for improving the accuracy and adaptability of time-series forecasting.

[0059] In some embodiments, the step of determining multiple candidate segmentation points in a multidimensional feature vector sequence and their corresponding local complexity indices may specifically include: Based on a multidimensional feature vector sequence, a mutation detection algorithm and a pre-trained machine learning classifier are used to identify multiple candidate segmentation points in the sequence; the local complexity index of each segmentation point among the multiple candidate segmentation points is calculated.

[0060] In the multidimensional feature vector sequences of wireless networks, sudden changes in data often indicate significant changes in network status, user behavior, etc. The core idea of ​​mutation detection algorithms is to identify the locations where data values ​​suddenly change significantly by analyzing the statistical characteristics of the sequence data.

[0061] Alternatively, mutation detection algorithms include statistical distribution-based methods, such as the CUSUM algorithm, which detects mutations by accumulating the deviation of data; and sliding window-based methods, which determine mutation points by comparing changes in statistical quantities such as the mean and variance of data within the window.

[0062] Machine learning classifiers, trained on large amounts of data, can learn the features of different patterns in a multidimensional feature vector sequence. For wireless network data, the classifier can distinguish between normal states and abnormal states that may lead to segmentation.

[0063] Local complexity is a metric used to measure the complexity of a multidimensional feature vector sequence near a candidate split point. It reflects the degree of drastic change and irregularity of the data within that local region. In wireless networks, high-complexity regions may indicate network interference or frequent changes in user behavior; low-complexity regions indicate a relatively stable network state.

[0064] In the embodiments of this application, the mutation detection algorithm can accurately capture the mutation location of data from the perspective of statistical characteristics, while the machine learning classifier identifies possible segmentation points from the perspective of pattern recognition. Combining the two can fully leverage their respective advantages and improve the accuracy of candidate segmentation point identification; by calculating the local complexity index of each candidate segmentation point, the complexity of the data at different locations can be understood more accurately. This provides a precise basis for subsequently dividing the multidimensional feature vector sequence into high-complexity and low-complexity regions.

[0065] In some embodiments, the step of dividing a multidimensional feature vector sequence into at least one high-complexity region and one low-complexity region based on a local complexity index may specifically include: Set a complexity threshold; define regions with local complexity indices higher than the complexity threshold as high complexity regions; define regions with local complexity indices lower than or equal to the complexity threshold as low complexity regions.

[0066] A complexity threshold is a critical value used to distinguish between high-complexity and low-complexity regions. In the multidimensional feature vector sequence of a wireless network, the complexity of the data varies due to different network scenarios and data characteristics. By setting an appropriate complexity threshold, the entire sequence can be divided into regions of different complexities.

[0067] Specifically, regions with local complexity indices above the threshold indicate drastic data changes and potential network anomalies or sudden changes in user behavior, thus they are classified as high-complexity regions. Conversely, regions with local complexity indices below or equal to the threshold indicate relatively smooth data changes and a more stable network state, thus they are classified as low-complexity regions.

[0068] As described above, this embodiment of the application uses a complexity threshold to divide regions, which can clearly distinguish between high-complexity and low-complexity regions in a multi-dimensional feature vector sequence. This facilitates in-depth analysis of the data characteristics of regions with different complexities.

[0069] In some embodiments, the step of fusing external context data and time series data at corresponding time points to generate a multidimensional feature vector sequence may specifically include: The feature vectors of the external context data are mapped to a low-dimensional space and encoded into low-dimensional vectors through the embedding layer; according to the corresponding time point, the low-dimensional vectors are fused with the time series data to generate a multi-dimensional feature vector sequence.

[0070] Specifically, external context data often has characteristics such as high dimensionality and sparsity. For example, in a wireless network scenario, external context data may include user location information, device type, time information, etc., and this information may have very high dimensionality when represented in the form of feature vectors.

