A network matrix system based on deep learning

Abstract

The present invention relates to the field of distributed computing technology, and specifically to a network matrix system based on deep learning, wherein the system includes a task allocation module, a task prediction module, a carrying capacity analysis module, and a computing unit optimization module. The present invention can maximize the performance of each computing unit and avoid overload and waste of resources by real-time monitoring of the carrying capacity of each computing unit and dynamically allocating matrix computing tasks. By analyzing historical task data and predicting future task volumes, data support is provided for the reasonable allocation of computing resources. By analyzing the matching of current and predicted task volumes with the carrying capacity of computing units, potential resource shortages can be discovered in advance and adjusted in time. By increasing the power consumption of computing units to handle additional task volumes, continuous computing efficiency and system stability are guaranteed, processing speed and accuracy are improved, and the adaptability of the system is enhanced, so that the ever-changing computing needs can be effectively responded to.

Description

Technical Field

[0001] The present invention relates to the field of distributed computing technology, and in particular to a network matrix system based on deep learning. Background Art

[0002] The field of distributed computing technology involves multiple computer systems or processors connected through a network to work together to complete tasks. The main advantages of distributed computing include increased processing power, resource sharing, improved reliability and fault tolerance. In the field of distributed computing, tasks are broken down into multiple small parts and processed in parallel by different machines in the network, thereby improving overall processing speed and efficiency. The technology is widely used in big data processing, cloud computing services, real-time data analysis, etc.

[0003] Among them, the network matrix system aims to use a distributed computing framework to achieve matrix operations and data processing. The network matrix system completes matrix operation tasks through multiple computers connected by a network, which is particularly suitable for large-scale data sets and complex operations, such as image processing, machine learning model training, etc. Through the network matrix system, large-scale matrix calculations can be effectively performed, data processing speed and accuracy can be optimized, and the system's adaptability to data processing of different scales can be improved.

[0004] Traditional network matrix systems have certain limitations in task allocation and energy management. Common problems include uneven utilization of computing resources, which causes some computing units to be overloaded while others are idle. This imbalance in resource allocation reduces the overall efficiency and responsiveness of the system. In addition, traditional systems lack the ability to predict future computing needs and find it difficult to dynamically adjust resources based on fluctuations in task volume. For example, if the peak task volume is not predicted, there will be a shortage of resources, which will affect the timely completion of tasks. These limitations not only prolong data processing time, but also increase energy consumption and operating costs. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and propose a network matrix system based on deep learning.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a network matrix system based on deep learning, the system comprising:

[0007] The task allocation module calculates the task carrying capacity ratio of each computing unit based on the matrix calculation tasks to be processed and the idle task carrying capacity of each computing unit. It also allocates the matrix calculation tasks to each computing unit through the network according to the matrix calculation task volume and the number of computing units, and obtains the real-time task allocation result.

[0008] The task volume prediction module uses deep learning algorithms to analyze the changing trend of matrix calculation task volume based on the processing records of matrix calculation tasks, predict the matrix calculation task volume in the future target period, and obtain the calculation task volume prediction results;

[0009] The carrying capacity analysis module detects the matrix calculation task processing rate of each computing unit under standard power consumption based on the real-time task volume allocation result and the task volume prediction result, predicts the total processing volume of the matrix calculation tasks in the target period, and determines whether the idle amount of the task carrying capacity in the target period can accommodate the predicted matrix calculation task volume, thereby obtaining the task carrying analysis result;

[0010] Based on the task load analysis results, if the predicted computing task volume cannot be accommodated, the computing unit optimization module will adjust the power consumption of the computing unit in real time according to the excess task volume, optimize the task processing rate of the computing unit, and obtain the computing unit power consumption adjustment result.

[0011] The present invention is improved in that the method for calculating the task carrying capacity ratio of each computing unit is:

[0012] Based on the distributed computing units connected to the network, the idle task carrying capacity of each computing unit is extracted to obtain the task carrying capacity information;

[0013] Based on the task carrying capacity information, the formula is:

[0014]

[0015] Calculate the task carrying capacity ratio of each computing unit , get the task carrying capacity ratio information, where Indicates The task carrying capacity ratio of each computing unit, It is The idle task carrying capacity of computing units, is the total number of computational units, No. The idle task carrying capacity of each computing unit.

