Edge cloud data co-processing method based on dynamic sliding window and federal learning

Abstract

The invention discloses an edge cloud data co-processing method based on a dynamic sliding window and federal learning, and particularly relates to the technical field of edge computing and data co-processing. Constructing a timestamp sliding window, adaptively adjusting the length of the window based on local sudden change intensity, extracting a key state segment, and generating a local feature abstract stream; inputting the feature abstract into a local federal sub-model, calculating a credible feature response value, constructing a collaborative sparse graph to recognize abnormal nodes, and triggering increment to retrain; lightweight model updating and feature abstracts are uploaded under bandwidth constraints, and cloud fusion constructs a global model and issues the global model; adjusting the next round of sliding window parameters according to the node abrupt change rate and the network state; the method can effectively improve the model training precision and the edge cloud cooperation efficiency, and has the advantages of low bandwidth, high dynamic and self-adaption.

Description

Technical Field

[0001] This invention relates to the field of edge computing and data collaborative processing technology, specifically to an edge-cloud data collaborative processing method based on dynamic sliding window and federated learning. Background Technology

[0002] Currently, in applications such as the Industrial Internet, smart manufacturing, and remote monitoring, a large number of edge nodes are typically deployed to collect and preprocess production data in real time. Because these edge nodes often operate in complex environments with limited network bandwidth and unstable communication latency, traditional centralized data processing solutions struggle to balance communication overhead with processing timeliness.

[0003] On the one hand, directly uploading all the raw data from edge nodes to the cloud not only causes severe bandwidth congestion but also poses risks to data security and privacy. On the other hand, data synchronization mechanisms based on static strategies cannot adapt to sudden situations in industrial settings (such as equipment malfunctions or sudden events), making it difficult to achieve efficient, dynamic, and differentiated data collaboration.

[0004] Furthermore, in the process of collaborative learning among multiple edge nodes, factors such as different sampling frequencies and misaligned data timing often lead to the accumulation of global model training errors, which seriously affects the federated learning effect. Summary of the Invention

[0005] The purpose of this invention is to provide a collaborative processing method for edge-cloud data based on dynamic sliding windows and federated learning, in order to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for collaborative processing of edge-cloud data based on dynamic sliding window and federated learning, comprising: For the time-series data collected from each edge node, a sliding data window with timestamps is constructed, and the window length is adjusted using a dynamic sliding window adjustment method. Based on the data mutation characteristics within the sliding window, key state segments within each time window are extracted to generate a local feature summary stream F_i; The local feature summary stream F_i is input into the local federated sub-model, and the output of the local federated sub-model is weighted and confidence evaluated to obtain the local credible feature response value R_i. The R_i of all edge nodes are aggregated to construct a collaborative sparse graph, and local data augmentation and retraining requests are triggered based on anomalous mutation nodes in the graph structure. Under the condition of satisfying bandwidth constraints, the selected local federated sub-model update results and local feature summary stream F_i are uploaded to the cloud server, and the global model M_g is constructed by cloud aggregation. The updated M_g is distributed to the edge nodes, and the initial parameters of the next round of sliding window are adjusted according to the node's own mutation rate and network status.

[0007] Preferably, the method of adjusting the window length using a dynamic sliding window adjustment includes: The raw time-series data collected from edge nodes is timestamped, and the data stream is initially segmented with a fixed time granularity. Based on the statistical volatility indicators within each data segment, including the rate of change of the mean, the increase in the standard deviation, and the frequency of mutations, a local mutation intensity vector is constructed. Based on the local mutation intensity vector and the current network bandwidth status of the edge nodes, a bandwidth-sensitive window adjustment algorithm is used to dynamically update the sliding window length.

[0008] Preferably, the step of extracting key state segments within each time window and generating a local feature summary stream F_i includes: Within the data segment covered by the current sliding window, each data point is judged to meet the mutation conditions based on a preset threshold. The mutation conditions include the mutation magnitude threshold where the data increase exceeds three times the standard deviation of the historical mean. Continuous data points that meet the mutation conditions are divided into state segments; For each state segment, feature compression processing is performed, including calculating the average rate of change, the maximum fluctuation range, and the segment duration, to form a feature vector; All feature vectors within the current time window are concatenated in chronological order to generate a local feature summary stream F_i corresponding to the window.