[0071] High-dimensional data not only increases computational complexity but can also lead to model overfitting. The role of the embedding layer is to map high-dimensional feature vectors to a low-dimensional space, where the data representation is more compact and dense. This is because the embedding layer learns a mapping function to transform feature vectors from the original high-dimensional space into vectors in the low-dimensional space, allowing the preservation of important features and semantic information of the data. Simultaneously, low-dimensional vectors reduce data redundancy and improve the efficiency of subsequent processing.

[0072] Optionally, the embedding layer is typically a fully connected neural network layer containing a set of learnable weight matrices. The input external context data feature vector is multiplied by the weight matrix and processed by a non-linear activation function to obtain a low-dimensional vector. During training, the weight matrix is ​​continuously adjusted using a backpropagation algorithm so that the low-dimensional vector can better represent the features of the original data.

[0073] Time-series data reflects the state changes of a wireless network at different points in time, while external context data provides information about the external environment and users corresponding to these points in time. This embodiment fuses low-dimensional vectors and time-series data at corresponding points in time. At each point in time, the low-dimensional vector and time-series data corresponding to that point are concatenated. This allows external context information to be integrated into the time-series data, resulting in a more comprehensive and richer feature representation that can more accurately describe the actual situation of the wireless network.

[0074] In the above embodiments, by mapping external context data to a low-dimensional space and fusing it with time-series data, the generated multi-dimensional feature vector sequence can more comprehensively express the state information of the wireless network and enhance the data feature expression capability. In addition, mapping external context data to a low-dimensional space can reduce data redundancy and noise.

[0075] Please see Figure 2 , Figure 2 This is the second flowchart illustrating the time-series data segmentation method in the field of wireless networks provided in this application embodiment.

[0076] 1) Data preprocessing 1.1) Noise Removal: Use filters (such as Kalman filters) to remove noise from the data to improve data quality and provide more accurate input for subsequent analysis.

[0077] 1.2) Smoothing: Use moving average or exponential smoothing methods to smooth the data, reduce the impact of short-term fluctuations, and highlight long-term trends.

[0078] 1.3) Feature extraction: Extract the basic features of the time series, such as trends and seasonality, to provide key information for the model and help capture the inherent patterns of the data.

[0079] 2) Contextual information integration To enable the model to more accurately capture data patterns and improve prediction accuracy, external contextual information can be integrated with time series data. This allows the model to comprehensively consider the influence of multiple factors in complex data environments. The specific steps are as follows: 2.1) External data collection: Collect external contextual information related to time series data, such as weather data, holiday data, etc., which may have a significant impact on the time series.

[0080] 2.2) Feature transformation: Transforming external information into numerical features, such as converting weather data into features like temperature and humidity, to provide additional explanatory variables for the model.

[0081] 2.3) Feature fusion: External features are added to each time point of the time series data to form a new multidimensional feature vector, which enhances the model's adaptability to complex data environments and its prediction accuracy.

[0082] 3) Intelligent segmentation To accurately segment time series data and effectively capture non-stationary characteristics, a combination of methods will be employed, from initial detection to further confirmation, and then to intelligent judgment using deep learning models, gradually improving the accuracy, reliability, and flexibility of segmentation. The specific steps are as follows: 3.1) CUSUM mutation detection: The CUSUM algorithm is used to detect abrupt changes in time series data. This algorithm accumulates the deviations in the data; when the accumulated deviation exceeds a certain threshold, it is considered to have detected an abrupt change.

[0083] The CUSUM algorithm is used to detect abrupt changes in time series data. These points can serve as potential segmentation boundaries and help capture non-stationary characteristics in the data.

[0084] 3.2) Page-Hinkley Test: The Page-Hinkley test, based on the cumulative sum and standard deviation of the data, determines whether a truly significant change has occurred by setting an appropriate threshold. The Page-Hinkley test further identifies significant changes in the data, enhancing the accuracy and reliability of segmentation.

[0085] 3.3) LSTM binary classifier To further enhance the intelligence and flexibility of segmentation, a deep learning model is introduced. An LSTM (Long Short-Term Memory) network model is trained as a binary classifier to determine whether each time point is suitable as a segmentation point.

