A method and system for scheduling edge computing tasks in a computing center

Abstract

This invention discloses a deep learning-based method and system for scheduling edge computing tasks in computing centers, relating to the fields of edge computing and intelligent task scheduling technology. By performing data preprocessing, fragment attribution and splicing, and redundancy reduction operations at the edge nodes, this invention effectively reduces redundant computation and repeated transmission, improving the integrity and consistency of task data, thereby providing higher-quality input for unified scheduling in computing centers. Furthermore, by combining deep learning methods to model and prioritize task features, it can dynamically adapt to abnormal fluctuations in task flows, enabling flexible adjustments to path distribution.

Description

Technical Field

[0001] This invention relates to the field of edge computing and intelligent task scheduling technology, and in particular to a method and system for scheduling edge computing tasks in a computing center. Background Technology

[0002] With the rapid development of computing centers and edge computing, real-time processing and task scheduling of large-scale data streams have become a research focus. Traditional computing center scheduling relies heavily on centralized resource allocation strategies, which struggle to handle the high-concurrency data streams and fragmented data segments generated at edge nodes. In recent years, the combination of deep learning and edge intelligence has driven the optimization of task scheduling methods. For example, by utilizing data preprocessing, task hierarchical classification, and intelligent allocation strategies, scheduling efficiency has been improved and network latency has been reduced.

[0003] CN114546608B discloses a task scheduling method based on edge computing, which achieves distributed scheduling updates by maintaining a policy network in each region and combining federated learning and deep reinforcement learning. This method has advantages in model scalability and data privacy protection, but its focus is mainly on the online update of the policy network, without addressing the abnormal fluctuations and fragmentation of input data at the edge. This results in insufficient stability and resource utilization of scheduling results in complex scenarios. Therefore, this method has limitations in handling data fragment attribution and reducing redundancy.

[0004] CN114650321A discloses a task scheduling method and edge computing terminal for edge computing. In the data preprocessing stage, it uses task priority partitioning, combined with a multi-objective optimization model and an improved Cuckoo algorithm for cloud task scheduling, effectively reducing communication overhead and network latency. However, this method focuses on optimizing task priority and overall completion time, lacking the identification and processing of fragmented segments and abnormal parameters within the edge node data stream, which can easily reduce scheduling efficiency when there is too much redundant data. Furthermore, this method does not involve differential judgment based on traceability parameters, making it difficult to ensure the robustness of scheduling results in scenarios with complex and fluctuating task flows. Summary of the Invention

[0005] In view of the problems existing in current edge computing task scheduling technology, this invention is proposed.

[0006] Therefore, the problem to be solved by this invention is how to solve the problems of high task data redundancy, unreasonable scheduling paths and unstable scheduling results in the prior art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for scheduling edge computing tasks in a computing center, comprising: edge nodes preprocessing received task data streams, extracting abnormal fluctuation parameters and fragmentation structure parameters of the task data streams, and outputting an initial task data set with annotations; based on the initial task data set, performing attribution analysis on fragmented data segments, splicing together segments with the same origin, and generating a spliced ​​task data set; comparing the spliced ​​task data set with other task data sets to be scheduled in the edge nodes, identifying duplicate or similar data segments, and generating a target task data set after redundancy reduction; inputting the target task data set into the computing center scheduling, sorting the task distribution paths according to the abnormal fluctuation parameters and redundancy reduction results, and generating a task scheduling scheme.

[0008] As a preferred embodiment of the edge computing task scheduling method for computing power centers described in this invention, the preprocessing of the received task data stream by the edge node includes: dividing the received task data stream into continuous time segments according to timestamps, generating a preprocessed data set with time indexes; detecting time segments whose fluctuation amplitude exceeds a set range based on the fluctuation amplitude of the time segments in the preprocessed data set, and buffering the time segments separately to generate a data set marked as abnormal segments; performing format difference detection on the data set marked as abnormal segments and the data set of other normal segments, and performing unified encoding on the edge node side to output a preprocessed data set with a consistent structure.