[0009] Preferably, the local feature summary stream F_i is input into the local federated sub-model, and the output of the local federated sub-model is subjected to a weighted confidence evaluation to obtain the local credible feature response value R_i, including: The local feature summary stream F_i generated within the current sliding window is input into the pre-initialized local federated sub-model, which is a lightweight deep neural network structure that uses a one-dimensional convolutional neural network to extract temporal features. Obtain the model's prediction output for F_i and calculate the corresponding confidence score. The confidence score is obtained through the Softmax activation function of the output layer and represents the credibility of the current prediction result. The confidence scores are weighted and adjusted, with the weighting factor calculated based on the fluctuation intensity and duration of the state segments in F_i; The weighted confidence score is used as the local credible feature response value R_i.

[0010] Preferably, the step of summarizing the R_i of all edge nodes to construct a collaborative sparse graph, and triggering local data augmentation and retraining requests based on anomalous mutation nodes in the graph structure, includes: Collect trusted feature response values ​​R_i uploaded by multiple edge nodes, and calculate the edge weights between nodes based on the physical distance, functional similarity and feature response correlation between each node. Construct an undirected graph structure between edge nodes, using edge weights as the weights of connecting edges, and perform sparsification on the graph structure, retaining only edges with weights higher than a set threshold. In sparse graph structures, anomalous mutation regions of nodes are identified, specifically by detecting nodes whose response values ​​differ significantly from those of their neighboring nodes and whose centrality decreases, as potential outliers. For potential outliers, the local original data is extracted and augmented, and incremental retraining of the local sub-model is performed based on the augmented data.

[0011] Preferably, uploading the selected local federated sub-model update results and the local feature summary stream F_i to the cloud server includes: Sparse encoding is performed on the current local federated sub-model parameter update results, retaining the portion of gradient change that exceeds a set change threshold, to form a lightweight model update vector; Based on the current available bandwidth of the edge node, determine whether the minimum upload condition is met. If the available bandwidth is higher than the upload threshold, bind and package the lightweight model update vector with the corresponding local feature summary stream F_i. The packaged results are uploaded to the cloud server, where an aggregation algorithm is used to weight and fuse the model update results uploaded from multiple edge nodes to generate a global model M_g.

[0012] Preferably, the updated M_g is distributed to the edge nodes, and the initial parameters of the next round of sliding window are adjusted according to the node's own mutation rate and network status, including: The global model M_g, aggregated in the cloud, is distributed to each participating edge node; After receiving M_g, the edge node counts the frequency of mutation events in the previous sliding window and calculates the mutation rate index V_i, where the mutation rate is the ratio of the number of data points that meet the mutation judgment conditions to the total number of data points. Collect the network status information of the current edge nodes, including bandwidth B_i and average latency D_i, and establish a joint regulation function of mutation rate and network status; Based on the output of the joint control function, adjust the initial length and moving step size of the sliding window for the next round.

[0013] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention presents a collaborative edge-cloud data processing method based on dynamic sliding windows and federated learning. This method can achieve a balance between data hierarchical filtering, model accuracy assurance, and communication efficiency in complex environments where edge node network bandwidth is extremely limited and state fluctuations are frequent. By constructing a mutation-aware dynamic sliding window mechanism, combined with statistical volatility analysis and adaptive window adjustment strategies, it effectively improves the response capability of edge nodes to local sudden states and reduces the risk of invalid data redundancy and the accumulation of model training errors.

