Intelligent Computing Resource Energy-saving Scheduling System Based on Load Prediction

Abstract

The present invention relates to the technical field of power management, specifically an intelligent computing resource energy-saving scheduling system based on load prediction. The system includes: a cache access monitoring module, a burst load determination module, a frequency fluctuation analysis module, a load classification and identification module, and an intelligent regulation strategy module. In the present invention, by monitoring the cache access data to analyze the fluctuation of access frequencies and identify changes in access patterns, the accuracy of load judgment is improved. Combining with the decrease in hit rate, the identification of burst loads is enhanced, making resource scheduling more accurate. By monitoring the CPU / GPU operating frequency and fluctuation rate, abnormal fluctuations are identified, the allocation of computing resources is optimized, and the energy consumption waste caused by frequency adjustment is reduced. Based on the time ratio and task characteristics, the load is classified, the resource matching degree is optimized, the CPU / GPU computing power and memory allocation are dynamically adjusted, the coordination of computing and data transmission is achieved, the energy consumption management is optimized, the stability of tasks is ensured, and the problems of resource idleness and imbalance are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power management, and particularly to an intelligent computing resource energy-saving scheduling system based on load prediction. Background Art

[0002] The technical field of power management includes the control and optimization of energy consumption in environments such as computer systems, server clusters, and data centers. Its core content is to monitor, analyze, and regulate the power consumption of computing resources to reduce energy consumption and improve energy utilization efficiency. This technical field involves dynamic power management, voltage regulation, frequency adjustment, and power distribution strategies to adapt to different workload requirements. The power management of computing devices usually relies on hardware-based energy consumption control mechanisms and software-based task scheduling strategies, such as dynamic voltage and frequency adjustment, load balancing, task migration, etc. In addition, in cloud computing and distributed computing environments, power management also includes the allocation optimization of virtual machine resources and the intelligent control of cooling systems to reduce unnecessary energy waste. With the popularization of artificial intelligence and large-scale computing, the application scope of power management technology has gradually expanded, not only focusing on the energy consumption of a single computing device but also involving the energy optimization of the entire computing infrastructure.

[0003] Among them, an intelligent computing resource energy-saving scheduling system refers to dynamically scheduling computing resources based on load prediction to achieve energy efficiency optimization of computing tasks. This system estimates future resource requirements using a load prediction model for the real-time load situation of computing tasks and performs task scheduling based on the power management mechanism of computing devices. Specifically, this system uses historical load data to train the prediction model, anticipates future computing requirements through methods such as regression analysis or time series modeling, and then allocates computing tasks based on the energy consumption characteristics of computing nodes. The scheduling methods include task reallocation based on load balancing, task migration based on energy efficiency ratio, and computing resource management based on dynamic frequency adjustment. This system also combines the temperature, power usage, and task execution characteristics of computing nodes and uses a multi-objective optimization method to schedule computing resources, enabling computing tasks to reduce power consumption while meeting performance requirements.

[0004] Existing technologies rely on historical statistical data, making it difficult to accurately predict real-time load changes, resulting in lagging resource allocation and untimely response to sudden loads. Load detection relies on static threshold judgment, with limited accuracy in identifying sudden loads, which may lead to mismatches in computing resource adjustment. CPU / GPU frequency adjustment is based on fixed power thresholds, lacking a comprehensive assessment of the dynamic characteristics of the load, which may affect computing performance or cause resource waste. Task scheduling adopts static load balancing, without fully considering the characteristics of computing tasks. Some nodes operate at high power consumption for a long time, and there is insufficient energy consumption optimization. The computing resource allocation method is fixed, making it difficult to adapt to complex load changes, affecting the overall energy consumption control efficiency. The energy consumption control ability of the cooling system is limited, further increasing energy waste. Summary of the Invention

[0005] The purpose of the present invention is to solve the drawbacks existing in the prior art, and a smart computing resource energy-saving scheduling system based on load prediction is proposed.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: The smart computing resource energy-saving scheduling system based on load prediction includes:

[0007] The cache access monitoring module obtains cache access data, monitors the cache access pattern, calculates the change in access frequency, evaluates the amplitude of access change, counts the hit rate and the number of migrations, and obtains cache access feature data;

[0008] The sudden load determination module calculates the fluctuation range of the access frequency based on the cache access feature data, judges the abnormal growth rate, compares the decline amplitude of the hit rate, and evaluates the continuity of data fluctuation to obtain the sudden load status result;

[0009] The frequency fluctuation analysis module obtains the CPU / GPU operating frequency, monitors the frequency change situation, calculates the fluctuation rate at adjacent moments, judges the abnormal fluctuation amplitude, and evaluates the change in the fluctuation rate to obtain the CPU / GPU frequency fluctuation status result;

[0010] The load classification and recognition module calculates the difference between the two pieces of data based on the sudden load status result and the CPU / GPU frequency fluctuation status result, compares the number of occurrences of data anomalies, classifies them into sudden, medium, or steady-state loads, calculates the proportion of multi-category time, and combines the computing task type to screen the load types to obtain the load classification result;

[0011] The intelligent regulation strategy module judges the load category based on the load classification result, compares resource allocation and data transmission, and adjusts the CPU / GPU resource allocation and memory allocation to obtain a load dynamic regulation plan.

[0012] As a further aspect of the present invention, the cache access feature data includes access frequency change, access change amplitude, hit rate, and migration times. The burst load status result includes access frequency fluctuation range, abnormal growth rate, decline amplitude of hit rate, and data fluctuation continuity. The CPU / GPU frequency fluctuation status result includes CPU / GPU operating frequency, frequency change situation, fluctuation rate, abnormal fluctuation amplitude, and fluctuation rate change. The load classification result includes load category, multi-category time proportion, and calculation task type to screen load types. The load dynamic regulation scheme includes CPU / GPU resource allocation adjustment, memory allocation adjustment, and comparison of resource allocation and data transmission.