[0009] As a preferred embodiment of the edge computing task scheduling method of the computing power center described in this invention, the extraction of abnormal fluctuation parameters and fragmentation structure parameters of the task data stream includes: extracting numerical differences of continuous time segments from the preprocessed data set, calculating the statistical distribution of the differences, and generating an abnormal fluctuation parameter vector; marking the loss points and overlap points between adjacent data segments based on the time index of the preprocessed data set, and generating a fragmentation structure boundary set; mapping the abnormal fluctuation parameter vector and the fragmentation structure boundary set to form a feature set that simultaneously includes fluctuation and fragmentation attribution.

[0010] As a preferred embodiment of the edge computing task scheduling method of the computing power center described in this invention, the following steps are included: performing attribution analysis on fragmented data segments: marking the source of the fragmented structure boundary set in the feature set, and comparing adjacent segments based on time index and data content features to generate a fragment attribution candidate set; and combining the abnormal fluctuation parameter vector to remove segments in the fragment attribution candidate set whose fluctuation differences exceed the fluctuation threshold, thereby generating an attribution set.

[0011] As a preferred embodiment of the edge computing task scheduling method of the computing power center described in this invention, the step of splicing together the same source segments and generating a spliced ​​task data set includes: rearranging the same source segments in the belonging set according to the time index order to generate a splicing order set; inserting markers or approximate padding values ​​into adjacent segments with missing points in the splicing order set to generate a spliced ​​segment set with enhanced continuity; and integrating the spliced ​​segment set according to the source marker to generate the spliced ​​task data set.

[0012] As a preferred embodiment of the edge computing task scheduling method for computing power centers described in this invention, the following steps are included: identifying duplicate or similar data segments: performing index matching between the spliced ​​task data set and other task data sets to be scheduled within the edge nodes, and extracting data segments with the same time index range; calculating the difference between the feature sets of the corresponding data segments, and marking data segments with difference values ​​lower than a preset difference threshold as similar segments, and outputting a set of similar segments; generating a target task data set after redundancy reduction includes: removing duplicate segments from the set of similar segments, retaining a single copy with a source tag, and generating a deduplication set; integrating the deduplication set and the remaining unmarked segments in the spliced ​​task data set to generate the target task data set after redundancy reduction.

[0013] As a preferred embodiment of the edge computing task scheduling method for computing power centers described in this invention, the generation of the task scheduling scheme includes: mapping the target task data set and the abnormal fluctuation parameter vector to generate a task feature combination set containing fluctuation characteristics and redundancy reduction markers; prioritizing candidate task distribution paths according to the fluctuation amplitude level and the number of redundancy markers in the task feature combination set to generate a path set with sorting weights; and allocating the path set to the target task data set according to priority order to generate and output the final task scheduling scheme.

[0014] Secondly, the present invention provides an edge computing task scheduling system for computing power centers, comprising: The preprocessing module is used by edge nodes to preprocess the received task data stream, extract abnormal fluctuation parameters and fragmentation structure parameters of the task data stream, and output an initial task data set with annotations. The splicing module is used to perform attribution analysis on fragmented data segments based on the initial task data set, splice segments with the same source, and generate a spliced ​​task data set. The redundancy reduction module is used to compare the spliced ​​task data set with other task data sets to be scheduled in edge nodes, identify duplicate or similar data segments, and generate a target task data set after redundancy reduction. The scheduling module is used to input the target task data set into the computing power center for scheduling, sort the task distribution paths according to the abnormal fluctuation parameters and redundancy reduction results, and generate a task scheduling scheme.

[0015] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the edge computing task scheduling method for computing centers as described in the first aspect of the present invention.

[0016] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the edge computing task scheduling method for computing centers as described in the first aspect of the present invention.

[0017] The beneficial effects of this invention are as follows: By performing data preprocessing, fragment attribution and splicing, and redundancy reduction operations at the edge nodes, this invention effectively reduces redundant calculations and repeated transmissions, improves the integrity and consistency of task data, and thus provides higher quality input for the unified scheduling of computing centers. On this basis, by combining deep learning methods to model and prioritize task features, it can dynamically adapt to the abnormal fluctuation characteristics of task flow and realize flexible adjustment of path distribution.

[0018] Overall, this invention not only improves the collaboration efficiency between the edge and the center, reduces the waste of system resources and scheduling latency, but also enhances robustness and adaptability in high-concurrency and complex task scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the edge computing task scheduling method for computing power centers.