[0086] LSTM models possess powerful memory capabilities and the ability to process sequential data, enabling them to learn complex patterns and long-term dependencies in time series data. Through extensive training data, LSTM models can automatically learn which features and patterns indicate a suitable time point as a segmentation point and which do not.

[0087] In practical applications, time-series data is input into a trained LSTM binary classifier. The model classifies each time point and outputs whether that time point is a segmentation point. This deep learning-based method can fully utilize the latent information in the data, improve the intelligent decision-making level of segmentation, and adapt to time-series data of different types and complexities.

[0088] 4) Dynamic segment length adjustment To enable the model to more accurately capture details and trends in time series data, and to improve prediction accuracy and efficiency, the segment length can be dynamically adjusted based on the local complexity of the data. The specific implementation steps are as follows: 4.1) Computational complexity index Time series data varies at different points in time. To measure this variation, the rate of change or volatility is calculated at each time point. These complexity metrics reflect the local complexity of the data.

[0089] 4.2) Divide the complexity region Based on the calculated complexity index, the time series is divided into different regions. Data within each region exhibits similar complexity characteristics. This division allows us to assign appropriate segment lengths to different regions. Such segmentation helps us to make targeted segmentations based on the actual situation of the data, thereby better capturing the details and trends of the data.

[0090] 4.3) Adjust segment length Based on the division of complexity regions, different segment lengths are used for different regions. In high-complexity regions, due to the rapid and complex changes in data, using shorter segment lengths can more closely track data changes and capture more detailed information.

[0091] In regions of low complexity, where the data is relatively stable, using longer segment lengths can reduce computational load, improve processing efficiency, and better capture the overall trend of the data. Through this flexible segment length adjustment strategy, the model can better adapt to changes in the data, improving prediction accuracy and efficiency.

[0092] In some embodiments, when processing time series data, in order to enable the model to understand the data more accurately, mine deeper information in the data, and improve predictive ability, a context-aware coding method can be used, which mainly includes the following three key steps: First, define the embedding layer. An embedding layer maps feature vectors to a low-dimensional space. This helps the model learn and represent complex feature relationships more efficiently, enabling it to better capture the intrinsic connections between features.

[0093] Secondly, feature encoding. An embedding layer encodes the feature vectors into low-dimensional vectors. This step not only helps extract deep features from the data and uncover hidden information, but also reduces the computational burden on the model and improves its efficiency.

[0094] Finally, feature fusion is performed. The encoded feature vectors are fused with the time series data to form a richer data representation. This fusion method enhances the model's understanding and predictive capabilities, enabling it to perform better in handling real-world problems.

[0095] In some embodiments, the following operations may be performed during the model training and prediction phases: The first step is to prepare the training data by dividing the time series data into training and test sets to ensure that the model can be validated and evaluated on independent data.

[0096] Next, model building is carried out, which involves defining the model's structure, including the input layer, hidden layers, and output layer. Simultaneously, based on the specific data characteristics and prediction task, an appropriate network structure and activation function must be selected to enable the model to better adapt to the data characteristics and accurately capture the patterns and regularities within the data.

[0097] Then, the model training phase begins, where the model is trained using the training set data, and the weights are updated and the model parameters are optimized through the backpropagation algorithm.

[0098] Next, model predictions are performed using the trained model, and the prediction results are recorded to evaluate the model's prediction accuracy and generalization ability.

[0099] Finally, cross-validation is performed to evaluate the effectiveness of different strategies and select the optimal model and parameters, thereby ensuring that the model has good stability and reliability.

[0100] For example, the specific implementation steps of the above method are as follows: 1) Input time series data.

[0101] Obtain the raw time series data to be processed .

[0102] 2) Data preprocessing and feature extraction.

[0103] The raw data undergoes preprocessing and feature extraction. This step aims to improve data quality and extract core information. Specifically, filters such as the Kalman filter are used to denoise the time series data T to remove random noise. Subsequently, a moving average method is used to smooth the denoised data.

[0104] Based on the preprocessing, the inherent basic features of the time series are further extracted, such as trend features that describe the long-term direction of data change and seasonal features that describe the periodic fluctuations of data.

