CN-122019306-A - Edge reasoning-based calculation power monitoring and regulating method, device and equipment

Abstract

The invention provides a computing power monitoring and regulating method, device and equipment based on edge reasoning, which are characterized in that a multisource computing power monitoring data set is completed by collecting real-time monitoring data of a starting and stopping transition state of a current computing power core board, steady state historical monitoring data and starting and stopping transition state monitoring data of an adjacent computing power core board, a processing precision standard of the completed transition state monitoring data is improved from an initial stage of the starting and stopping transition state of the current computing power core board to a steady state switching stage, computing power state analysis processing is respectively carried out on the completed transition state monitoring data of each stage to obtain computing power state analysis results of different stages of the transition state, and a computing power state analysis result corresponding to the current computing power core board steady state historical monitoring data in the multisource computing power monitoring data set is linked to obtain a full-period computing power state analysis result and perform computing power regulating operation. The invention can ensure the smooth state switching of the calculation force core board.

Inventors

• YANG YANHUI
• YUAN JUN
• FAN XIANBO
• JIANG MIN
• MIAO HAICHAO

Assignees

• 电子科技大学

Dates

Publication Date

20260512

Application Date

20260204

Claims (10)

1. 1. An edge reasoning-based computational power monitoring and regulating method is characterized by comprising the following steps: Collecting real-time monitoring data of the start-stop transition state of a current power computing core board, steady state historical monitoring data of the current power computing core board and the start-stop transition state monitoring data of an adjacent power computing core board to obtain a multi-source power computing monitoring data set; the method comprises the steps of completing a broken data segment in current power computing core board start-stop transition state real-time monitoring data in the multi-source power computing monitoring data set through current power computing core board steady state historical monitoring data and adjacent power computing core board start-stop transition state monitoring data in the multi-source power computing monitoring data set, and obtaining completed transition state monitoring data; Gradually improving the processing precision standard of the completed transition state monitoring data from the initial stage of the start-stop transition state of the current power calculation core board to the steady state switching stage, and respectively executing power calculation state analysis processing on the completed transition state monitoring data of each stage to obtain power calculation state analysis results of different stages of the transition state; Carrying out connection processing on the calculation force state analysis results of the transition states at different stages and calculation force state analysis results corresponding to the current calculation force core plate steady state historical monitoring data in the multi-source calculation force monitoring data set to obtain full-period calculation force state analysis results; and performing calculation force regulation and control operation based on the analysis result of the full-period calculation force state, and completing calculation force regulation and control connection between the start-stop transition state and the steady state of the current calculation force core plate.
2. 2. The edge reasoning-based power monitoring and controlling method according to claim 1, wherein the step of supplementing the broken data segment in the current power core board start-stop transition state real-time monitoring data in the multi-source power monitoring data set with the current power core board steady state history monitoring data and the adjacent power core board start-stop transition state monitoring data in the multi-source power monitoring data set to obtain the complemented transition state monitoring data comprises the following steps: Traversing the current power core board steady state historical monitoring data in the multi-source power computing monitoring data set, screening the time period content corresponding to the start-stop transition process, arranging according to time sequence, and marking the start-stop trigger nodes of each time period to obtain a current core board historical start-stop time period sequence; Traversing the adjacent power core board start-stop transition state monitoring data in the multi-source power calculation monitoring data set, screening the time period content corresponding to the start-stop transition process, normalizing the time period content to eliminate the influence of hardware and load difference, arranging according to time sequence, and marking the start-stop trigger node of each time period to obtain a normalized adjacent core board start-stop time period sequence; performing time sequence alignment processing on the current core board historical start-stop time period sequence and the adjacent core board start-stop time period sequence, enabling start-stop node time period marks of the two sequences to be kept synchronous, aligning the recording positions of state change nodes, and generating a start-stop time period comparison sequence with synchronous time sequence; Traversing current power core board start-stop transition state real-time monitoring data in the multi-source power monitoring data set, identifying blank periods of all content missing, recording start and end marks of each blank period, associating corresponding start-stop action types, and obtaining a broken data period mark set; Filling blank intervals covered by the broken data section time period mark sets time by time period based on time period contents matched with the broken data section time period mark sets in the time sequence synchronous start-stop time period comparison sequence, so that the filling contents are continuous with state changes of the front time period and the rear time period, and preliminary full transition state monitoring data are obtained; and traversing the preliminarily complemented transition state monitoring data, checking state change connection conditions among adjacent time period entries, and correcting content conflict parts at time period connection positions to ensure that the state change of the whole transition state monitoring data is kept continuous, so as to obtain the complemented transition state monitoring data.