[0021] Figure 2 This is a structural diagram of the edge computing task scheduling system of the computing center. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Figure 1 This is a flowchart of a computing center edge computing task scheduling method according to an embodiment of the present invention. Figure 1 As shown, the edge computing task scheduling method in the computing center includes: S1: The edge node preprocesses the received task data stream, extracts the abnormal fluctuation parameters and fragmentation structure parameters of the task data stream, and outputs an initial task data set with annotations.

[0026] S1.1: The edge node divides the received task data stream into continuous time segments according to the timestamp, and generates a preprocessed data set with time index.

[0027] Specifically, the edge node first reads the received task data stream one by one and extracts the timestamp field from each record; if the data stream does not have a timestamp field, the edge node assigns a monotonically increasing reception time to the record at the time of reception to ensure the integrity of the timing.

[0028] Secondly, a time window is set, and the task data stream is divided into continuous time segments using the time window as the unit of division, with the start and end times of each time segment not overlapping; finally, each time segment is stored as a matrix structure, and a unified time index field is added to the structure to form a preprocessed data set with a time index.

[0029] As can be seen, this invention ensures that the boundaries between time segments strictly correspond to the actual time range by using a timestamp-based division method, so that subsequent fluctuation detection and fragmentation analysis can be based on time as a unified benchmark.

[0030] S1.2: Based on the fluctuation amplitude of time series segments in the preprocessed data set, detect time series segments whose fluctuation amplitude exceeds the set range, buffer the time series segments separately, and generate a data set that marks abnormal segments.

[0031] Because existing technologies often only perform anomaly detection at the computing center stage, anomalies cannot be detected in advance at the edge nodes. Abnormal data is mixed with normal data during transmission and encoding, which increases the complexity of subsequent processing.

[0032] The specific operations of this invention include: within each time segment, calculating the maximum and minimum values ​​of the data sequence, with the fluctuation amplitude defined as the difference between the two; edge nodes setting a dynamic threshold based on the statistical results of historical data, for example, the mean plus twice the standard deviation. When the fluctuation amplitude of a certain time segment exceeds the dynamic threshold, the corresponding time segment is marked as an abnormal segment and stored in the abnormal buffer set; otherwise, it is stored in the normal data set, and the set of marked abnormal segments and the set of normal segments are output.

[0033] S1.3: Perform format difference detection on the data set of the marked abnormal segments and the data set of the remaining normal segments, and perform unified encoding on the edge node side to output a preprocessed data set with consistent structure.

[0034] Specifically, the edge nodes sequentially check the data field types of the abnormal segment set and the normal segment set, such as integer, floating-point, and time types. If there are differences in the fields, a unified mapping is performed to unify them into floating-point numeric fields and ISO standard format time fields. At the same time, a binary field is added to the unified data structure. If the segment is an abnormal segment, the value is 1; otherwise, the value is 0. Finally, the abnormal segment set and the normal segment set are merged to output a preprocessed data set with a consistent structure.

[0035] S1.4: Extract numerical differences of continuous time segments from the preprocessed dataset, calculate the statistical distribution of the differences, and generate anomaly fluctuation parameter vectors.

[0036] The calculation of numerical differencing involves: within each time series segment, calculating the differences between adjacent data points sequentially to obtain a differencing sequence, which represents the difference between subsequent data points and preceding data points. Subsequently, statistics are calculated for the differencing sequence, including mean, variance, skewness, and kurtosis, and these statistics are combined into a parameter vector of [mean, variance, skewness, kurtosis]. Finally, the parameter vector is written into the corresponding time series record, forming the abnormal fluctuation parameter field.

[0037] S1.5: Based on the time index of the preprocessed data set, mark the missing points and overlapping points between adjacent data segments to generate a fragmented structure boundary set.

[0038] The confirmation of lost points includes the following steps: Edge nodes compare the end time of adjacent time segments with the start time of the next time segment. If the end time plus the time window is less than the start time, it is determined that there is a lost point at the corresponding location, and the boundary information is recorded.

[0039] The confirmation of overlapping points includes: if the end time is greater than the start time, then it is determined that there are overlapping points and the boundary information is recorded.