[0105] 3) Context information integration.

[0106] Contextual information integration can incorporate external environmental factors and enhance the accuracy of segmentation.

[0107] Specifically, collect external contextual information that is temporally relevant to time series data, such as weather data and holiday data.

[0108] Then, the collected non-numerical external information is transformed into computable numerical features, such as converting weather conditions into specific values ​​like temperature and humidity.

[0109] Finally, feature fusion is performed, adding external features to each time point of the time series data to form a new multidimensional feature vector, which contains both the original data information and the external context information.

[0110] 4) Intelligent segmentation Intelligent segmentation of time series data involves dividing the time series into different segments by detecting significant changes in the data. Specific detection mechanisms can be divided into: ①The CUSUM mutation detection algorithm has the following calculation formula: ; ; in, and These represent the cumulative sums in the positive and negative directions, respectively. It is the first Data points, It is the mean of the data. It is a threshold.

[0111] ②The Page-Hinkley test, its calculation formula is: ; when When a certain threshold is exceeded, the data is considered to have changed significantly.

[0112] ③LSTM binary classifier Time series data is divided into training and test sets for model training and evaluation. An LSTM network structure is defined, including input, hidden, and output layers to process the sequence data. The LSTM model is trained using the training set data, updating the model's weights through backpropagation to allow the model to learn patterns and features in the data. The model's performance is evaluated using the test set data, and the best model is selected based on the evaluation results to ensure good performance in practical applications.

[0113] Specifically, a binary classification model containing an input layer, an LSTM hidden layer, and an output layer is trained by constructing a training set from time series segments labeled "mutation" or "non-mutation". After training, the model can classify new data points to determine whether they are mutation points.

[0114] 5) Dynamic segment length adjustment.

[0115] This embodiment further performs dynamic segment length adjustment. This step adaptively optimizes the segment granularity based on the complexity of the data itself.

[0116] First, calculate the complexity metric for each time point, which can include the rate of change or volatility, as shown in the following example: ; ; Subsequently, based on the calculated complexity index, the entire time series was divided into several high-complexity regions and low-complexity regions.

[0117] Finally, the segmentation strategy is adjusted according to the regional attributes: for high-complexity regions with drastic data fluctuations, shorter segment lengths are used to capture the details of changes in detail; for low-complexity regions with relatively stable data, longer segment lengths are used to reduce redundant calculations and focus on long-term trends.

[0118] In some embodiments, context-aware encoding can transform segmented data incorporating multiple information sources into a format suitable for machine learning models. Specifically, an embedding layer is defined, which maps the high-dimensional multidimensional feature vectors generated in the preceding steps to a low-dimensional dense space of a predetermined dimension. Then, the multidimensional feature vectors within each segment are input into this embedding layer for encoding, generating corresponding low-dimensional feature vectors. Finally, these encoded context feature vectors are fused with the original time-series data segments to form a final data representation rich in contextual information, which can be used for subsequent downstream tasks such as prediction, classification, or anomaly detection.

[0119] The above embodiments of this application provide a time series data segmentation method in the field of wireless networks, which can more effectively process and predict complex multivariate time series data. This method mainly has three core strategies: first, dynamically adjusting the segment length based on the local complexity of the time series data to capture more useful information; second, combining external factors with the time series data and integrating them into the model through encoding to improve prediction accuracy; and third, automatically detecting key points or pattern change points in the time series to segment the sequence. Its specific advantages are as follows: 1. Improve prediction accuracy: The segment length adjustment strategy based on data complexity, as well as the feature encoding technology that combines external context information, enables the model to adapt to the dynamic changes of data and make full use of external information; by dynamically adjusting the segment length and context-aware encoding, the model can more accurately capture the features of time series data, thereby improving prediction accuracy.

[0120] 2. Enhanced adaptability and flexibility: A dynamic segment length adjustment strategy based on local data complexity is adopted, enabling the model to automatically adjust the segmentation strategy according to the actual changes in the data, thereby better adapting to different data characteristics; the model can automatically adjust the segment length to flexibly adapt to the characteristics of different time series data, improving the model's adaptability and flexibility.