[0013] As a further aspect of the present invention, the cache access monitoring module includes:

[0014] The cache access data acquisition sub-module obtains cache access records, collects access time, access address, and access interval, extracts the access frequency change trend, calculates the incremental interval of multiple accesses in the access time series, sets the access density index based on the time interval difference, classifies and summarizes the access density index, matches the corresponding access pattern according to the classification and summarization result, and obtains the cache access time series characteristics;

[0015] The access pattern calculation sub-module calculates the change value of multiple address access frequencies based on the cache access time series characteristics, counts the access increment within a short time, judges the access growth rate and its change trend, and uses the formula:

[0016] ;

[0017] Calculate and obtain the cache access frequency change coefficient, and combine the access pattern classification to calculate the cache access pattern characteristics;

[0018] Wherein, represents the cache access frequency change coefficient, which measures the degree of change of the cache access frequency over time, represents the address number of the th access, which is used to record the specific address information during cache access, represents the address number of the th access, which is used as a comparison benchmark for calculating the change between adjacent accesses, represents the time stamp of the th access, which records the specific time point when this access occurs, represents the time stamp of the th access, which is used as a comparison benchmark for calculating the time change, represents the total number of accesses, which indicates the number of accesses within the statistical time window;

[0019] The hit rate and migration statistics sub-module calculates the cache access hit rate based on the cache access pattern characteristics, counts the number of hits according to the number of accesses, obtains the hit ratio, counts the number of cache migrations, calculates the migration ratio, determines the cache load status based on the migration ratio, and obtains the cache access characteristic data.

[0020] As a further solution of the present invention, the burst load determination module includes:

[0021] The access frequency fluctuation calculation sub-module calculates the change range of the access frequency based on the cache access characteristic data, counts the change amplitude of the number of accesses per unit time, calculates the difference between the highest and lowest access numbers, summarizes the change amplitude interval, calculates the access fluctuation mean value, using the formula:

[0022] ;

[0023] Calculates the access frequency fluctuation range, obtains the access frequency fluctuation range through operation, and combines with the fluctuation trend to obtain the access frequency fluctuation characteristics;

[0024] Among them, represents the access frequency fluctuation range, which is used to measure the fluctuation of the access frequency per unit time, represents the number of the th access, indicating the access quantity counted at the th access, represents the number of the th access, serving as a comparison benchmark for calculating the change in adjacent access frequencies, represents the time interval between the th access and the th access, indicating the time length from the th access to the th access,

[0025] The hit rate decline evaluation sub-module calculates the change amplitude of the cache hit rate based on the access frequency fluctuation characteristics, calls the access records to count the number of hits, calculates the current hit rate and compares it with the historical data, counts the proportion of the hit rate decline, and determines the stability of the cache resources based on the hit rate decline trend to obtain the hit rate decline amplitude.

[0026] The data fluctuation continuity judgment sub-module calculates the fluctuation of the access data within a continuous period based on the hit rate decline amplitude, analyzes the change stability of the access frequency, calls the access data within multiple time periods, calculates the fluctuation duration, and determines the load status based on the duration of the burst access to obtain the burst load status result.

[0027] As a further solution of the present invention, the frequency fluctuation analysis module includes:

[0028] The operating frequency monitoring sub-module obtains the CPU / GPU operating frequency, continuously collects frequency data according to the time series, calls the system monitoring module to record the frequency values at each moment, statistically analyzes the frequency change trend between adjacent moments, calculates the difference between the maximum and minimum frequency values, and obtains the CPU / GPU operating frequency data;

[0029] The fluctuation rate calculation sub-module calculates the frequency change rate between adjacent moments based on the CPU / GPU operating frequency data, statistically analyzes the rate fluctuation amplitude at all moments, analyzes the fluctuation trend, and uses the formula:

[0030] ;

[0031] Performs operations to obtain the CPU / GPU frequency fluctuation rate, and combines the rate fluctuation trend to obtain the CPU / GPU fluctuation rate characteristics;

[0032] Wherein, represents the CPU / GPU frequency fluctuation rate, represents the CPU / GPU frequency values at time points, represents the cumulative rate value at time points,

[0033] the frequency fluctuation state evaluation sub-module analyzes the rate change trend based on the CPU / GPU fluctuation rate characteristics, calculates the outlier of the fluctuation amplitude, compares the rate fluctuation ranges in each time period, and combines the rate change interval to obtain the CPU / GPU frequency fluctuation state result.

[0034] As a further solution of the present invention, the load classification and recognition module includes:

[0035] The data difference calculation sub-module extracts the time series of the two items of data based on the burst load state result and the CPU / GPU frequency fluctuation state result, calculates the mean value, variance, and change rate respectively, obtains the absolute difference between the two and performs normalization processing, calculates the difference degree on its time series, and uses the formula:

[0036] ;

[0037] Combines its change trend for analysis, and performs operations to obtain the data difference coefficient;

[0038] Wherein, represents the data difference coefficient, which is used to measure the numerical difference degree between the burst load state and the CPU / GPU frequency change, represents the burst load state value at time indicating at time The burst load level detected by the system, representing the moment The CPU / GPU frequency value at, indicating the actual CPU or GPU frequency at which the computing device is operating at the moment The actual CPU or GPU frequency at which the computing device is operating, representing the mean value of the burst load state, indicating the average level of all burst load state values within the statistical time series range, representing the mean value of the CPU / GPU frequency, indicating the average level of all CPU / GPU frequency values within the statistical time series range, representing the total length of the time series, indicating the total number of time points included in the statistics during the calculation process;

[0039] The abnormal number statistics sub-module calls the data coefficient of variation, sets the burst load abnormal threshold, traverses based on the time series data, calculates the difference value for each data point and compares it with the abnormal threshold, records the number of abnormal points exceeding the threshold, calculates the frequency of abnormal occurrence and its proportion in the overall data, and obtains the load abnormal number ratio;

[0040] The load classification attribution sub-module sets the classification rules for burst load, medium load, and steady load based on the load abnormal number ratio and the data coefficient of variation, divides the time series data into intervals, calculates the time proportion of different categories, and filters the load types in combination with the calculation task type to obtain the load classification result.