[0040] Ultimately, all boundaries are stored as a fragmented boundary set, which includes time location and boundary type.

[0041] S1.6: Map the abnormal fluctuation parameter vector and the fragmented structure boundary set to form a feature set that simultaneously includes both volatility and fragment attribution.

[0042] The mapping process includes binding the corresponding abnormal fluctuation parameter vector to the record of each time series segment and adding a fragmentation marker field. If the time series segment involves lost points or overlapping points, the fragmentation marker field records the corresponding boundary information.

[0043] Finally, the complete record of each time series is written into the feature set, and each record in the set contains both volatility features and fragmentation features.

[0044] S2: Based on the initial task data set, perform attribution analysis on fragmented data segments, splice together segments with the same source, and generate a spliced ​​task data set.

[0045] S2.1: Mark the source of the fragmented structure boundary set in the feature set, and compare adjacent segments based on time index and data content features to generate a candidate set of fragment attribution.

[0046] The specific operations include: Extract the fragmented structure boundary set from the feature set, and assign a source tag to each segment according to the time index of the original task data set, recording the source node, original task ID and start and end time of the segment; For each fragmented segment, content features are obtained, including numerical mean, variance, and higher-order difference features, to form a segment feature vector; During the candidate matching process, the time index interval and content feature similarity of adjacent segments are compared one by one. If the time interval is less than the set time threshold and the content feature distance is less than the feature distance threshold, the corresponding data segment pair is added to the fragment belonging candidate set.

[0047] It should be noted that existing technologies typically splice segments only in chronological order, ignoring differences in content characteristics. This leads to incorrect splicing of data from different sources or for different tasks, resulting in data conflicts or confusion. This invention, by simultaneously considering both time indexes and segment content characteristics, can accurately identify segments from the same source, improving splicing accuracy. At the same time, it retains the source information of each segment, ensuring traceability for subsequent processing and reducing the risk of data mismatch.

[0048] S2.2: Combining the abnormal fluctuation parameter vector, fragments with fluctuation differences exceeding the fluctuation threshold in the fragment attribution candidate set are removed to generate the attribution set.

[0049] The determination of fluctuation differences includes: extracting the abnormal fluctuation parameter vector of each fragment from the fragment attribution candidate set; for each pair of candidate fragments, calculating the degree of difference between the abnormal fluctuation parameter vectors, and defining the fluctuation difference as |mean difference| + |variance difference| + |skewness difference| + |kurtosis difference|.

[0050] S2.3: Rearrange the same-origin segments in the set of belonging to the set according to the time index order to generate a set of splicing order.

[0051] Specifically, for each group of related fragments in the set of origin, the fragments are sorted in ascending order according to their start time index to obtain the rearranged fragment sequence; the rearrangement order of each fragment is written into the sequence field to identify the order of the fragments in the splicing process; the splicing order set is output, where each record contains a fragment source tag, start and end time index, content feature vector and rearrangement order field.

[0052] S2.4: In the splicing order set, insert markers or approximate fill values ​​for adjacent segments with missing points to generate a splicing segment set with enhanced continuity. Integrate the splicing segment set according to the source markers to generate the spliced ​​task data set.

[0053] As can be seen, by inserting markers or approximate padding values ​​while retaining source markers, this invention can generate task data sets with enhanced continuity, ensuring the integrity of the data sequence and tracing the source of each data segment. This results in concatenated data that is both continuous and traceable, providing reliable input for subsequent computing center scheduling.

[0054] S3: Compare the spliced ​​task data set with other task data sets to be scheduled in the edge nodes, identify duplicate or similar data segments, and generate the target task data set after redundancy reduction.

[0055] S3.1: Perform index matching between the spliced ​​task data set and other task data sets to be scheduled within the edge node, and extract data segments with the same time index range.

[0056] Specifically, for the spliced ​​task data set and the task data set to be scheduled at the edge nodes, the start time index and end time index fields of each segment are extracted respectively; The time index ranges of the segments in the two sets are compared sequentially. If the time index ranges of the two segments overlap, that is, the difference between the start time and the difference between the end time are both less than the time threshold Δt_index, then the segments are identified as candidate duplicate segments and the data segments are extracted for subsequent feature comparison. Output a set of fragments with the same time index range. Each record contains the spliced ​​fragment ID, the fragment ID to be scheduled, the start and end fields of the time index, and the source tag.