[0014] 2. This invention, by introducing a trusted feature response evaluation and collaborative sparse graph construction mechanism, achieves state collaborative awareness and anomaly identification among edge nodes. This enables dynamic triggering of local data augmentation and incremental retraining operations without increasing communication load. The cloud utilizes a bandwidth-aware compression upload strategy and a differentiated weighted aggregation algorithm to further enhance the stability and generalization ability of the global model in multi-source heterogeneous data environments, making it suitable for various edge-cloud collaborative computing scenarios such as industrial manufacturing, intelligent transportation, and remote monitoring. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] 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 embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] For examples, please refer to Figure 1 As shown in this embodiment, a method for collaborative processing of edge-cloud data based on dynamic sliding window and federated learning includes: For the time-series data collected from each edge node, a sliding data window with timestamps is constructed, and the window length is adjusted using a dynamic sliding window adjustment method.

[0019] First, the raw time-series data collected by the edge nodes is timestamped. Specifically, for each piece of raw data received within the sampling period of the edge nodes, a local timestamp is appended based on the data collection time. The timestamp precision can be set to the millisecond level to ensure that the timing order of the data within the sliding window is accurate and controllable.

[0020] After timestamping, the continuous data stream is initially segmented according to a fixed time granularity. This time granularity is a preset parameter, preferably 5 seconds or 10 seconds, meaning each 5-second or 10-second interval forms an initial data segment. The purpose of this step is to provide a basic data structure for subsequent mutation analysis and dynamic window adjustment. Each data segment contains several timestamped data points, forming an intra-segment dataset.

[0021] Next, for each of the above data segments, statistical volatility indicators are extracted for subsequent mutation intensity modeling. These include the following three statistical indicators: Rate of change of mean: The ratio of the change between the average value of each data dimension in the current data segment and the average value of the previous data segment, reflecting the long-term trend change; Standard deviation increase: The ratio of the standard deviation within the current data segment to the historical average standard deviation, used to assess the degree of abnormality in the volatility. Mutation frequency: The percentage of data points in the current data segment that meet a certain mutation judgment condition (such as exceeding a set threshold ΔT). The mutation judgment threshold ΔT is a value set based on domain experience, such as temperature change exceeding 5 degrees Celsius, voltage fluctuation exceeding 2 volts, etc.

[0022] The three indicators mentioned above are combined to form a three-dimensional numerical vector, denoted as the local mutation intensity vector. This vector is used to represent the comprehensive dynamic characteristics of the current data segment in terms of stability, continuity, and trend of change.

[0023] After obtaining the local mutation intensity vector, the length of the sliding data window is jointly adjusted based on the current network bandwidth status of the edge nodes. The bandwidth status of the edge nodes is determined by continuously monitoring the size of the data packets successfully uploaded by the nodes per second and the transmission latency, denoted as the current bandwidth utilization rate B, in bits per second.

[0024] Based on the mutation intensity vector and bandwidth utilization B, a bandwidth-sensitive window adjustment algorithm is executed. This algorithm is implemented through the following logic: If any dimension of the mutation intensity vector exceeds its set change threshold (e.g., mean change rate exceeds 0.3, standard deviation increase exceeds 0.5, mutation frequency exceeds 20%), and the bandwidth utilization B is lower than the set bandwidth threshold B_min (e.g., 1000 bits per second), then the sliding window length is shortened to the minimum allowed value T_min (e.g., 5 seconds). When the mutation intensity is low and the bandwidth utilization B is higher than the set upper limit B_max (e.g., 8000 bits per second), the window length is expanded to the maximum allowed value T_max (e.g., 60 seconds). In other cases, keep the window length unchanged, or perform linear interpolation adjustments between T_min and T_max.

[0025] Based on the data mutation characteristics within the sliding window, key state fragments within each time window are extracted to generate a local feature summary stream F_i.

[0026] After the sliding window is dynamically adjusted, key state segments are extracted from the data segment currently covered by the sliding window to generate a local feature summary stream for subsequent model input. First, a mutation condition is assessed for all data points within the window. The mutation condition is set as follows: for any data dimension, if the value at a certain time point increases by more than three times the historical standard deviation of that dimension compared to its mean within the historical sliding window, then the data point is considered to have undergone a mutation.