[0041] As a further solution of the present invention, the intelligent regulation strategy module includes:

[0042] The load category determination sub-module, based on the load classification result, calls the CPU utilization rate, GPU utilization rate, memory occupancy rate, and data transmission rate, compares multiple parameters with the load type reference values, filters the corresponding threshold intervals, determines the load category, and obtains the load category determination result;

[0043] The resource allocation comparison sub-module, based on the load category determination result, calls the CPU / GPU resource allocation amount, memory occupancy amount, and data transmission rate, calculates the matching degree of multiple resources, and uses the formula:

[0044] ;

[0045] Performs an operation to obtain the resource matching degree value, compares it with the resource allocation reference, and obtains the resource allocation deviation value;

[0046] Wherein, represents the resource matching degree value, which is used to measure the matching degree between the actually allocated resource amount and the load demand resource amount, represents the The actual allocated resource amount of a resource, indicating the specific values such as CPU computing power, GPU computing power, memory occupancy, or data transfer rate allocated by the system on the resource, represents the load demand resource amount of the resource, indicating the specific values such as CPU computing power, GPU computing power, memory occupancy, or data transfer rate required by the system on the resource, represents the load benchmark resource amount of the resource, indicating the preset reference value or standard value on the resource category, represents the number of resource types, indicating the total number of resource types involved in the system, including different categories such as CPU computing power, GPU computing power, memory occupancy, data transfer rate, etc.;

[0047] Based on the resource allocation deviation value, the dynamic regulation scheme generation sub-module calls the current load thresholds of the CPU / GPU, the memory usage rate, and the data transfer rate adjustment range, compares the adjustable space, adjusts the resource allocation ratio, and obtains a load dynamic regulation scheme.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0049] In the present invention, by monitoring the cached access data, analyzing the fluctuation of access frequencies, and identifying the change of access patterns, the accuracy of load characteristic judgment is improved. The calculation of the access frequency fluctuation range is combined with the comparison of the decrease amplitude of the hit rate to enhance the ability to identify burst loads, making the resource scheduling more targeted. The monitoring of the CPU / GPU operating frequency is combined with the calculation of the fluctuation rate to identify abnormal fluctuations, optimize the allocation of computing resources, and reduce the energy consumption waste caused by frequency adjustment. The load classification is based on the multi-category time ratio and the characteristics of computing tasks to optimize the resource matching degree and improve the computing efficiency. Dynamically adjust the computing power of the CPU / GPU and the memory allocation, coordinate the computing resources and data transfer, optimize the energy consumption management, ensure the stable execution of tasks, and reduce the problems of resource idleness and imbalance. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is the system flow chart of the present invention;

[0051] Figure 2 is the flow chart of the cache access monitoring module of the present invention;

[0052] Figure 3 is the flow chart of the burst load determination module of the present invention;

[0053] Figure 4 is the flow chart of the frequency fluctuation analysis module of the present invention;

[0054] Figure 5 This is the flowchart of the load classification and recognition module of the present invention;

[0055] Figure 6 This is the flowchart of the intelligent regulation strategy module of the present invention. Specific embodiments

[0056] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0058] Embodiment 1:

[0059] Please refer to Figure 1 , the intelligent computing resource energy-saving scheduling system based on load prediction includes:

[0060] The cache access monitoring module obtains cache access data, monitors the cache access pattern, calculates the change in access frequency, evaluates the change amplitude of access, counts the hit rate and the number of migrations, and obtains cache access feature data;

[0061] The burst load determination module calculates the fluctuation range of access frequency based on the cache access feature data, determines that the growth rate is abnormal, compares the decline amplitude of the hit rate, evaluates the continuity of data fluctuation, and obtains the burst load status result;

[0062] The frequency fluctuation analysis module obtains the CPU / GPU operating frequency, monitors the frequency change situation, calculates the fluctuation rate at adjacent moments, determines that the fluctuation amplitude is abnormal, evaluates the change in the fluctuation rate, and obtains the CPU / GPU frequency fluctuation status result;

[0063] The load classification and recognition module calculates the difference between the two items of data based on the burst load status result and the CPU / GPU frequency fluctuation status result, compares the number of occurrences of data anomalies, classifies them into burst, medium or steady-state loads, calculates the proportion of multi-category time, and filters the load types in combination with the calculation task type to obtain the load classification result;

[0064] Based on the load classification results, the intelligent control strategy module determines the load category, compares resource allocation and data transmission, adjusts CPU / GPU resource allocation and memory allocation, and obtains a load dynamic control scheme.

[0065] The cache access feature data includes the change in access frequency, the amplitude of access change, the hit rate, and the number of migrations. The burst load status results include the range of access frequency fluctuations, abnormal growth rate, the decline amplitude of the hit rate, and the continuity of data fluctuations. The CPU / GPU frequency fluctuation status results include the CPU / GPU operating frequency, the frequency change situation, the fluctuation rate, the abnormal fluctuation amplitude, and the change in the fluctuation rate. The load classification results include the load category, the proportion of multi-category time, and the calculation task type to screen the load types. The load dynamic control scheme includes the adjustment of CPU / GPU resource allocation, the adjustment of memory allocation, and the comparison of resource allocation and data transmission.

[0066] Please refer to Figure 2 , the cache access monitoring module includes:

[0067] The cache access data acquisition sub-module obtains the cache access records, collects the access time, access address, and access interval, extracts the trend of the change in access frequency, calculates the incremental interval of multiple accesses in the access time series, sets the access density index based on the time interval difference, classifies and summarizes the access density index, and matches the corresponding access pattern according to the classification and summary results to obtain the cache access time series features;

[0068] The cache access data acquisition sub-module is used to obtain the cache access records and collect the access time, access address, and access interval. First, by parsing the cache log data, the time stamps of each access request, the corresponding cache address numbers, and the time intervals between two adjacent accesses are extracted and stored in a structured data table for subsequent processing, as shown in Table 1. Subsequently, the access time series is analyzed, and the fluctuation trend of the access frequency is obtained by calculating the increment of the change in the continuous access interval. In the analysis of time series data, a benchmark for the time interval difference is set, and the benchmark value can be determined by statistically calculating the mean and standard deviation of all access intervals. Assume the cache access data is as follows:

[0069] Table 1 Cache access time series data

[0070] ;

[0071] As shown in Table 1, for the benchmark value of the access interval, it can be set as the mean of all access intervals plus one standard deviation. For example: seconds. If the standard deviation is calculated to be seconds, then the benchmark value is set to Seconds. For the classification of access intensity metrics, it is set that if the access interval is less than 5 seconds, it is regarded as high-intensity access, 5 - 10 seconds as medium-intensity access, and more than 10 seconds as low-intensity access. Based on this classification, the access pattern classification can be obtained. For example, if the access interval of access A1 drops from 15 seconds to 2 seconds, it indicates a sudden increase in access frequency, and this pattern can be classified as a burst access pattern. While the access interval of access A2 is relatively stable, it is classified as a regular access pattern. Through the classification of these patterns, the cache access time series characteristics are obtained.