[0016] The present invention is improved in that the step of obtaining the real-time task allocation result is:

[0017] Based on the matrix calculation tasks to be processed and the task carrying capacity ratio information, the formula is:

[0018]

[0019] Calculate the amount of matrix calculation tasks assigned to each computing unit ,in, is the total matrix calculation task, For the The task carrying capacity ratio of each computing unit, For the The amount of matrix calculation tasks assigned to the computing units, is the dynamic adjustment coefficient, is the amount of redundant tasks, is the load balancing factor;

[0020] Based on the matrix calculation task amount assigned to each computing unit , the matrix calculation tasks are allocated to each computing unit through the network to obtain the real-time task allocation results.

[0021] The present invention is improved in that the method for analyzing the changing trend of the matrix calculation task volume is:

[0022] Based on the processing records of matrix calculation tasks, daily matrix calculation task volume data within a period of time is extracted to obtain task volume data information;

[0023] Based on the task volume data information, through the formula:

[0024]

[0025] Calculate the rate of change of task volume ,in, For time point The amount of matrix calculation tasks, For time point The amount of matrix calculation tasks, For time point The rate of change of the task load, is the time interval of the rate of change, is the smoothing coefficient;

[0026] Based on the task volume change rate , according to the rate of change of task volume The size and positive and negative values ​​of the evaluation matrix are used to calculate the changing trend of the task volume and obtain the changing trend information of the task volume.

[0027] The present invention is improved in that the step of obtaining the prediction result of the computing task volume is:

[0028] Based on the processing records of matrix computing tasks, the historical computing task volume of the target period is extracted, the historical average task volume of the target period is calculated, and the average task volume data is obtained;

[0029] Based on the average task volume data and the task volume change trend information, using the deep learning algorithm, through the formula:

[0030]

[0031] Calculate the matrix calculation task volume for the predicted future target period , and get the prediction result of computing task volume, where is the historical average task volume during the target period, is the rate of change of the task load, Calculate the amount of work for the matrix of predicted future target periods, is the smoothing adjustment coefficient, is the time interval factor, is the data reliability coefficient, Adjust the factor for the time period.

[0032] The present invention is improved in that the method for predicting the total processing capacity of the matrix calculation task in the target time period is:

[0033] Based on the real-time task allocation result, extract the task processing rate of each computing unit under standard power consumption conditions to obtain task processing rate information;

[0034] Based on the task processing rate information, combined with the time interval from the current time to the target time period , through the formula:

[0035]

[0036] Calculate the total amount of tasks expected to be processed before the target period , get the expected task processing data, where For the The task processing rate of each computing unit, is the time interval from the current time to the target time period, is the total amount of tasks expected to be processed before the target period, is the adjustment factor, is the balance coefficient, is the attenuation factor, is the total number of computational units.

[0037] The present invention is improved in that the step of obtaining the task load analysis result is:

[0038] Based on the estimated task processing volume data, combined with the matrix calculation task volume data, the remaining task volume is calculated to obtain the remaining task volume information;

[0039] Based on the remaining task amount information and in combination with the total task carrying capacity of the computing unit, the idle amount of the task carrying capacity in the target period is calculated to obtain the idle task carrying capacity information;

[0040] Based on the task carrying capacity idle information and task volume prediction results, the formula is:

[0041]

[0042] Calculate Deviation Task Volume ,like If is positive, then the idle capacity of the task carrying capacity in the target period cannot accommodate the predicted matrix calculation task volume. If it is negative or zero, it means that it can accommodate and obtain the task load analysis result, where Calculate the amount of work for the matrix of predicted future target periods, is the deviation task amount, It is the idle amount of task carrying capacity in the target period.

[0043] The present invention is improved in that the step of obtaining the power consumption adjustment result of the computing unit is:

[0044] Based on the task load analysis result, if the predicted computing task volume cannot be accommodated, the additional task volume that each unit needs to process is calculated according to the deviation task volume;

[0045] Based on the additional amount of tasks that each unit needs to process and the time interval from the current time to the target time period, calculate the task processing rate that each unit needs to increase;

[0046] Based on the task processing rate that each unit needs to increase , through the formula:

[0047]

[0048] Calculate adjusted power consumption , adjust the computing unit immediately, optimize the task processing rate of the computing unit, and obtain the power consumption adjustment result of the computing unit, where, is the standard power consumption, is the power consumption adjustment factor, is the nonlinear response coefficient, For each computing unit, the task processing rate needs to be increased. is the adjusted power consumption.