3. 3. The method for computing power monitoring and controlling based on edge reasoning according to claim 2, wherein the step of performing time sequence alignment processing on the current core board historical start-stop time period sequence and the adjacent core board start-stop time period sequence to enable start-stop node time period marks of the two sequences to keep synchronous, aligning recording positions of state change nodes, and generating a time sequence synchronous start-stop time period comparison sequence comprises the steps of: Extracting start-stop trigger node marks in the current core plate historical start-stop time period sequence, recording time period positions and action types corresponding to each mark, and sorting according to the action types to obtain a current core plate start-stop trigger mark set; Extracting start-stop trigger node marks in the adjacent core plate start-stop time period sequence, recording time period positions and action types corresponding to each mark, and sorting according to the action types to obtain an adjacent core plate start-stop trigger mark set; Performing action type matching on the current core board start-stop trigger mark set and the normalized adjacent core board start-stop trigger mark set, calibrating matched time period positions based on performance baselines of the core boards, enabling trigger marks of the same type of start-stop actions to correspond on a normalized time axis, and recording the calibrated matched time period positions to obtain a trigger mark matched pair set; Comparing the time interval position of each calibrated matched pair in the trigger mark matched pair set, calculating the relative time sequence deviation on a normalized time axis, and adjusting the time interval mark of the preliminary alignment sequence based on the relative time sequence deviation to keep the relative time sequence relation of the matched pair consistent, so as to obtain an adjusted alignment sequence; traversing the adjusted alignment sequence, checking the continuity of state change between adjacent start-stop trigger marks, correcting the time period entry of state mutation, and keeping the state change of the whole sequence continuous to obtain a checked alignment sequence; And associating the verified alignment sequence with the start-stop action type of the broken data section time period mark set to enable the sequence content to be matched with the action requirement of the broken section, and obtaining a start-stop time period comparison sequence with synchronous time sequence.
4. 4. The edge reasoning-based computational power monitoring and control method as set forth in claim 3, wherein comparing the time period positions of each calibrated matched pair in the trigger mark matched pair, calculating the relative time sequence deviation thereof on a normalized time axis, adjusting the time period mark of the preliminary alignment sequence based on the relative time sequence deviation, keeping the relative time sequence relationship of the matched pair consistent, and obtaining the adjusted alignment sequence, comprising: Extracting the time period starting and ending marks of each matched pair in the trigger mark matched pair set, calculating the time difference of the corresponding marks of the current core plate and the adjacent core plate, and sorting according to the action type to obtain a time period deviation value set; Traversing adjacent time period entries of the current core plate historical start-stop time period sequence, calculating time period intervals between adjacent start-stop trigger marks, and sorting according to action types to obtain a time period interval reference value set; Based on the time interval deviation value set and the time interval reference value set, adjusting the time interval of adjacent trigger marks in the preliminary alignment sequence, so that the adjusted interval is consistent with the reference value set, and a preliminary adjusted alignment sequence is obtained; traversing the preliminarily adjusted alignment sequence, checking state change content corresponding to each start-stop trigger mark, correcting entries of which the state content is not matched with the time interval, and enabling the state change to be associated with the time interval to obtain an alignment sequence after content verification; Comparing the aligned sequence after the content verification with the time period content of the adjacent core board start-stop time period sequence, and adjusting the time period mark of the sequence to improve the content matching degree so as to obtain an aligned sequence with matched time period content; And traversing the alignment sequence matched with the time period content, checking the overall continuity of state change, and correcting conflict content at the joint of the time period to obtain an adjusted alignment sequence.
5. 5. The method for computing power monitoring and controlling based on edge reasoning as set forth in claim 1, wherein the step of gradually increasing the processing precision standard of the completed transition state monitoring data from an initial stage of a start-stop transition state of a current computing power core board to a steady state switching stage, and performing computing power state analysis processing on the completed transition state monitoring data of each stage to obtain computing power state analysis results of different stages of the transition state, includes: Dividing the time interval of the initial stage, the intermediate conversion stage and the steady state switching stage of the current computing core board start-stop transition state, marking the start and end time interval marks of each stage, and associating the corresponding start-stop action types to obtain a transition state stage time interval mark set; Extracting all contents corresponding to the transition state stage period mark set in the complemented transition state monitoring data, sorting and arranging according to stages, and recording the state change content of each stage to obtain a transition state monitoring data subsequence of each stage; analyzing the state change frequency of the transition state monitoring data subsequence of each stage, correspondingly setting sampling interval rules of each stage, so that the sampling interval of the initial stage corresponds to low-frequency change, the interval of the middle stage corresponds to gradual change frequency, and the interval of the switching stage corresponds to high-frequency change, and obtaining a stage sampling rule set; Extracting the content of the transition state monitoring