[0072] The access pattern calculation sub-module calculates the change value of multi-address access frequency based on the cache access time series characteristics, counts the access increment within a short period, judges the access growth rate and its change trend, using the formula:

[0073] ;

[0074] Calculate and obtain the cache access frequency change coefficient, and combine with the access pattern classification to calculate the cache access pattern characteristics;

[0075] Among them, represents the cache access frequency change coefficient, which measures the degree of change of cache access frequency over time, represents the address number of the th access, used to record the specific address information during cache access, represents the address number of the th access, serving as a comparison benchmark for calculating the change between adjacent accesses, represents the timestamp of the th access, recording the specific time point when this access occurs, represents the timestamp of the th access, serving as a comparison benchmark for calculating the time change, represents the total number of accesses, indicating the number of accesses within the statistical time window;

[0076] The cache access pattern calculation sub-module needs to calculate the change value of multi-address access frequency based on the cache access time series characteristics. First, for a certain cache system, record the access situations of all cache addresses within a specific time window. Each access operation is accompanied by the access address number and the timestamp . This module first needs to traverse all access records and sort the data in ascending order of the timestamp to construct the access sequence . Then, calculate the absolute difference between adjacent access address numbers and average the sum. In addition, it is necessary to calculate the time-weighted access rate change, that is, the access frequency of each time and the access frequency The square root of the sum of the squared differences is taken to obtain the rate change. To illustrate this process, we can assume a practical scenario. For example, in a certain server cache, the access address number may represent the data block ID. Suppose there are 6 accesses within 5 seconds, and the access address number sequence is and the access timestamp sequence is , then calculate: 1. Calculate the mean of the access address changes:

[0077] ;

[0078] Calculate the time-weighted access rate change:

[0079] ;

[0080] Calculate separately:

[0081] ;

[0082] ;

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] Finally, obtain :

[0088] ;

[0089] This value reflects the degree of change in the cache access frequency. If this value is much higher than the set benchmark value (e.g., 50), it indicates that the current cache access pattern has changed significantly, and further analysis of its access trend is required. See Table 1, which shows the calculation results of the access frequency change coefficient within different time windows.

[0090] The hit rate and migration statistics sub-module calculates the cache access hit rate based on the cache access pattern characteristics, counts the number of hits according to the number of accesses, obtains the hit ratio, counts the number of cache migrations, calculates the migration ratio, and determines the cache load status based on the migration ratio to obtain the cache access characteristic data.

[0091] The hit rate and migration statistics sub-module calculates the cache access hit rate based on the cache access pattern characteristics. First, count the total number of accesses and the number of hits , and calculate the hit rate: ;

[0092] In Table 1, assuming that all access data to cache A1 hits, then , the total number of accesses , then the hit rate is , then, count the number of cache migrations. If an address appears in different cache ranges, it is counted as a migration. For example, if A1 is moved from the main cache to the L2 cache, then the number of migrations increases, and calculate the migration ratio: ;

[0093] Assuming the number of migrations is 1, then the migration ratio , finally, judge the cache load status based on the migration ratio. If the migration ratio exceeds 50%, the cache load is high. If it is less than 20%, the cache load is low. In this example, the migration ratio is 20%, belonging to the low load status, and finally obtain the cache access characteristic data.

[0094] Please refer to Figure 3 , the burst load determination module includes:

[0095] The access frequency fluctuation calculation sub-module calculates the change range of the access frequency based on the cache access characteristic data, counts the change amplitude of the access times within a unit time, calculates the difference between the highest and lowest access times, summarizes the change amplitude range, and calculates the access fluctuation mean value, using the formula:

[0096] ;

[0097] Calculate the access frequency fluctuation range, operate to obtain the access frequency fluctuation range, and combine the fluctuation trend to obtain the access frequency fluctuation characteristics;

[0098] Among them, represents the access frequency fluctuation range, which is used to measure the fluctuation of the access frequency within a unit time, represents the number of the th access, indicating the access quantity counted at the th access, represents the number of the th access, which is used as a comparison benchmark for calculating the change of adjacent access frequencies, represents the time interval between the th access and the th access, indicating the time length from the th access to the th access,

[0099] The access frequency fluctuation calculation sub-module calculates the change range of the access frequency based on the cache access feature data. First, it extracts the access count sequence and the corresponding access time intervals in the cache access records, and arranges the access count data in chronological order to statistically analyze the change amplitude of the access count per unit time. For the setting of the unit time window, a fixed duration can be adopted, such as 10 seconds, 30 seconds, or 1 minute, etc., and the maximum and minimum values of each access count within the time window are statistically analyzed, and the difference is calculated to obtain the fluctuation amplitude of the access count. For example, assume that the access count data of a certain cache address is as follows:

[0100] Table 2 Access Count Fluctuation Statistical Table

[0101] ;

[0102] As shown in Table 2, within the time window of 0 - 40 seconds, the maximum access count is 8 and the minimum access count is 3, then the change range of the access count is , and further calculate the average access fluctuation, using the formula:

[0103] ;

[0104] where, represents the number of the rd access, represents the time interval of the th access, represents the total number of access counts statistically analyzed. Substitute the data for calculation:

[0105] ;

[0106] ;

[0107] ;

[0108] Calculate to obtain the access frequency fluctuation range . After obtaining the access frequency fluctuation range through calculation, compare this value with the fluctuation data within different time windows. If the fluctuation trend of shows an increasing or sudden increase, it indicates that the cache access has a significant fluctuation feature. If the fluctuation tends to be stable, the access change is stable. Finally, combine the fluctuation trend to obtain the access frequency fluctuation feature.

[0109] The hit rate decline evaluation sub-module calculates the change amplitude of the cache hit rate based on the access frequency fluctuation feature, calls the access records to count the number of hits, calculates the current hit rate and compares it with the historical data, statistically analyzes the proportion of the hit rate decline, and judges the stability of the cache resources based on the hit rate decline trend to obtain the hit rate decline amplitude;

[0110] The hit rate decline evaluation sub-module calculates the change range of the cache hit rate based on the access frequency fluctuation characteristics. First, it extracts the historical hit count data from the cache access records and calculates the hit rate within the current time window. The statistical formula is as follows ;

[0111] where is the number of hits, is the total number of accesses. For example, assuming the number of accesses within the current time window is , and the number of hits is , then the current hit rate is calculated as follows:

[0112] ;

[0113] Then, it compares with the historical data. Assuming the hit rate in the previous time window is 85%, the hit rate decline range is calculated as follows:

[0114] ;

[0115] Furthermore, it statistically calculates the hit rate decline ratio, which is calculated as follows:

[0116] ;

[0117] If the decline ratio exceeds 20%, it is determined that there is a problem with the stability of the cache resources. If the decline ratio is lower than 10%, it is determined that the cache resources are stable. In this example, the decline ratio is 17.65%, which is between 10% - 20%, and it is determined to be in a mild fluctuation state. Finally, the hit rate decline range is obtained.