[0027] Specifically, let the historical mean of a certain dimension within the current sliding window be μ, the standard deviation be σ, and the observed value of the current data point be x_i. If the absolute value of x_i minus μ is greater than 3 multiplied by σ, then the following condition is met. If the condition is met, the point is marked as a mutation point. This method can effectively distinguish between normal fluctuations and abnormal mutations, improving the accuracy of state recognition.

[0028] The continuous data points marked as mutation points are divided into several state segments. Continuity refers to the fact that, with consecutive timestamps, multiple data points that meet the mutation conditions are adjacent to each other without interruption. To avoid interference from short-term anomalies or signal noise, a continuous mutation is only considered a valid state segment if its duration exceeds a set minimum segment duration threshold (e.g., 1 second).

[0029] For each valid state segment, feature compression is performed, condensing the original multiple time points into a single feature vector. Feature compression includes the calculation of the following three metrics: Average rate of change: Represents the average speed at which data changes within a state segment, calculated as the sum of the differences between adjacent data points within the segment divided by the duration; Maximum fluctuation range: This represents the difference between the maximum and minimum values ​​within a segment, used to measure the magnitude of the mutation; Fragment duration: the time difference between the start and end times of the state fragment, in seconds.

[0030] The feature compression results corresponding to each state segment are used to form a set of three-dimensional feature vectors, which are then arranged sequentially according to their timestamps. Finally, the feature vectors of all state segments within the current sliding window are concatenated to form the local feature summary stream corresponding to that window, denoted as F_i.

[0031] The local feature summary stream F_i is input into the local federated sub-model, and the output of the local federated sub-model is weighted and confidence evaluated to obtain the local credible feature response value R_i.

[0032] After the local feature summary stream F_i is constructed, it is passed as input to the federated sub-model deployed locally on the edge nodes. The federated sub-model is a lightweight deep neural network structure. To adapt to the computing resource constraints in the edge computing environment, a one-dimensional convolutional neural network structure is preferred. This network includes an input layer, two one-dimensional convolutional layers, a fully connected layer, and an output layer. The convolutional kernel size is set to 3, the stride is 1, and the ReLU activation function is used to extract the local patterns and trends of temporal features in F_i.

[0033] After receiving input F_i, the model performs forward propagation and outputs a corresponding prediction vector, representing the model's judgment of the state within the current time window. To measure the confidence level of the prediction result, a Softmax activation function is applied to the output, normalizing each dimension of the prediction vector to a probability value between [0,1]. The higher the probability, the more confident the model is in its judgment of that category. This confidence score is denoted as C_i, representing the model's basic confidence response to the classification result of the current state.

[0034] To further enhance the model's adaptability to changes in sensitivity to abrupt changes in state, the confidence score C_i needs to be weighted. The weighting factor W_i is generated based on the statistical characteristics of each state segment in the local feature summary stream F_i. Specifically, it involves calculating the weighted product of the fluctuation intensity (i.e., the maximum fluctuation range) and duration of each state segment, averaging the results across all segments to obtain the overall fluctuation index S_i, and then constructing the weighting function. , where α is an empirical adjustment parameter, preferably a floating-point value between 0.5 and 1.

[0035] Multiplying the confidence score C_i by the weighting factor W_i yields the final weighted confidence response value R_i, which represents the credible output of the edge node's judgment on the feature state within the current time window.

[0036] The R_i values ​​of all edge nodes are aggregated to construct a collaborative sparse graph, and local data augmentation and retraining requests are triggered based on anomalous mutation nodes in the graph structure.

[0037] After all edge nodes have completed the calculation of the trusted feature response value R_i for the current round, it is uploaded to the central aggregation processing end. First, all collected R_i values ​​are aggregated and processed, and based on this, the connection weights between edge nodes are calculated as the edge weights for constructing the graph structure. The following three factors are considered comprehensively during the specific calculation process: Physical distance similarity: The closer the geographical locations of edge nodes are, the higher their correlation; modeling is performed using a normalized Euclidean distance inverse proportional function; Functional similarity: The consistency of the types of devices monitored or tasks performed by edge nodes is assessed using classification matching or functional label similarity scoring; Feature response correlation: The Pearson correlation coefficient between the reliable feature response values ​​R_i and R_j of different nodes is calculated to reflect the degree of synchronization of state changes.