[0049] Compared with the prior art, the advantages and positive effects of the present invention are:

[0050] In the present invention, by real-time monitoring of the carrying capacity of each computing unit and dynamically allocating matrix computing tasks, the performance of each computing unit can be maximized to avoid overload and waste of resources. By analyzing historical task data and predicting future task volumes, data support is provided for the reasonable allocation of computing resources. By analyzing the matching of current and predicted task volumes with the carrying capacity of the computing units, potential resource shortages can be discovered in advance and adjusted in time. By increasing the power consumption of the computing units to handle additional tasks, continuous computing efficiency and system stability are guaranteed, processing speed and accuracy are improved, the adaptability of the system is enhanced, and it can effectively respond to changing computing needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a system flow chart of the present invention;

[0052] Figure 2 A flow chart of calculating the task carrying capacity ratio of each computing unit in the present invention;

[0053] Figure 3 A flowchart of obtaining the real-time task allocation result of the present invention;

[0054] Figure 4 A flow chart of the changing trend of the matrix calculation task volume analyzed by the present invention;

[0055] Figure 5 A flowchart for obtaining a prediction result of a computing task volume in the present invention;

[0056] Figure 6 A flow chart of the total processing capacity of the matrix calculation task for predicting the target time period according to the present invention;

[0057] Figure 7 A flowchart of obtaining task load analysis results for the present invention;

[0058] Figure 8 The present invention is a flow chart for obtaining the calculation unit power consumption adjustment result. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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 intended to limit the present invention.

[0060] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are 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 cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0061] See also Figure 1 ,The present invention provides a technical solution: a network matrix system based on deep learning, the system includes;

[0062] The task allocation module is based on the matrix calculation tasks to be processed. According to the idle task carrying capacity of each computing unit and the total task carrying capacity of all computing units, it calculates the task carrying capacity ratio of each computing unit, and calculates the matrix calculation task amount allocated to each computing unit according to the matrix calculation task amount and the number of computing units. The matrix calculation tasks are allocated to each computing unit through the network to obtain the real-time task allocation result.

[0063] The task volume prediction module is based on the processing records of matrix calculation tasks and uses deep learning algorithms to analyze the changing trend of matrix calculation task volume according to the matrix calculation task volume data over a period of time. It also predicts the matrix calculation task volume in the future target period based on the average matrix calculation task volume in the target period and obtains the calculation task volume prediction result.

[0064] The carrying capacity analysis module detects the matrix calculation task processing rate of each computing unit under standard power consumption based on the real-time task volume allocation results and task volume prediction results, predicts the total processing volume of the matrix calculation tasks in the target period, calculates the remaining volume of the current task before the target period, analyzes the idle volume of the task carrying capacity in the target period, determines whether the idle volume of the task carrying capacity in the target period can accommodate the predicted matrix calculation task volume, and obtains the task carrying analysis results;

[0065] The computing unit optimization module is based on the task load analysis results. If it cannot accommodate the predicted computing task volume, it will adjust the power consumption of the computing unit in real time according to the excess task volume, optimize the task processing rate of the computing unit, and obtain the computing unit power consumption adjustment result.

[0066] The real-time task allocation results include the task volume and task type information of the computing unit. The computing task volume prediction results include the task growth rate information and the total task volume prediction information within the target period. The task load analysis results include the expected full load rate and expected idle amount of the computing unit. The computing unit power consumption adjustment results include the adjusted power consumption value and the expected processing rate increase.

[0067] See also Figure 2 ,The method for calculating the task carrying capacity ratio of each computing unit is:

[0068] Based on the distributed computing units connected to the network, the idle task carrying capacity of each computing unit is extracted to obtain the task carrying capacity information;

[0069] Based on the task carrying capacity information, the formula is:

[0070]

[0071] Calculate the task carrying capacity ratio of each computing unit , get the task carrying capacity ratio information, where Indicates The task carrying capacity ratio of each computing unit, It is The idle task carrying capacity of computing units, is the total number of computational units, No. The idle task carrying capacity of each computing unit.