[0118] The data fluctuation continuity judgment sub-module calculates the access data fluctuation situation within consecutive time periods based on the hit rate decline range, analyzes the change stability of the access frequency, calls the access data within multiple time periods, calculates the fluctuation duration, and determines the load state based on the duration of burst accesses to obtain the burst load state result.

[0119] The data fluctuation continuity judgment sub-module calculates the access data fluctuation situation within consecutive time periods based on the hit rate decline range. First, it calls the access data within multiple time periods, calculates the fluctuation duration, sets the time window, and analyzes the access frequency change trend within multiple time windows. For example, assuming the access frequencies within the past five time windows are [50, 40, 55, 35, 60], to calculate the fluctuation duration, first calculate the change range of the access frequency within each time window:

[0120] ;

[0121] The average value of the fluctuation change range is calculated as follows:

[0122] ;

[0123] If the fluctuation mean value is greater than the set threshold (e.g., 15), it is determined that the fluctuation is continuous. Next, the load status is determined based on the duration of the burst access. Set the duration threshold for the burst access. For example, if the burst access lasts for more than 3 time windows and the fluctuation mean value is greater than 15, it is determined as a high load status. In this example, the fluctuation amplitude exceeds 15 within three consecutive windows, so it is determined as a burst load status, and finally the burst load status result is obtained.

[0124] Please refer to Figure 4 , the frequency fluctuation analysis module includes:

[0125] The operating frequency monitoring sub-module obtains the CPU / GPU operating frequency, continuously collects frequency data according to the time series, calls the system monitor to record the frequency values at each moment, statistically analyzes the frequency change trend between adjacent moments, calculates the difference between the maximum and minimum frequency values, and obtains the CPU / GPU operating frequency data;

[0126] The operating frequency monitoring sub-module obtains the CPU / GPU operating frequency, continuously collects frequency data according to the time series. First, call the system monitoring module to collect the current CPU / GPU operating frequency at fixed time intervals (e.g., 100 ms), record the time stamp and the corresponding frequency value, and store them in the cache monitoring log. Subsequently, perform structured storage on the collected time series data. As shown in Table 3, arrange the frequency data in chronological order, calculate the frequency change trend between adjacent moments. The specific steps include obtaining the frequency values at two adjacent time points and calculating their change amount, setting a threshold to judge the change trend. For example, set the threshold to 50 MHz. If the frequency increase exceeds 50 MHz, it is determined as an upward trend. If the decrease exceeds 50 MHz, it is determined as a downward trend. If the change is within the threshold, it is determined as stable. As shown in Table 3, statistically analyze the difference between the maximum and minimum frequency values to obtain the CPU / GPU operating frequency data.

[0127] Table 3 CPU / GPU Operating Frequency Data Record Table

[0128] ;

[0129] As shown in Table 3, the maximum CPU frequency value is 3300 MHz, and the minimum CPU frequency value is 3100 MHz, then the frequency change range is MHz. Similarly, the GPU frequency change range is MHz. The CPU / GPU operating frequency data is obtained through calculation.

[0130] The fluctuation rate calculation sub-module calculates the frequency change rate at adjacent times based on the CPU / GPU operating frequency data, statistically analyzes the rate fluctuation amplitude at all times, and analyzes the fluctuation trend. The formula is as follows:

[0131] ;

[0132] Operate to obtain the CPU / GPU frequency fluctuation rate, and combine the rate fluctuation trend to obtain the CPU / GPU fluctuation rate characteristics;

[0133] Among them, represents the CPU / GPU frequency fluctuation rate, represents the CPU / GPU frequency values at represents the cumulative rate values at represents the number of time points for statistics;

[0134] The fluctuation rate calculation sub-module calculates the frequency change rate at adjacent times based on the CPU / GPU operating frequency data. First, extract the frequency data at adjacent times from Table 3, calculate the frequency change amount between two adjacent time points, and divide by the time interval to obtain the frequency change rate. For example, set the time interval ms, and calculate the frequency change rate of the CPU as follows:

[0135] ;

[0136] Similarly, calculate the frequency change rate of the GPU:

[0137] ;

[0138] ;

[0139] Then, use the formula:

[0140] ;

[0141] Among them, is the CPU / GPU frequency value at the th time point, is the cumulative rate value at the th time point, is the number of time points for statistics, and substitute the CPU data for calculation:

[0142] ;

[0143] ;

[0144] ;

[0145] ;

[0146] Calculate the fluctuation rate of the CPU MHz. The GPU calculation method is the same. Finally, combining the rate fluctuation trend, the CPU / GPU fluctuation rate characteristics are obtained.

[0147] Based on the CPU / GPU fluctuation rate characteristics, the frequency fluctuation state evaluation sub-module analyzes the rate change trend, calculates the outlier of the fluctuation amplitude, compares the rate fluctuation ranges in each time period, and combines the rate change interval to obtain the CPU / GPU frequency fluctuation state result.

[0148] Based on the CPU / GPU fluctuation rate characteristics, the frequency fluctuation state evaluation sub-module analyzes the rate change trend. First, obtain the CPU / GPU frequency fluctuation rates in different time periods, and calculate the difference between the maximum and minimum rates. Assume that the maximum rate of the GPU is 0.4 MHz / ms and the minimum rate is 0.1 MHz / ms. Then the fluctuation range is calculated as follows:

[0149] ;

[0150] Then, set the abnormal fluctuation threshold. For example, set the threshold to 0.35 MHz / ms. If the fluctuation range is greater than the threshold, it is determined as abnormal fluctuation, otherwise it is normal fluctuation. In this example, the GPU fluctuation range is 0.3 MHz / ms, which is less than the set threshold, so it is determined as the normal fluctuation state. Subsequently, compare the rate fluctuation ranges in each time period, calculate the fluctuation mean, and finally combine the rate change interval to obtain the CPU / GPU frequency fluctuation state result.