[0038] Based on the three indicators mentioned above, an edge weight matrix is ​​constructed to generate an undirected weighted graph structure. Each node in the graph represents an edge node, and each edge represents the state association between two nodes. The edge weight is used to characterize the connection strength. To compress the graph structure and highlight significant relationships, sparsification is performed. Specifically, an edge weight threshold T_e is set, for example, T_e=0.6. Only edges with weights higher than T_e are retained, and the connections of other edges are removed, forming a sparse collaborative graph.

[0039] Based on the above sparse graph structure, anomaly node identification is performed. This identification process includes the following two criteria: Significant difference in response values: The difference between the R_i of a node and the average R_j value of all its neighboring nodes is greater than the set anomaly detection threshold ΔR, for example, ΔR>0.4; Decreased graph structure centrality: In the current graph structure, the degree centrality, eigenvector centrality, or betweenness centrality of this node is significantly lower than the historical average, indicating that its connectivity or transitivity in the local subgraph is weakened.

[0040] Nodes that meet any of the above conditions are marked as potentially anomalous mutation nodes.

[0041] For nodes identified as anomalous mutations, a process of re-extracting and augmenting their local raw data is immediately triggered. This data augmentation includes extending the sliding window time range, increasing the sampling frequency of mutation point density, and introducing historical control fragments to construct an enhanced sample set. After data augmentation, the local feature summary stream is re-input into the local federated sub-model, and incremental retraining is performed while retaining the original model parameters.

[0042] Under the condition of satisfying bandwidth constraints, the selected local federated sub-model update results and local feature summary stream F_i are uploaded to the cloud server, and the global model M_g is constructed by cloud aggregation.

[0043] To reduce communication costs and adapt to the bandwidth limitations of edge nodes, sparse encoding is applied to the model parameter update results generated after each round of local training. Specifically, let ΔP_i be the model parameter change vector obtained by the edge node after local training. The absolute value of the gradient change magnitude of each parameter dimension is calculated, and a change magnitude threshold T_g is set, preferably 10% of the maximum value of the original gradient vector. Only parameter dimensions that satisfy the change magnitude greater than T_g are retained, and the weakly changing parts are discarded to form the sparse vector ΔP_i′, which is the lightweight model update vector.

[0044] After obtaining the lightweight model update vector ΔP_i′, the edge nodes monitor the current network bandwidth status in real time. The bandwidth status can be obtained by the ratio of the total number of bytes of data packets successfully transmitted per unit time to the transmission latency, and the result is denoted as the available bandwidth B_i, in bits per second. A minimum upload threshold B_min is set, preferably 2000 bits per second. When B_i ≥ B_min, the node is considered to meet the upload condition.

[0045] If the upload conditions are met, the lightweight model update vector ΔP_i′ and the local feature summary stream F_i within the current sliding window are encapsulated together into an upload data packet U_i. The encapsulation format adopts a structured binary format, including the model parameter part and the feature summary part, to ensure data integrity and parsing efficiency. U_i is uploaded to the cloud aggregation server through an encrypted transmission channel.

[0046] After receiving uploaded data packets U_i from multiple edge nodes, the cloud extracts ΔP_i′ and performs model aggregation. The aggregation process uses a weighted average algorithm, with each edge node i having an aggregation weight of W_i. W_i is calculated based on the node's historical upload frequency, model update sparsity, and the stability of the feature summary stream F_i. Finally, by calculating the weighted sum of all ΔP_i′ and their corresponding W_i, new global model parameters P_g are generated, serving as the output of the current global model M_g.

[0047] The updated M_g is distributed to the edge nodes, and the initial parameters of the next round of sliding window are adjusted according to the node's own mutation rate and network status.