[0057] It should be noted that existing technologies often perform full comparisons of the entire dataset or rely solely on content matching, without considering time index matching. This results in low comparison efficiency and may overlook highly overlapping time segments. This invention, through time index matching, can significantly reduce the number of segments requiring subsequent difference calculations, improving comparison efficiency, while ensuring that redundancy is eliminated only for time-overlapping data segments, thus improving accuracy.

[0058] S3.2: Perform difference calculation on the feature set of the corresponding data segment, and mark the data segments with difference values ​​lower than the preset difference threshold as similar segments, and output the set of similar segments.

[0059] A preferred method for calculating the difference includes: For each pair of task segments to be compared, feature vectors are extracted from their respective feature sets. Let be the feature vectors. and The vector has a length of N, and each element in the vector corresponds to a feature term, such as numerical difference features, abnormal fluctuation parameters, fragmented structure markers, etc. For each pair of eigenvalues ​​at corresponding positions in the vector, calculate the difference; that is, for the k-th eigenvalue element, calculate its absolute difference. .

[0060] The total sum of differences is obtained by summing the absolute differences of all N feature elements.

[0061] Divide the summation result by the vector dimension N to calculate the average difference value.

[0062] S3.3: Remove duplicate segments from the set of similar segments, retain a single copy with a source tag, and generate a deduplicated set.

[0063] Specifically, for a set of similar segments, a segment retention strategy is determined based on the priority of the source tag and time index. For example, segments with lower source node numbers or earlier splicing order are retained. Duplicate segments are removed while retaining the complete source tag field of the segment for subsequent scheduling. A deduplicated set is output, in which each record contains the deduplicated segment ID, source tag, time index, and joint feature vector.

[0064] It should be noted that if duplicate segments are frequently deleted or overwritten without retaining source information, the original source of the segments may be unidentifiable during task scheduling, affecting data traceability and distribution strategies. By removing duplicate segments while retaining source markers, this invention ensures redundancy reduction while preserving the traceability and original source information of each segment, improving the flexibility and security of task scheduling.

[0065] S3.4: Integrate the remaining unlabeled segments in the deduplication set and the concatenated task data set to generate the target task data set after redundancy reduction.

[0066] The fragments in the spliced ​​task data set that were not included in the similar fragment set are filtered to generate the remaining unlabeled fragment set; the deduplication set and the remaining unlabeled fragment set are merged in time index order to maintain the continuity of the data sequence and retain the source tag and joint feature vector field of each fragment.

[0067] The output is a set of target task data after redundancy reduction. Each record includes a fragment ID, time index, source marker, abnormal fluctuation parameter vector, and fragmentation structure marker field, providing unified, continuous, and redundancy-free input data for subsequent task scheduling in the computing center.

[0068] S4: Input the target task data set into the computing center for scheduling, sort the task distribution paths according to the abnormal fluctuation parameters and redundancy reduction results, and generate the final task scheduling scheme.

[0069] First, the target task data set and the abnormal fluctuation parameter vector are mapped to generate a task feature combination set containing fluctuation characteristics and redundancy reduction markers.

[0070] Specifically, task segment information is extracted one by one from the target task dataset, including segment ID, time index range, and source marker. For each task segment, the abnormal fluctuation amplitude, differential statistics, and fluctuation level of the corresponding segment are extracted from the abnormal fluctuation parameter vector set, and these features are mapped one-to-one with the segment ID, time index, and source marker. For each segment, the deduplication marker generated during the redundancy reduction process is queried, and the marker is appended to the corresponding feature information of the segment to form a complete task feature entry. The feature entries of all task segments are summarized to generate a task feature combination set, where each record includes segment ID, time index, source marker, abnormal fluctuation parameter vector, and redundancy reduction marker. The numerical features in the task feature combination set are normalized to unify the units for subsequent sorting and path weight calculation.

[0071] This invention maps abnormal fluctuation parameters and redundancy reduction markers to a unified set of task feature combinations, which can simultaneously consider task stability and redundancy reduction effects in scheduling calculations, making task distribution more reasonable and taking into account both data reliability and processing efficiency.