[0151] Please refer to Figure 5 , the load classification and recognition module includes:

[0152] Based on the burst load state result and the CPU / GPU frequency fluctuation state result, the data difference calculation sub-module extracts the time series of the two items of data, calculates the mean, variance and change rate respectively, obtains the absolute difference between the two and performs normalization processing, calculates the difference degree on its time series, and uses the formula:

[0153] ;

[0154] Analyze by combining its change trend, and calculate the data difference coefficient through operation;

[0155] Among them, represents the data difference coefficient, which is used to measure the numerical difference degree between the burst load state and the CPU / GPU frequency change, represents the moment of the burst load state value, indicating the burst load state value at the moment The sudden load level detected by the system, representing the moment of the CPU / GPU frequency value, indicating the actual CPU or GPU frequency at which the computing device is operating at the moment The actual CPU or GPU frequency at which the computing device is operating representing the mean value of the sudden load status, indicating the average level of all sudden load status values within the statistical time series range, representing the mean value of the CPU / GPU frequency, indicating the average level of all CPU / GPU frequency values within the statistical time series range, representing the total length of the time series, indicating the total number of time points included in the statistics during the calculation process;

[0156] The data difference calculation sub-module extracts the time series of the two items of data based on the sudden load status result and the CPU / GPU frequency fluctuation status result. First, obtain the time series data from the sudden load monitoring record , and at the same time, extract the same time series from the CPU / GPU frequency fluctuation monitoring record , align the two according to the time stamp to ensure data synchronization, and then calculate the mean values of the sudden load status and the CPU / GPU frequency. The mean value calculation formula is as follows:

[0157] ;

[0158] ;

[0159] Assume the time series length , the sudden load status data , the CPU / GPU frequency data , then the mean value calculation is as follows:

[0160] ;

[0161] ;

[0162] Then, calculate the variances of the sudden load status and the CPU / GPU frequency. The variance calculation formula is as follows:

[0163] ;

[0164] ;

[0165] Substitute the data for calculation:

[0166] ;

[0167] ;

[0168] ;

[0169] ;

[0170] Then, calculate the normalization of the absolute difference between the burst load status and the CPU / GPU frequency, using the formula:

[0171] ;

[0172] Substitute the data for calculation:

[0173] ;

[0174] ;

[0175] ;

[0176] Finally, calculate the data difference coefficient , and analyze it in combination with the change trend to obtain the data difference coefficient through operation.

[0177] The abnormal number statistics sub-module calls the data difference coefficient, sets the abnormal threshold for burst load, traverses based on the time series data, calculates the difference value for each data point and compares it with the abnormal threshold, records the number of abnormal points exceeding the threshold, calculates the frequency of abnormal occurrence and its proportion in the overall data, and obtains the load abnormal number ratio;

[0178] The abnormal number statistics sub-module calls the data difference coefficient and sets the abnormal threshold for burst load. First, determine the abnormal threshold of the data difference coefficient. For example, set the threshold , then, traverse based on the time series data, calculate the data difference value for each time point and compare it with the abnormal threshold, record the number of abnormal points exceeding the threshold. Assuming the calculated data difference coefficient of the time series data is , the number of abnormal points exceeding the threshold obtained through traversal calculation is as follows:

[0179] ;

[0180] Then, calculate the frequency of abnormal occurrence and its proportion in the overall data:

[0181] ;

[0182] Finally, the load abnormal number ratio is obtained as 60%.

[0183] The load classification and attribution sub-module sets the classification rules for burst load, medium load, and steady-state load based on the load exception frequency ratio and data difference coefficient, divides the time series data into intervals, calculates the time proportion of different categories, and filters the load types in combination with the calculation task type to obtain the load classification result.

[0184] The load classification and attribution sub-module sets the classification rules for burst load, medium load, and steady-state load based on the load exception frequency ratio and data difference coefficient. First, the classification criteria are defined as shown in Table 4.

[0185] Table 4 Load Classification Rules

[0186] ;

[0187] Then, according to the calculated exception ratio of 60% and data difference coefficient of 3.15, and in accordance with the classification criteria in Table 4, it is determined that this data sequence belongs to burst load. Next, the proportion of different load types in the time series is calculated. Assuming the classification results obtained in different time periods are [burst load, medium load, medium load, burst load, burst load], the proportion of burst load is statistically calculated:

[0188] ;

[0189] Similarly, the proportion of medium load is calculated:

[0190] ;

[0191] Finally, the load types are filtered in combination with the calculation task type to obtain the load classification result, that is, the proportion of burst load is the highest, and it is determined that the system is in the burst load state.

[0192] Please refer to Figure 6 , the intelligent regulation strategy module includes:

[0193] The load category determination sub-module, based on the load classification result, calls the CPU utilization rate, GPU utilization rate, memory occupancy rate, and data transfer rate, compares multiple parameters with the load type benchmark values, filters the corresponding threshold intervals, and determines the load category to obtain the load category determination result;

[0194] The load category determination sub-module, based on the load classification result, calls the CPU utilization rate, GPU utilization rate, memory occupancy rate, and data transfer rate. First, the CPU utilization rate, GPU utilization rate, memory occupancy rate, and data transfer rate are obtained from the system resource monitoring data and matched with the load classification result. For example, the benchmark values of different load types are set as shown in Table 5.

[0195] Table 5 Load Type Benchmark Values

[0196] ;

[0197] Extract the current load parameters from the system monitoring data. For example, the CPU utilization rate is 65%, the GPU utilization rate is 50%, the memory occupancy rate is 75%, and the data transfer rate is 180 MB / s. Match the data range in Table 5 and find that all parameters fall within the medium load range. Therefore, it is determined that the current load category is medium load, and finally the load category determination result is obtained.