[0121] 3. Enhanced Context Awareness: Context-aware coding techniques are adopted, including external data collection, feature transformation and feature fusion, as well as embedding layer coding, which enhances the model's adaptability to complex data environments. By introducing external factors as part of the model input, the model can comprehensively consider time series data and its related environmental conditions, thereby improving the accuracy of predictions.

[0122] 4. Optimize computational efficiency: By combining statistical mutation detection, deep learning models, and unsupervised learning techniques, the intelligence and automation of the segmentation algorithm are improved, reducing manual intervention and the consumption of computational resources. Through the intelligent segmentation algorithm, the model can automatically detect key change points in the time series, reducing unnecessary calculations and improving the computational efficiency of the model.

[0123] This application also provides a time series data segmentation device in the field of wireless networks. The time series data segmentation device in the field of wireless networks provided in this application is described below. The time series data segmentation device in the field of wireless networks described below can be referred to in correspondence with the time series data segmentation method in the field of wireless networks described above.

[0124] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a time series data segmentation device in the field of wireless networks provided in an embodiment of this application. In this embodiment, the time series data segmentation device in the field of wireless networks may include an acquisition module 310, a fusion module 320, an index module 330, a segmentation module 340, and an output module 350.

[0125] The acquisition module 310 is used to acquire time-series data to be processed in the wireless network, as well as external context data that is time-synchronized with the time-series data.

[0126] The fusion module 320 is used to fuse external context data and time series data at corresponding time points to generate a multidimensional feature vector sequence.

[0127] The indicator module 330 is used to determine multiple candidate segmentation points in a multidimensional feature vector sequence and their corresponding local complexity indicators.

[0128] The segmentation module 340 is used to divide a multidimensional feature vector sequence into at least one high-complexity region and one low-complexity region based on the local complexity index.

[0129] Output module 350 is used to generate shorter data segments in high-complexity regions and longer data segments in low-complexity regions, thereby obtaining a set of dynamically lengthed structured data segments.

[0130] In some embodiments, the index module 330 can be specifically used to: identify multiple candidate segmentation points in the sequence based on a multidimensional feature vector sequence, using a mutation detection algorithm and a pre-trained machine learning classifier; and calculate the local complexity index of each segmentation point among the multiple candidate segmentation points.

[0131] In some embodiments, external context data includes at least one of weather information, holiday information, network configuration information, and user information.

[0132] In some embodiments, the local complexity metric is an indicator used to characterize the rate of change, volatility, or information entropy of a multidimensional feature vector sequence within a local time window.

[0133] In some embodiments, the segmentation module 340 may be specifically used to: set a complexity threshold; determine regions with local complexity indices higher than the complexity threshold as high complexity regions; and determine regions with local complexity indices lower than or equal to the complexity threshold as low complexity regions.

[0134] In some embodiments, the fusion module 320 may specifically be used for: The feature vectors of the external context data are mapped to a low-dimensional space and encoded into low-dimensional vectors through the embedding layer; according to the corresponding time point, the low-dimensional vectors are fused with the time series data to generate a multi-dimensional feature vector sequence.

[0135] On the other hand, this application also provides an electronic device, please refer to... Figure 4 , Figure 4 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of this application, such as... Figure 4As shown, the electronic device may include memory 420, processor 410, and a computer program stored in memory 420 and executable on processor 410. When processor 410 executes the program, it can implement a time-series data segmentation method in the field of wireless networks. This method may include: The process involves acquiring time-series data to be processed from a wireless network, along with external context data synchronized with the time-series data in time; fusing the external context data and time-series data at corresponding time points to generate a multidimensional feature vector sequence; determining multiple candidate segmentation points and corresponding local complexity indices in the multidimensional feature vector sequence; dividing the multidimensional feature vector sequence into at least one high-complexity region and one low-complexity region based on the local complexity indices; generating shorter data segments in the high-complexity region and longer data segments in the low-complexity region, thereby obtaining a set of dynamically lengthed structured data segments.