[0072] formula:

[0073]

[0074] The meaning and acquisition method of parameters:

[0075] : No. The idle task carrying capacity of a computing unit is a value obtained by evaluating the performance of the computing unit. It is usually the result of a comprehensive evaluation of hardware parameters such as the unit's processor speed, memory capacity, and network bandwidth. For example, the maximum number of tasks or amount of data that each computing unit can handle can be used as a quantitative indicator of its task carrying capacity.

[0076] : The total number of computing units is the number of units actually involved in the calculation, which is usually a specific value obtained based on system design or available resource configuration.

[0077] Calculation example:

[0078] There are three computing units, and their idle task carrying capacities are:

[0079] , , , total number of units .

[0080] Calculate the proportion of task carrying capacity of each unit:

[0081]

[0082] The calculation example shows that computing unit 2 undertakes 40% of the system's tasks, while computing units 1 and 3 undertake 30% and 30% respectively. This proportion information is crucial for the formulation of task allocation and load balancing strategies, and can help system administrators optimize resource utilization and improve overall system efficiency.

[0083] See also Figure 3 ,The steps to obtain the real-time task allocation results are:

[0084] Based on the matrix calculation tasks to be processed and the task carrying capacity ratio information, the formula is:

[0085]

[0086] Calculate the amount of matrix calculation tasks assigned to each computing unit ,in, is the total matrix calculation task, For the The task carrying capacity ratio of each computing unit, For the The amount of matrix calculation tasks assigned to the computing units, is the dynamic adjustment coefficient, is the amount of redundant tasks, is the load balancing factor;

[0087] Based on the amount of matrix calculation tasks assigned to each computing unit , the matrix calculation tasks are allocated to each computing unit through the network to obtain the real-time task allocation results.

[0088] formula:

[0089]

[0090] The meaning and acquisition method of parameters:

[0091] : The total amount of matrix calculation tasks, directly obtained from the matrix calculation tasks to be processed.

[0092] : No. The task carrying capacity ratio of each computing unit is calculated by the previous step.

[0093] : Dynamic adjustment coefficient, used to adapt to dynamic adjustment requirements when system performance fluctuates or load increases. The coefficient is usually preset by experts based on the system's real-time monitoring data and performance analysis.

[0094] : The amount of redundant tasks is set as a certain proportion of the system processing capacity, such as of , to handle potential data peaks or burst tasks.

[0095] : Load balancing coefficient, used to adjust the fairness and uniformity of task allocation to prevent some computing units from being overloaded while other units are idle. It is usually determined based on past load balancing data and system optimization goals.

[0096] Calculation example:

[0097] Set to process The data task volume of the unit, the task carrying capacity of the three computing units is , , , the task carrying capacity of each unit accounts for , , ,set up , (Assume that the amount of redundant tasks is the system processing capacity ), .

[0098] calculate :

[0099]

[0100] calculate :

[0101]

[0102] calculate :

[0103]

[0104] The calculation results show that the second computing unit will be assigned more computing tasks, and each computing unit is assigned additional computing tasks, ensuring flexible response to system performance fluctuations and task peaks while maintaining the balance and fairness of task allocation.

[0105] See also Figure 4 ,The method for analyzing the changing trend of matrix calculation task volume is:

[0106] Based on the processing records of matrix calculation tasks, daily matrix calculation task volume data within a period of time is extracted to obtain task volume data information;

[0107] Based on the task volume data information, through the formula:

[0108]

[0109] Calculate the rate of change of task volume ,in, For time point The amount of matrix calculation tasks, For time point The amount of matrix calculation tasks, For time point The rate of change of the task load, is the time interval of the rate of change, is the smoothing coefficient;

[0110] Based on the rate of change of task volume , according to the rate of change of task volume The size and positive and negative values ​​of the evaluation matrix are used to calculate the changing trend of the task volume and obtain the changing trend information of the task volume.

[0111] formula:

[0112]

[0113] The meaning and acquisition method of parameters:

[0114] : It is at the time point The amount of matrix computing tasks, usually obtained through system logs or real-time monitoring systems. The data reflects the computing load of the system at a specific point in time.