data subsequences of each stage according to the stage sampling rule set, selecting corresponding time period items according to sampling intervals, and recording the state change content of each sampling item to obtain the monitoring data after sampling of each stage; traversing the monitoring data sampled at each stage, combing the time sequence relation of state change, recording the state type and the change direction of each sampling item, and sorting the state type and the change direction into a structured sequence according to the stages to obtain the analysis result of the calculation force state at each stage; And the initial stage calculation force state analysis result, the intermediate conversion stage calculation force state analysis result and the steady state switching stage calculation force state analysis result are connected in series in time sequence, the state change continuity at the stage connection position is checked, the connection conflict content is corrected, and the calculation force state analysis results of different stages of the transition state are obtained.
6. 6. The method for computing power monitoring and controlling based on edge reasoning according to claim 5, wherein analyzing the state change frequency of the sub-sequence of the transition state monitoring data of each stage, correspondingly setting sampling interval rules of each stage, enabling the sampling interval of the initial stage to correspond to low frequency change, the interval of the middle stage to correspond to gradual change frequency, and the interval of the switching stage to correspond to high frequency change, and obtaining the stage sampling rule set comprises: Traversing the initial stage transition state monitoring data subsequence, counting time intervals of adjacent state change nodes, calculating average length of the intervals, and analyzing low-frequency distribution characteristics of state change to obtain initial stage state frequency distribution description; setting a sampling rule with interval length equal to an average interval based on the initial stage state frequency distribution description, so that a sampling item covers a main state change node of an initial stage to obtain an initial stage sampling rule; traversing the intermediate transition state monitoring data subsequence, counting the time intervals of adjacent state change nodes, calculating gradual narrowing trend of the intervals, and analyzing gradual change frequency characteristics of state change to obtain intermediate stage state frequency distribution description; setting sampling rules with interval length gradually reduced along with frequency improvement based on the intermediate stage state frequency distribution description, so that sampling items cover detail state changes of the intermediate stage, and obtaining intermediate conversion stage sampling rules; Traversing the transition state monitoring data subsequence in the steady state switching stage, counting the time interval of adjacent state change nodes, calculating the minimum length of the interval, and analyzing the high-frequency distribution characteristics of the state change to obtain the state frequency distribution description in the switching stage; setting sampling rules with interval length equal to minimum interval based on the switching stage state frequency distribution description, enabling the sampling items to cover all state change nodes of the switching stage to obtain stable state switching stage sampling rules, and integrating the sampling rules of the three stages to obtain a stage sampling rule set.
7. 7. The method for computing power monitoring and controlling based on edge reasoning according to claim 6, wherein setting a sampling rule that a gap length gradually decreases with frequency elevation based on the intermediate stage state frequency distribution description, so that a sampling item covers a detail state change of an intermediate stage, and obtaining an intermediate conversion stage sampling rule includes: traversing the intermediate transition state monitoring data subsequence, marking the time interval positions of all state change nodes, arranging according to time sequence, and recording the state type corresponding to each node to obtain an intermediate state change node set; calculating the time intervals of adjacent nodes in the intermediate stage state change node set, arranging the nodes according to time sequence, and analyzing the gradual narrowing trend of interval values to obtain intermediate stage interval change trend description; Setting a sampling interval sequence from large to small based on the intermediate stage interval change trend description, so that each sampling interval corresponds to one interval in the interval change trend, and obtaining a preliminary intermediate sampling interval sequence; Associating the preliminary intermediate sampling interval sequence with the intermediate stage state change node set, checking node coverage conditions corresponding to sampling intervals, and adjusting interval length to enable sampling items to cover main detail change nodes to obtain an adjusted intermediate sampling interval sequence; Traversing the adjusted intermediate sampling interval sequence, associating corresponding time interval positions, and marking the starting time interval and the ending time interval of each sampling interval to obtain an intermediate sampling rule time interval mark set; And associating the intermediate sampling rule time period mark set with the intermediate conversion stage transition state monitoring data subsequence, verifying the matching degree of the sampling item and the state change node, and correcting the interval length with insufficient matching degree to obtain the intermediate conversion stage sampling rule.