[0136] Optionally, the electronic device may further include a communication bus 430 and a communication interface 440, wherein the processor 410, the communication interface 440, and the memory 420 communicate with each other through the communication bus 430. The processor 410 can call the computer program in the memory 420 to execute the time-series data segmentation method in the field of wireless networks provided by the above methods.

[0137] Furthermore, the logical instructions in the aforementioned memory 420 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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 this application. 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.

[0138] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the time-series data segmentation method in the field of wireless networks provided by the above methods. The steps and principles of the method have been described in detail in the above methods and will not be repeated here.

[0139] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the time-series data segmentation method in the field of wireless networks provided by the above methods. The steps and principles of this method have been described in detail in the above methods and will not be repeated here.

[0140] Non-transitory computer-readable storage media can be any available medium or data storage device that can be accessed by a processor, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0141] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0142] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for segmenting time-series data in the field of wireless networks, characterized in that, include: Acquire time-series data to be processed in the wireless network, as well as external context data that is time-synchronized with the time-series data; The external context data and the time series data are fused at corresponding time points to generate a multidimensional feature vector sequence. Determine multiple candidate segmentation points and their corresponding local complexity indices in the multidimensional feature vector sequence; Based on the local complexity index, the multidimensional feature vector sequence is divided into at least one high-complexity region and one low-complexity region. Shorter data segments are generated in the high-complexity region, and longer data segments are generated in the low-complexity region, thus obtaining a set of dynamically lengthped structured data segments.

2. The time-series data segmentation method in the field of wireless networks according to claim 1, characterized in that, Determining multiple candidate segmentation points and their corresponding local complexity indices in the multidimensional feature vector sequence includes: Based on the multidimensional feature vector sequence, a mutation detection algorithm and a pre-trained machine learning classifier are used to identify multiple candidate segmentation points in the sequence. Calculate the local complexity index of each of the multiple candidate segmentation points.

3. The time-series data segmentation method in the field of wireless networks according to claim 1, characterized in that, The external context data includes at least one of weather information, holiday information, network configuration information, and user information.

4. The time-series data segmentation method in the field of wireless networks according to any one of claims 1 to 3, characterized in that, The local complexity index is an indicator used to characterize the rate of change, volatility, or information entropy of the multidimensional feature vector sequence within a local time window.

5. The time-series data segmentation method in the field of wireless networks according to claim 1, characterized in that, The step of dividing the multidimensional feature vector sequence into at least one high-complexity region and one low-complexity region based on the local complexity index includes: Set a complexity threshold; Regions with local complexity indices higher than the complexity threshold are defined as high-complexity regions; Regions with local complexity indices lower than or equal to the complexity threshold are defined as low-complexity regions.

6. The time-series data segmentation method in the field of wireless networks according to claim 1, characterized in that, The step of fusing the external context data and the time series data at corresponding time points to generate a multidimensional feature vector sequence includes: The feature vectors of the external context data are mapped to a low-dimensional space and encoded into low-dimensional vectors through an embedding layer. Based on the corresponding time point, the low-dimensional vector is fused with the time series data to generate the multi-dimensional feature vector sequence.

7. A time-series data segmentation device in the field of wireless networks, characterized in that, include: The acquisition module is used to acquire time-series data to be processed in the wireless network, as well as external context data that is time-synchronized with the time-series data; The fusion module is used to perform feature fusion between the external context data and the time series data at corresponding time points to generate a multi-dimensional feature vector sequence. The indicator module is used to determine multiple candidate segmentation points and their corresponding local complexity indicators in the multidimensional feature vector sequence. The segmentation module is used to divide the multidimensional feature vector sequence into at least one high-complexity region and one low-complexity region according to the local complexity index. The output module is used to generate shorter data segments in the high-complexity region and longer data segments in the low-complexity region, thereby obtaining a set of dynamically lengthed structured data segments.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the time-series data segmentation method in the field of wireless networks as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the time-series data segmentation method in the field of wireless networks as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the time-series data segmentation method in the field of wireless networks as described in any one of claims 1 to 6.

Images