[0115] : At a previous time point The matrix calculation task volume is also obtained through system logs or monitoring system records, and is used to Compare and calculate the changes in the amount of tasks.

[0116] : Smoothing coefficient, used to adjust the sensitivity of the rate of change. It is an empirical value, usually set based on past data analysis to ensure that the rate calculation reflects actual changes while not overreacting to short-term fluctuations.

[0117] : The time interval of the change rate is used to define the time frame for calculating the change rate.

[0118] Calculation example:

[0119] Set at time The amount of tasks is 100 units, time point The amount of tasks is 150 units, smoothing coefficient and time interval day, in order to calculate the rate of change of the task volume within that day.

[0120] The calculation process is as follows:

[0121] Calculate the difference in workload between two time points :

[0122]

[0123] calculate :

[0124]

[0125] Calculation results It means that in the past day, the amount of matrix computing tasks increased at a rate of 55 units per day. The rate can help understand the dynamic changes of matrix computing tasks and can be used to further analyze system load trends or make future load forecasts.

[0126] See also Figure 5 , the steps to obtain the prediction results of the computing task volume are:

[0127] Based on the processing records of matrix calculation tasks, the historical calculation task volume of the target period is extracted and calculated by the formula:

[0128]

[0129] Calculate the historical average task volume for the target period , get the average task volume data, where Indicates Historical task volume data points for target periods, is the number of data points, is the historical average task volume during the target period;

[0130] Based on the average task volume data and the trend information of task volume, using deep learning algorithm, through the formula:

[0131]

[0132] Calculate the matrix calculation task volume for the predicted future target period , and get the prediction result of computing task volume, where is the historical average task volume during the target period, is the rate of change of the task load, Calculate the amount of work for the matrix of predicted future target periods, is the smoothing adjustment coefficient, is the time interval factor, is the data reliability coefficient, Adjust the factor for the time period.

[0133] formula:

[0134]

[0135] The meaning and acquisition method of parameters:

[0136] :The historical average task volume for the target period is obtained by accumulating and averaging historical data points. Usually, the calculation process corresponds to the calculation of the task volume change rate. For example, the task volume change rate is the task change rate within 5 days. In the calculation process of the historical average task volume for the target period, the task volume of the target period within these 5 days needs to be selected.

[0137] : The rate of change of the matrix calculation task volume is calculated using the previous method.

[0138] : The smoothing adjustment coefficient is an empirical value, which is usually adjusted based on the historical forecasting performance of the model to increase or decrease the weight of historical data in the model.

[0139] : Time interval factor, representing the length of time from the current to the target period.

[0140] : Time period adjustment factor, for example, set according to the difference in task volume at night and during the day.

[0141] : Data reliability coefficient is adjusted according to the quality and reliability of historical data. The information provided by high-quality data sources will have a greater impact on the output of the model.

[0142] Calculation example:

[0143] The following data and parameters are set:

[0144] Historical data points during the target period :105, 120, 130, 140, 145 (unit: task volume), record quantity , the rate of change of task volume within 5 days (Unit: Task volume / period), smoothing adjustment coefficient , time interval factor , data reliability coefficient , time period adjustment factor .

[0145] Calculate the historical average task volume :

[0146]

[0147] calculate :

[0148]

[0149] Calculation results , indicating that based on the historical average task volume, change trend, external influencing factors and data quality, the matrix calculation task volume for the future target period is predicted to be approximately Task units. This forecast helps decision makers adjust resources in advance to meet expected computing needs.

[0150] See also Figure 6 ,The method to predict the total processing capacity of matrix calculation tasks in the target period is:

[0151] Based on the real-time task allocation result, the task processing rate of each computing unit under standard power consumption conditions is extracted to obtain the task processing rate information;

[0152] Based on the task processing rate information, combined with the time interval from the current time to the target time period , through the formula:

[0153]

[0154] Calculate the total amount of tasks expected to be processed before the target period , get the expected task processing data, where For the The task processing rate of each computing unit, is the time interval from the current time to the target time period, is the total amount of tasks expected to be processed before the target period, is the adjustment factor, is the balance coefficient, is the attenuation factor, is the total number of computational units.