[0198] Based on the load category determination result, the resource allocation comparison sub-module calls the CPU / GPU resource allocation amount, memory occupancy, and data transfer rate to calculate the matching degree of multiple resources, using the formula:

[0199] ;

[0200] Perform operations to obtain the resource matching degree value, and compare it with the resource allocation benchmark to obtain the resource allocation deviation value;

[0201] Among them, represents the resource matching degree value, which is used to measure the matching degree between the actual allocated resource amount and the load demand resource amount. represents the actual allocated resource amount of the th resource, indicating the specific values such as CPU computing power, GPU computing power, memory occupancy, or data transfer rate allocated by the system on the th resource. represents the load demand resource amount of the th resource, indicating the specific values such as CPU computing power, GPU computing power, memory occupancy, or data transfer rate required by the system on the th resource. represents the load benchmark resource amount of the th resource, indicating the preset benchmark reference value or standard value on the th resource category. represents the number of resource types, indicating the total number of resource types involved in the system, including different categories such as CPU computing power, GPU computing power, memory occupancy, and data transfer rate;

[0202] Based on the load category determination result, the resource allocation comparison sub-module calls the CPU / GPU resource allocation amount, memory occupancy, and data transfer rate. First, extract the current resource allocation situation of the system, including the CPU computing resource allocation amount , the GPU computing resource allocation amount , the memory occupancy , and the data transfer rate . At the same time, determine the resource amount corresponding to the load demand , and the load benchmark resource amount , and then use the formula:

[0203] ;

[0204] Assume that the current CPU resource allocation is 70%, the GPU resource allocation is 55%, the memory occupancy is 75%, and the data transfer rate is 190 MB / s, while the corresponding load requirements are 65%, 50%, 70%, and 180 MB / s respectively, and the load benchmark resource amounts are 80%, 60%, 85%, and 220 MB / s respectively. Calculate the resource matching degree:

[0205] ;

[0206] ;

[0207] ;

[0208] ;

[0209] ;

[0210] Finally, the calculated resource matching degree , and then compare it with the resource allocation benchmark. Set the matching degree threshold to 5.0. If exceeds the threshold, it is determined that there is a deviation in resource allocation. In this example , which is greater than 5.0, it is determined that the current resource allocation deviation is relatively large, and finally the resource allocation deviation value is obtained.

[0211] Based on the resource allocation deviation value, the dynamic regulation scheme generation sub-module calls the current load thresholds of the CPU / GPU, the memory usage rate, and the data transfer rate adjustment range, compares the adjustable space, and adjusts the resource allocation ratio to obtain a dynamic load regulation scheme.

[0212] Based on the resource allocation deviation value, the dynamic regulation scheme generation sub-module calls the current load thresholds of the CPU / GPU, the memory usage rate, and the data transfer rate adjustment range. First, set the adjustable resource range. For example, the adjustable range of the CPU is ±10%, the adjustable range of the GPU is ±15%, the adjustable range of the memory occupancy rate is ±10%, and the adjustable range of the data transfer rate is ±20%. Then compare the current resource allocation deviation value exceeds the threshold of 5.0, so adjustment is required. Assume that the CPU resource usage rate is 5% higher, then adjust the CPU resource allocation to 65%, the GPU resource allocation to 50%, the memory occupancy to 70%, and the data transfer rate to 180 MB / s. The final adjusted resource allocation is as follows:

[0213] Table 6 Adjusted Resource Allocation Scheme

[0214] ;

[0215] According to Table 6, check whether the adjusted parameters meet the load requirements by calculating the adjusted resource matching degree :

[0216] ;

[0217] ;

[0218] The adjusted matching degree drops to , indicating that the resource regulation reaches the optimal state, and finally a load dynamic regulation scheme is obtained.

[0219] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent computing resource energy-saving scheduling system based on load prediction, characterized in that The system includes: The cache access monitoring module obtains cache access data, monitors the cache access pattern, calculates the change in access frequency, evaluates the amplitude of access change, counts the hit rate and the number of migrations, and obtains cache access characteristic data; The burst load determination module calculates the fluctuation range of access frequency, determines the abnormal growth rate, compares the decrease amplitude of the hit rate, and evaluates the continuity of data fluctuation based on the cache access characteristic data, and obtains the burst load status result; The frequency fluctuation analysis module obtains the CPU / GPU operating frequency, monitors the frequency change situation, calculates the fluctuation rate at adjacent moments, determines the abnormal fluctuation amplitude, and evaluates the change in the fluctuation rate, and obtains the CPU / GPU frequency fluctuation status result; The load classification and identification module calculates the difference between the two data, compares the number of occurrences of data anomalies, classifies them as burst, medium, or steady-state loads, calculates the proportion of different load types in the time series, and filters the load types in combination with the calculation task type based on the burst load status result and the CPU / GPU frequency fluctuation status result, and obtains the load classification result; The intelligent regulation strategy module determines the load category based on the load classification result, compares the resource allocation and data transmission, and adjusts the CPU / GPU resource allocation and memory allocation to obtain a load dynamic regulation plan.

2. The intelligent computing resource energy-saving scheduling system based on load prediction according to claim 1, wherein The cache access characteristic data includes the change in access frequency, the amplitude of access change, the hit rate, and the number of migrations. The burst load status result includes the fluctuation range of access frequency, the abnormal growth rate, the decrease amplitude of the hit rate, and the continuity of data fluctuation. The CPU / GPU frequency fluctuation status result includes the CPU / GPU operating frequency, the frequency change situation, the fluctuation rate, the abnormal fluctuation amplitude, and the change in the fluctuation rate. The load classification result includes the load category, the proportion of different load types in the time series, and the filtering of load types by the calculation task type. The load dynamic regulation plan includes the adjustment of CPU / GPU resource allocation, the adjustment of memory allocation, and the comparison of resource allocation and data transmission.

3. The intelligent computing resource energy-saving scheduling system based on load prediction according to claim 2, characterized in that The cache access monitoring module includes: The cache access data acquisition sub-module obtains cache access records, collects the access time, access address, and access interval, extracts the change trend of access frequency, calculates the incremental interval of multiple accesses in the access time series, sets the access density index based on the time interval difference, classifies and summarizes the access density index, matches the corresponding access pattern according to the classification and summary result, and obtains the cache access time series characteristics; The access pattern calculation sub-module calculates the change value of the multi-address access frequency based on the cache access time series characteristics, counts the access increment within a short time, determines the access growth rate and its change trend, and uses the formula: ； Calculate and obtain the cache access frequency change coefficient, and calculate the cache access pattern characteristics in combination with the access pattern classification; Among them, represents the cache access frequency change coefficient, which measures the degree of change of the cache access frequency over time. represents the address number of the th access, which is used to record the specific address information during cache access. represents the address number of the th access, serving as a comparison benchmark for calculating the change in adjacent accesses. represents the timestamp of the th access, recording the specific time point when this access occurs. represents the timestamp of the th access, serving as a comparison benchmark for calculating the change in time. represents the total number of accesses, indicating the number of accesses within the statistical time window; The hit rate and migration statistics sub-module calculates the cache access hit rate based on the cache access pattern characteristics, counts the number of hits according to the number of accesses, obtains the hit ratio, counts the number of cache migrations, calculates the migration ratio, determines the cache load status based on the migration ratio, and obtains the cache access characteristic data.