8. 8. The method for monitoring and controlling the computing power based on the edge reasoning according to claim 1, wherein the step of linking the computing power state analysis results of the different stages of the transition state with the computing power state analysis results corresponding to the current computing power core board steady state history monitoring data in the multi-source computing power monitoring data set to obtain the full-period computing power state analysis results comprises the following steps: Traversing the calculation force state analysis results of different stages of the transition state, positioning the last period item of the steady state switching stage, recording the state content of the last period item and the node mark entering the steady state, and obtaining the calculation force state record at the tail end of the switching stage; traversing a calculation force state analysis result corresponding to the current calculation force core plate steady state history monitoring data in the multi-source calculation force monitoring data set, positioning a steady state forefront period item, recording the state content of the forefront period item and a node mark entering the steady state, and obtaining a steady state initial calculation force state record; Content comparison is carried out on the terminal calculation state record of the switching stage and the initial calculation state record of the stable state, the state content difference at the position of the mark of the node entering the stable state is analyzed, and the content and the appearance position of the state content difference are recorded to obtain the content difference description of the joint node; based on the content difference description of the connection nodes, correcting the state content of the terminal calculation force state record of the switching stage, so that the corrected content is consistent with the initial content of the initial calculation force state record of the stable state, and the regulated terminal calculation force state record of the switching stage is obtained; Correlating the tail end calculation force state record of the adjusted switching stage with the initial calculation force state record of the stable state to generate a calculation force state record covering the transition state to stable state connection period, and finishing the calculation force state record into a structured sequence to obtain a calculation force state analysis result of the transition state and the stable state connection period; The calculation force state analysis results of different stages of the transition state, the calculation force state analysis results of the transition state and steady state connection section and the calculation force state analysis results corresponding to the current calculation force core plate steady state history monitoring data in the multi-source calculation force monitoring data set are connected in series according to time sequence, the state change continuity of connection positions of all parts is checked, the connection conflict content is corrected, and the full-period calculation force state analysis results are obtained; Comparing the terminal calculation force state record of the switching stage with the initial calculation force state record of the steady state, analyzing the state content difference at the mark of the node entering the steady state, and recording the content and the appearance position of the state content difference to obtain the content difference description of the joint node, wherein the method comprises the following steps: extracting all state content items in the terminal calculation power state record of the switching stage, classifying and arranging according to the state type, and recording the time period position of each item to obtain a terminal switching content set; Extracting all state content items in the steady state initial computing power state record, classifying and arranging according to the state type, and recording the time period position of each item to obtain a steady state initial content set; Matching the switching terminal content set with the steady state initial content set according to the state type, setting matching tolerance based on a reasonable change range of state values, and enabling items of the same type of state content within the tolerance range to form a corresponding relation to obtain a content matching pair set; Traversing the content matching pair sets, analyzing the state content difference of each matching pair, recording the concrete performance of the difference, and sorting the difference content according to the state type to obtain a content difference description set; Traversing the content difference description set, recording the time period position corresponding to each difference content, and associating the node marks entering a stable state to obtain a difference position mark set; Integrating the content difference description set and the difference position mark set, and arranging the difference content and the corresponding position according to time sequence to obtain the content difference description of the connection node.
9. 9. A computing force monitoring and regulating device, comprising: the data acquisition module is used for acquiring real-time monitoring data of the start-stop transition state of the current power calculation core board, steady state history monitoring data of the current power calculation core board and start-stop transition state monitoring data of the adjacent power calculation core board to obtain a multi-source power calculation monitoring data set; The data complement module is used for complementing broken data segments in the current power computing core board start-stop transition state real-time monitoring data in the multi-source power computing monitoring data set through the current power computing core board steady state historical monitoring data in the multi-source power computing monitoring data set and the adjacent power computing core board start-stop transition state monitoring data to obtain complemented transition state monitoring data; The state analysis module is used for gradually improving the processing precision standard of the completed transition state monitoring data from the initial stage of the start-stop transition state of the current power calculation core board to the steady state switching stage, and respectively executing power calculation state analysis processing on the completed transition state monitoring data of each stage to obtain power calculation state analysis results of different stages of the transition state; the result connection module is used for connecting the calculation force state analysis results of the transition states at different stages with calculation force state analysis results corresponding to the current calculation force core plate steady state historical monitoring data in the multi-source calculation force monitoring data set to obtain full-period calculation force state analysis results; And the calculation force regulation and control module is used for carrying out calculation force regulation and control operation based on the analysis result of the full-period calculation force state and completing calculation force regulation and control connection between the start-stop transition state and the steady state of the current calculation force core plate.
10. 10. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that the processor implements the steps of the method of any of claims 1 to 9 when the program is executed.