[0048] After completing this round of edge collaborative training and aggregating the global model parameters M_g by the cloud server, the global model distribution process is initiated. Specifically, the global model M_g generated in the cloud is distributed to all edge nodes participating in this round of training through a secure transmission channel. Upon receiving M_g, each node immediately replaces its local model parameters to synchronously enter the next training cycle.

[0049] After updating model parameters, edge nodes count the number of mutation events that occurred during the previous sliding window run to measure the drastic change in the current node's state. The criteria for determining a mutation event remain consistent with the previously set mutation magnitude threshold: if the change in a data point compared to the historical mean of the sliding window exceeds three standard deviations, it is considered a mutation point. Let the total number of data points be N, and the total number of mutation points be N_v. Then, the mutation rate index Vi is defined as N_v divided by N, i.e. .

[0050] Simultaneously, edge nodes perceive the current communication network status in real time, collecting the available bandwidth B_i (in bits per second) and the average communication latency D_i (in milliseconds). Available bandwidth B_i can be calculated from the total amount of data successfully transmitted per unit time, and average latency D_i is obtained by taking the arithmetic mean of multiple measurements of upload feedback time.

[0051] To achieve adaptive adjustment of the sliding window parameters, a joint control function U_i is constructed based on the mutation rate V_i and the network state indices B_i and D_i. The joint control function takes the following form: α, β, and γ are empirically set positive weighting factors used to balance the impact of mutation intensity and network load factors on window adjustment.

[0052] Based on the value range of the joint control function U_i, the initial length L_i and the step size S_i of the sliding window are dynamically adjusted. If U_i exceeds the upper threshold T_high (e.g., set to 1.0), it indicates that the current node state is changing drastically and the network is unstable. In this case, L_i is reduced (e.g., to 10 seconds) and S_i is reduced (e.g., to 2 seconds). If U_i is below the lower threshold T_low (e.g., set to 0.3), the window length is appropriately increased (e.g., increased to 60 seconds) and the step size (e.g., increased to 15 seconds) to improve processing efficiency and reduce resource consumption.

[0053] The adjusted sliding window parameters will be used in the next round of edge node data acquisition and key state fragment extraction to ensure that the model training process continuously adapts to changes in node states and network conditions, thereby improving the dynamic response capability of edge-cloud collaboration.

[0054] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An edge cloud data collaborative processing method based on a dynamic sliding window and federated learning, characterized in that: The method comprises the following steps: For each edge node, the time series data collected is used to construct a sliding data window with a timestamp, and a dynamic sliding window adjustment method is used to adjust the window length; Based on the data mutation characteristics in the sliding window, the key state segments in each time window are extracted to generate a local feature abstract stream F_i; The local feature abstract stream F_i is input into the local federated sub-model, and the output results of the local federated sub-model are subjected to weighted confidence evaluation to obtain a local trusted feature response value R_i; All R_i of the edge nodes are collected to construct a collaborative sparse graph, and the abnormal mutation nodes in the graph structure trigger local data enhancement and retraining requests; Under the condition of satisfying the bandwidth constraint, the selected local federated sub-model update results and the local feature abstract stream F_i are uploaded to the cloud server, and the cloud server aggregates to construct a global model M_g; The updated M_g is distributed to the edge nodes, and the initial parameters of the next round of sliding window are adjusted according to the mutation rate and network state of the node itself.

2. The edge cloud data collaborative processing method based on dynamic sliding window and federated learning according to claim 1, characterized in that: The method for adjusting the window length by using the dynamic sliding window adjustment method comprises the following steps: The original time series data collected by the edge node is marked with a timestamp, and the data stream is preliminarily segmented at a fixed time granularity; Based on the statistical volatility indicators in each data segment, including the mean change rate, the standard deviation increase and the mutation frequency, a local mutation intensity vector is constructed; According to the local mutation intensity vector and the current network bandwidth state of the edge node, a bandwidth-sensitive window adjustment algorithm is used to dynamically update the sliding window length.