4. The intelligent computing resource energy-saving scheduling system based on load prediction according to claim 3, wherein The burst load determination module includes: The access frequency fluctuation calculation sub-module calculates the change range of the access frequency based on the cache access feature data, counts the change amplitude of the access times within a unit time, calculates the difference between the highest and lowest access times, summarizes the change amplitude interval, calculates the access fluctuation mean value, using the formula: ； Calculate the access frequency fluctuation range, operate to obtain the access frequency fluctuation range, and combine the fluctuation trend to obtain the access frequency fluctuation characteristics; Among them, represents the access frequency fluctuation range, which is used to measure the fluctuation of the access frequency within a unit time, represents the number of the th access, indicating the access quantity counted at the th access, represents the number of the th access, serving as a comparison benchmark for calculating the change in adjacent access frequencies, represents the time interval between the th access and the th access, indicating the time length between the th access and the represents the total number of accesses counted, indicating the total number of accesses included in the calculation process; The hit rate decrease evaluation sub-module calculates the change amplitude of the cache hit rate based on the access frequency fluctuation characteristics, calls the access record to count the hit times, calculates the current hit rate and compares it with the historical data, counts the proportion of the hit rate decrease, and judges the stability of the cache resources according to the hit rate decrease trend to obtain the hit rate decrease amplitude; The data fluctuation continuity judgment sub-module calculates the access data fluctuation situation within a continuous time period based on the hit rate decrease amplitude, analyzes the change stability of the access frequency, calls the access data within multiple time periods, calculates the fluctuation duration, and determines the load status according to the duration of the burst access to obtain the burst load status result.

5. The intelligent computing resource energy-saving scheduling system based on load prediction according to claim 4, characterized in that, The frequency fluctuation analysis module includes: The operating frequency monitoring sub-module obtains the CPU / GPU operating frequency, continuously collects frequency data according to the time series, calls the system monitoring module to record the frequency values at each moment, counts the frequency change trend between adjacent moments, calculates the difference between the maximum and minimum frequency values, and obtains the CPU / GPU operating frequency data; The fluctuation rate calculation sub-module calculates the frequency change rate between adjacent moments based on the CPU / GPU operating frequency data, counts the rate fluctuation amplitude at all moments, analyzes the fluctuation trend, using the formula: ； Operate to obtain the CPU / GPU frequency fluctuation rate, and combine the rate fluctuation trend to obtain the CPU / GPU fluctuation rate characteristics; Among them, represents the CPU / GPU frequency fluctuation rate, represents the CPU / GPU frequency values at represent the cumulative rate values at represents the number of time points for statistics; The frequency fluctuation state evaluation sub-module analyzes the rate change trend based on the CPU / GPU fluctuation rate characteristics, calculates the abnormal value of the fluctuation amplitude, compares the rate fluctuation range of each time period, and combines the rate change interval to obtain the CPU / GPU frequency fluctuation state result.

6. The intelligent computing resource energy-saving scheduling system based on load prediction according to claim 5, wherein, The load classification and recognition module includes: The data difference calculation sub-module extracts the time series of the two items of data based on the burst load status result and the CPU / GPU frequency fluctuation state result, calculates the mean value, variance and change rate respectively, obtains the absolute difference between the two and performs normalization processing, and calculates the difference degree on its time series, using the formula: ； Analyze in combination with its change trend, and operate to obtain the data difference coefficient; Among them, represents the data difference coefficient, which is used to measure the numerical difference degree between the burst load state and the CPU / GPU frequency change, represents the moment of the burst load state value, indicating the burst load level detected by the system at the moment ; represents the moment of the CPU / GPU frequency value, indicating the actual CPU or GPU frequency at which the computing device is running at the moment ; represents the mean value of the burst load state, indicating the average level of all burst load state values within the statistical time series range, represents the mean value of the CPU / GPU frequency, indicating the average level of all CPU / GPU frequency values within the statistical time series range, represents the total length of the time series, indicating the total number of time points included in the statistics during the calculation process; The abnormal number statistics sub-module calls the data difference coefficient, sets the burst load abnormal threshold, traverses based on the time series data, calculates the difference value for each data point and compares it with the abnormal threshold, records the number of abnormal points exceeding the threshold, calculates the frequency of abnormal occurrence and its proportion in the overall data to obtain the load abnormal number ratio; Based on the load exception frequency ratio and the data difference coefficient, the load classification attribution sub-module sets the classification rules for burst load, medium load, and steady-state load, divides the time series data into intervals, calculates the proportion of different load types in the time series, and combines the calculation task type to screen the load types to obtain the load classification result.

7. The intelligent computing resource energy-saving scheduling system based on load prediction according to claim 6, wherein The intelligent regulation strategy module includes: Based on the load classification result, the load category determination sub-module calls the CPU utilization rate, GPU utilization rate, memory occupancy rate, and data transmission rate, compares multiple parameters with the load type reference values, screens the corresponding threshold intervals, determines the load category, and obtains the load category determination result; Based on the load category determination result, the resource allocation comparison sub-module calls the CPU / GPU resource allocation amount, memory occupancy, and data transmission rate, calculates the matching degree of multiple resources, and uses the formula: ； Performs operations to obtain the resource matching degree value, and compares it with the resource allocation reference to obtain the resource allocation deviation value; Among them, represents the resource matching degree value, which is used to measure the matching degree between the actually allocated resource amount and the resource amount demanded by the load, represents the actually allocated resource amount of the th type of resource, indicating the specific value of the CPU computing power, GPU computing power, memory occupancy or data transmission rate allocated by the system on the th type of resource, represents the resource amount demanded by the load of the th type of resource, indicating the specific value of the CPU computing power, GPU computing power, memory occupancy or data transmission rate required by the system on the th type of resource, represents the benchmark resource amount of the th type of resource, indicating the preset reference value or standard value on the th type of resource category, represents the number of resource types, indicating the total number of resource types involved in the system, including different categories such as CPU computing power, GPU computing power, memory occupancy, and data transmission rate; Based on the resource allocation deviation value, the dynamic regulation scheme generation sub-module calls the current load threshold of CPU / GPU, the memory usage rate, and the data transmission rate adjustment range, compares the adjustable space, adjusts the resource allocation ratio, and obtains the load dynamic regulation scheme.

Images