[0155] formula:

[0156]

[0157] The meaning and acquisition method of parameters:

[0158] :This means The task processing rate of a computing unit, that is, the number of tasks that the unit can process per time unit. The parameter is usually obtained through evaluation of actual observation data and can be determined by long-term monitoring of the average rate at which the unit processes a specific task type.

[0159] : The time interval from the current time to the target period. The parameter is calculated based on the target period and the current date and time.

[0160] : Adjustment coefficient, which is used to adjust the weight of each computing unit in the total task volume calculation according to its historical performance data. The coefficient can be obtained by statistical analysis based on past performance data, for example, by analyzing the performance fluctuations of each unit under different workloads.

[0161] : The balance coefficient is usually a fixed value, which aims to add a stabilizing factor to the total amount calculation and help balance the calculation results to cope with potential unforeseen task volume fluctuations. The coefficient can be optimized by simulating different task fluctuation scenarios to ensure calculation stability under various conditions.

[0162] : Attenuation factor, used to reduce the negative impact of long time intervals on prediction accuracy and ensure that recent prediction results are more authoritative. The parameters are usually set by analyzing the relationship between the time span and prediction accuracy in historical data.

[0163] Calculation example:

[0164] Set the parameters as follows:

[0165] Task processing rate of three computing units , , Tasks / hour, target time span Hours, adjustment factor , balance coefficient , attenuation factor , total number of units .

[0166] Calculation process :

[0167]

[0168] Calculation results The task indicates that in the next 24 hours, all computing units together are expected to process approximately 9194 task units.

[0169] See also Figure 7 , the steps to obtain the task load analysis results are:

[0170] Based on the estimated task processing data, combined with the matrix calculation task data, the formula is:

[0171]

[0172] Calculate the remaining tasks , get the remaining task information, where is the remaining task amount, is the total matrix calculation task, The total amount of tasks expected to be processed before the target period;

[0173] Based on the remaining task information and the total task carrying capacity of the computing unit, the formula is:

[0174]

[0175] Calculate the idle capacity of the task carrying capacity in the target period , get the task carrying capacity idle information, where, is the sum of the task carrying capacity of the computing units, is the remaining task amount, is the idle amount of task carrying capacity in the target period;

[0176] Based on the task carrying capacity idle information and task volume prediction results, the formula is:

[0177]

[0178] Calculate Deviation Task Volume ,like If is positive, then the idle capacity of the task carrying capacity in the target period cannot accommodate the predicted matrix calculation task volume. If it is negative or zero, it means that it can accommodate and obtain the task load analysis result, where Calculate the amount of work for the matrix of predicted future target periods, is the deviation task amount, It is the idle amount of task carrying capacity in the target period.

[0179] formula:

[0180]

[0181] and

[0182]

[0183] and

[0184]

[0185] The meaning and acquisition method of parameters:

[0186] : The total amount of matrix calculation tasks, directly obtained from the matrix calculation tasks to be processed.

[0187] : The total amount of tasks that all units are expected to handle, calculated using the previous method.

[0188] : The sum of the task carrying capacity of all computing units, calculated as , is the total number of computational units, No. The idle task carrying capacity of each computing unit.

[0189] : The matrix calculation task volume for the predicted future target period is obtained by the previous method.

[0190] Calculation example:

[0191] Set the parameters as follows:

[0192] Total matrix calculation tasks Task, total amount of computing unit processing Task, the sum of the task carrying capacity of all computing units Task, the matrix calculation task volume for the predicted future target period Task.

[0193] Calculation process:

[0194] Calculate the remaining tasks :

[0195]

[0196] Calculate the idle capacity of the task carrying capacity in the target period :

[0197]

[0198] Calculate Deviation Task Volume :

[0199]

[0200] Calculation results Task, indicating that the idle amount of task carrying capacity in the target period can accommodate the predicted matrix calculation task volume, and there is additional idle capacity, indicating that the current system configuration and plan can meet future task requirements, and there is still room to handle more tasks or respond to emergencies, indicating that the system will not be overloaded during the predetermined target period.