3. The edge cloud data collaborative processing method based on dynamic sliding window and federated learning according to claim 1, characterized in that: The method for extracting the key state segments in each time window to generate a local feature abstract stream F_i comprises the following steps: In the data segment covered by the current sliding window, it is judged whether each data point satisfies the mutation condition based on a preset threshold, and the mutation condition includes a mutation amplitude threshold that the data increase exceeds three times the standard deviation of the historical mean; The continuous data points satisfying the mutation condition are divided into state segments; Each state segment is subjected to feature compression processing, including calculating the average change rate, the maximum fluctuation range and the segment duration to form a feature vector; All feature vectors in the current time window are spliced in time sequence to generate a local feature abstract stream F_i corresponding to the window.

4. The edge cloud data collaborative processing method based on dynamic sliding window and federated learning according to claim 3, characterized in that: The method for inputting the local feature abstract stream F_i into the local federated sub-model and performing weighted confidence evaluation on the output results of the local federated sub-model to obtain a local trusted feature response value R_i comprises the following steps: The local feature abstract stream F_i generated in the current sliding window is input into a pre-initialized local federated sub-model, and the federated sub-model is a lightweight deep neural network structure using a one-dimensional convolutional neural network to extract time series features; The prediction output of the model to F_i is obtained, and the corresponding confidence score is calculated, which is obtained by the Softmax activation function of the output layer and represents the credibility of the current prediction result; The confidence score is subjected to weighted adjustment, and the weight factor is calculated according to the fluctuation intensity and duration of the state segment in F_i; The weighted confidence score is taken as the local trusted feature response value R_i.

5. The edge cloud data collaborative processing method based on dynamic sliding window and federated learning according to claim 4, characterized in that: The R_i of all edge nodes is aggregated to construct a collaborative sparse graph, and a local data enhancement and retraining request is triggered according to the abnormal mutation nodes in the graph structure, including: Collecting the trusted feature response values R_i uploaded by multiple edge nodes, and calculating the edge weight between nodes according to the physical distance, functional similarity and feature response correlation between nodes; Constructing an undirected graph structure between edge nodes, taking the edge weight as the weight of the connecting edge, and performing sparse processing on the graph structure, only retaining the edges with edge weight higher than the set threshold; Identify the node abnormal mutation area in the sparse graph structure, specifically detect the nodes with significant response value difference and center degree decline with adjacent nodes as potential abnormal points; For potential abnormal points, trigger the re-extraction and expansion of local raw data, and perform incremental retraining of the local sub-model based on the enhanced data.

6. The edge cloud data collaborative processing method based on dynamic sliding window and federated learning according to claim 5, characterized in that: The selected local federated sub-model update result and local feature summary stream F_i are uploaded to the cloud server, including: Sparse coding processing is performed on the current local federated sub-model parameter update result, and the part with gradient change amplitude exceeding the set change threshold is retained to form a lightweight model update vector; According to the current available bandwidth of the edge node, it is judged whether the minimum upload condition is met, if the available bandwidth is higher than the upload threshold, the lightweight model update vector and the corresponding local feature summary stream F_i are bound and packaged; The packaged result is uploaded to the cloud server, and the cloud server uses an aggregation algorithm to weight and fuse the model update results uploaded by multiple edge nodes to generate a global model M_g.

7. The edge cloud data collaborative processing method based on dynamic sliding window and federated learning according to claim 1, characterized in that: The updated M_g is distributed to the edge nodes, and the initial parameters of the next round of sliding window are adjusted according to the mutation rate and network state of the node itself, including: The global model M_g aggregated by the cloud server is distributed to each participating edge node; After receiving M_g, the edge node calculates the mutation rate V_i by counting the frequency of mutation events in the last round of sliding window, the mutation rate is the ratio of the number of data points meeting the mutation determination condition to the total number of data points; Collect the network state information of the current edge node, including bandwidth B_i and average delay D_i, and establish a joint regulation function of mutation rate and network state; According to the output result of the joint regulation function, adjust the initial length and moving step of the next round of sliding window.

Images