Description

Edge reasoning-based calculation power monitoring and regulating method, device and equipment Technical Field The present invention relates to the field of data processing technologies, and in particular, to a method, an apparatus, and a device for computing power monitoring and controlling based on edge reasoning. Background The reinforced edge reasoning notebook uses a multi-computing force core board as a hardware carrier, and a core technology system is constructed around the dynamic regulation and control of computing force resource monitoring and reasoning tasks, so that the computing potential of the core board is maximally mined through accurate computing force sensing and task scheduling. The current mainstream scheme is mainly characterized in that operation data are independently collected for a single core board, calculation force resources are distributed according to preset rules, or task loads are adjusted by means of single core board historical steady state records, and regulation and control actions are completed through single core state feedback. In actual operation, the problems of unsmooth state engagement, mismatching of calculation force scheduling rhythm and the like, which are caused by the start-stop switching of a core board and the fluctuation of working conditions, are easy to cause insufficient suitability of calculation force distribution and reasoning task demands, the calculation force efficiency of multi-core cooperation cannot be fully released, and even the abnormal operation of the edge reasoning task can be possibly caused. Disclosure of Invention In view of the above, the invention provides a method, a device and equipment for monitoring and regulating computing power based on edge reasoning. The technical scheme of the embodiment of the invention is realized as follows: On one hand, the embodiment of the invention provides a calculation force monitoring and regulating method based on edge reasoning, which comprises the steps of collecting real-time monitoring data of a current calculation force core plate start-stop transition state, steady state historical monitoring data of the current calculation force core plate and start-stop transition state monitoring data of an adjacent calculation force core plate to obtain a multi-source calculation force monitoring data set, carrying out completion on broken data segments in the current calculation force core plate start-stop transition state real-time monitoring data in the multi-source calculation force monitoring data set through the current calculation force core plate steady state historical monitoring data and the adjacent calculation force core plate start-stop transition state monitoring data in the multi-source calculation force monitoring data set to obtain complete transition state monitoring data, gradually improving the processing precision standard of the complete transition state monitoring data from the initial stage of the current calculation force core plate start-stop transition state to the steady state switching stage, respectively executing calculation force state analysis processing on the complete transition state monitoring data of each stage to obtain a calculation force state analysis result of different stages, carrying out connection between the calculation force state analysis result of the different stages of the transition state and the calculation force core plate steady state analysis result of the multi-source calculation force core plate steady state operation on the basis of the calculation force monitoring and the calculation force core plate steady state operation result. In another aspect, an embodiment of the present invention provides a computing power monitoring and controlling device, including: the data acquisition module is used for acquiring real-time monitoring data of the start-stop transition state of the current power calculation core board, steady state history monitoring data of the current power calculation core board and start-stop transition state monitoring data of the adjacent power calculation core board to obtain a multi-source power calculation monitoring data set; The data complement module is used for complementing broken data segments in the current computing power core board start-stop transition state real-time monitoring data in the multi-source computing power monitoring data set through the current computing power core board steady state historical monitoring data in the multi-source computing power monitoring data set and the adjacent computing power core board start-stop transition state monitoring data to obtain complemented transition state monitoring data; The state analysis module is used for gradually improving the processing precision standard of the completed transition state monitoring data from the initial stage of the start-stop transition state of the current power calculation core board to the steady state switching stage, and respectively e