[0072] Secondly, based on the fluctuation amplitude level and the number of redundant markers in the task feature combination set, the candidate task distribution paths are prioritized to generate a path set with sorting weights.

[0073] The specific steps are as follows: For each candidate task distribution path, the fluctuation amplitude level and the number of redundant tags of the included task segments are counted. The characteristics of the task segments corresponding to the path are weighted and summarized to generate the path weight. Among them, the segment with the lower fluctuation amplitude and the fewer redundant tags has a higher contribution value. Based on the path weight, all candidate task distribution paths are sorted to generate a set of paths with sorting weights. Each path record includes path ID, path weight, task fragment ID and its source tag. The weight parameters and thresholds considered during the sorting process are recorded as scheduling strategy parameters for subsequent path selection and dynamic adjustment.

[0074] Finally, the path set is allocated according to priority order and the target task data set to generate the final task scheduling scheme and output it.

[0075] Specifically, the sorted set of paths and the target task data set are mapped to each other, and task segments are allocated according to the path priority from high to low. Each segment maintains the continuity of the time index and the integrity of the source tag. During the allocation process, if it is found that the same segment can be allocated to multiple paths, the best path is selected by comparing the path weights to ensure that highly volatile or redundant segments are preferentially allocated to paths with lower load or more stable conditions. The final task scheduling scheme is output. Each record contains the task fragment ID, distribution path ID, sorting weight, time index, abnormal fluctuation parameter vector, and redundancy reduction flag, providing complete input for the computing center to execute task scheduling.

[0076] As can be seen, by combining path priority, fluctuation parameters and redundancy markers in the final allocation, the present invention can generate a balanced, continuous and efficient task scheduling scheme, while maintaining task data integrity, source traceability and processing reliability.

[0077] Furthermore, such as Figure 2 As shown, this embodiment also provides a computing center edge computing task scheduling system, including, The preprocessing module is used by edge nodes to preprocess the received task data stream, extract abnormal fluctuation parameters and fragmentation structure parameters of the task data stream, and output an initial task data set with annotations. The splicing module is used to perform attribution analysis on fragmented data segments based on the initial task data set, splice segments with the same source, and generate a spliced ​​task data set. The redundancy reduction module is used to compare the spliced ​​task data set with other task data sets to be scheduled in the edge nodes, identify duplicate or similar data segments, and generate the target task data set after redundancy reduction. The scheduling module is used to input the target task data set into the computing center for scheduling, sort the task distribution paths according to the abnormal fluctuation parameters and redundancy reduction results, and generate a task scheduling scheme.

[0078] This embodiment also provides a computer device applicable to the edge computing task scheduling method of computing power centers, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the edge computing task scheduling method of computing power centers as proposed in the above embodiment.

[0079] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0080] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the edge computing task scheduling method for computing power centers as proposed in the above embodiments.

[0081] In summary, the beneficial effects of this invention are as follows: by performing data preprocessing, fragment attribution and splicing, and redundancy reduction operations at the edge nodes, this invention effectively reduces redundant calculations and repeated transmissions, improves the integrity and consistency of task data, and thus provides higher quality input for the unified scheduling of computing centers; on this basis, by combining deep learning methods to model and prioritize task features, it can dynamically adapt to the abnormal fluctuation characteristics of task flows and realize flexible adjustment of path distribution.

[0082] Overall, this invention not only improves the collaboration efficiency between the edge and the center, reduces the waste of system resources and scheduling latency, but also enhances robustness and adaptability in high-concurrency and complex task scenarios.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for scheduling edge computing tasks in a computing center, characterized in that: include: The edge node preprocesses the received task data stream, extracts the abnormal fluctuation parameters and fragmentation structure parameters of the task data stream, and outputs an initial task data set with annotations. Based on the initial task data set, the fragmented data segments are analyzed for their attribution, and the segments with the same origin are spliced ​​together to generate a spliced ​​task data set. The spliced ​​task data set is compared with other task data sets to be scheduled in the edge nodes to identify duplicate or similar data segments and generate the target task data set after redundancy reduction. The target task data set is input into the computing center for scheduling. The task distribution paths are sorted according to the abnormal fluctuation parameters and redundancy reduction results to generate a task scheduling scheme. The edge node preprocesses the received task data stream by: dividing the received task data stream into continuous time segments according to timestamps, generating a preprocessed data set with time indexes; detecting time segments whose fluctuation amplitude exceeds a set range based on the fluctuation amplitude of the time segments in the preprocessed data set, and buffering the time segments separately to generate a data set of marked abnormal segments; performing format difference detection between the data set of marked abnormal segments and the data set of other normal segments, and performing unified encoding on the edge node side to output a preprocessed data set with a consistent structure. Extracting the abnormal fluctuation parameters and fragmentation structure parameters of the task data stream includes: extracting the numerical differences of continuous time segments from the preprocessed data set, calculating the statistical distribution of the differences, and generating an abnormal fluctuation parameter vector; marking the lost points and overlapping points between adjacent data segments based on the time index of the preprocessed data set, and generating a fragmentation structure boundary set; mapping the abnormal fluctuation parameter vector and the fragmentation structure boundary set to form a feature set that simultaneously includes fluctuation and fragmentation attribution.

2. The edge computing task scheduling method for computing power centers as described in claim 1, characterized in that: The attribution analysis of fragmented data segments includes: The fragmented structure boundary set in the feature set is marked with its source, and adjacent fragments are compared based on time index and data content features to generate a candidate set of fragment attribution; By combining the abnormal fluctuation parameter vector, fragments in the fragment attribution candidate set that have fluctuation differences exceeding the fluctuation threshold are removed to generate an attribution set.

3. The edge computing task scheduling method for computing power centers as described in claim 2, characterized in that: The process of splicing together homogeneous fragments and generating a spliced ​​task data set includes: For the homologous segments in the aforementioned set of attribution, rearrange them according to the time index order to generate a set of splicing order; In the splicing order set, markers or approximate padding values ​​are inserted for adjacent segments with missing points to generate a splicing segment set with enhanced continuity; The spliced ​​fragment set is integrated according to the source tag to generate a spliced ​​task data set.

4. The edge computing task scheduling method for computing centers as described in claim 3, characterized in that: The identification of duplicate or similar data segments includes: The spliced ​​task data set and other task data sets to be scheduled within the edge node are indexed and matched to extract data segments with the same time index range. The feature set of the corresponding data segments is used to calculate the difference, and data segments with difference values ​​lower than the preset difference threshold are marked as similar segments, and a set of similar segments is output. The generated target task data set after redundancy reduction includes: Duplicate segments in the set of similar segments are removed, and a single copy is retained with an accompanying source tag to generate a deduplicated set. The remaining unlabeled segments in the deduplicated set and the spliced ​​task data set are integrated to generate the target task data set after redundancy reduction.

5. The edge computing task scheduling method for computing power centers as described in claim 4, characterized in that: The generation of the task scheduling scheme includes: The target task data set and the abnormal fluctuation parameter vector are mapped to generate a task feature combination set containing fluctuation characteristics and redundancy reduction markers; Based on the fluctuation amplitude level and the number of redundant markers in the task feature combination set, the candidate task distribution paths are prioritized and a path set with sorting weights is generated. The path set is allocated according to priority order and corresponding to the target task data set to generate the final task scheduling scheme and output it.

6. A computing center edge computing task scheduling system, based on the computing center edge computing task scheduling method according to any one of claims 1 to 5, characterized in that: Also includes: The preprocessing module is used by edge nodes to preprocess the received task data stream, extract abnormal fluctuation parameters and fragmentation structure parameters of the task data stream, and output an initial task data set with annotations. The splicing module is used to perform attribution analysis on fragmented data segments based on the initial task data set, splice segments with the same source, and generate a spliced ​​task data set. The redundancy reduction module is used to compare the spliced ​​task data set with other task data sets to be scheduled in the edge nodes, identify duplicate or similar data segments, and generate the target task data set after redundancy reduction. The scheduling module is used to input the target task data set into the computing center for scheduling, sort the task distribution paths according to the abnormal fluctuation parameters and redundancy reduction results, and generate a task scheduling scheme.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the edge computing task scheduling method of any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the edge computing task scheduling method of any one of claims 1 to 5.

Images