[0201] See also Figure 8 , the steps for obtaining the calculation unit power consumption adjustment result are:

[0202] Based on the task load analysis results, if the predicted computing task volume cannot be accommodated, the deviation task volume is calculated by the formula:

[0203]

[0204] Calculate the amount of additional work each unit needs to handle ,in, is the deviation task amount, is the total number of computational units, The amount of additional tasks that need to be handled by each unit;

[0205] Based on the amount of additional work each unit needs to handle , combined with the time interval from the current time to the target time period, through the formula:

[0206]

[0207] Calculate the task processing rate that needs to be increased for each unit ,in, is the time interval from the current time to the target time period, The amount of additional work that each unit needs to handle, The task processing rate that needs to be increased for each unit;

[0208] Based on the need for each unit to increase the task processing rate , through the formula:

[0209]

[0210] Calculate adjusted power consumption , adjust the computing unit immediately, optimize the task processing rate of the computing unit, and obtain the power consumption adjustment result of the computing unit, where, is the standard power consumption, is the power consumption adjustment factor, is the nonlinear response coefficient, For each computing unit, the task processing rate needs to be increased. is the adjusted power consumption.

[0211] formula:

[0212]

[0213] and

[0214]

[0215] and

[0216]

[0217] The meaning and acquisition method of parameters:

[0218] : Deviation task amount, calculated by the previous method.

[0219] :The number of units participating in the calculation is the actual number of units participating in the calculation, which is usually a specific value obtained based on system design or available resource configuration.

[0220] : The time interval from the current time to the target time period.

[0221] : The standard power consumption of the computing unit. Based on the design parameters of the computing unit or the specifications provided by the manufacturer, it can also be set to the most energy-efficient power consumption.

[0222] : Power consumption adjustment factor, which indicates the additional power consumption required for each unit increase in processing rate. The factor is set based on the energy efficiency evaluation or historical energy consumption data of the device.

[0223] : Nonlinear response coefficient, used to describe the nonlinear response of increased power consumption to performance improvement. The value can be determined through experiments or simulations based on the specific energy consumption performance curve of the computing unit.

[0224] Calculation example:

[0225] Set the parameters as follows:

[0226] Deviation of the task volume Task, number of computing units , time interval hours, standard power consumption Watt, power consumption adjustment factor Watts / task / hour.

[0227] Calculate the additional task processing requirements for each unit:

[0228]

[0229] Calculate the task processing rate difference:

[0230]

[0231] Calculate the adjusted power consumption:

[0232]

[0233] Calculation results Watts means that to handle the additional workload, the power consumption of each computing unit needs to increase from 100 watts to about 163 watts. This increase reflects that each computing unit must consume more energy to meet performance requirements under a given workload.

[0234] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A network matrix system based on deep learning, characterized in that: The system comprises: The task allocation module calculates the task carrying capacity ratio of each computing unit based on the matrix calculation tasks to be processed and the idle task carrying capacity of each computing unit. It also allocates the matrix calculation tasks to each computing unit through the network according to the matrix calculation task volume and the number of computing units, and obtains the real-time task allocation result. The task volume prediction module uses deep learning algorithms to analyze the changing trend of matrix calculation task volume based on the processing records of matrix calculation tasks, predict the matrix calculation task volume in the future target period, and obtain the calculation task volume prediction results; The method for analyzing the changing trend of the task volume calculated by the matrix is: Based on the processing records of matrix calculation tasks, daily matrix calculation task volume data within a period of time is extracted to obtain task volume data information; Based on the task volume data information, through the formula: Calculate the rate of change of task volume ,in, For time point The amount of matrix calculation tasks, For time point The amount of matrix calculation tasks, For time point The rate of change of the task load, is the time interval of the rate of change, is the smoothing coefficient; Based on the task volume change rate , according to the rate of change of task volume The size and positive and negative of the value are used to evaluate the changing trend of the task volume calculated by the matrix, and the changing trend information of the task volume is obtained; The steps for obtaining the prediction result of the computing task volume are as follows: Based on the processing records of matrix computing tasks, the historical computing task volume of the target period is extracted, the historical average task volume of the target period is calculated, and the average task volume data is obtained; Based on the average task volume data and the task volume change trend information, using the deep learning algorithm, through the formula: Calculate the matrix calculation task volume for the predicted future target period , and get the prediction result of computing task volume, where is the historical average task volume during the target period, is the rate of change of the task load, Calculate the amount of work for the matrix of predicted future target periods, is the smoothing adjustment coefficient, is the time interval factor, is the data reliability coefficient, is the time period adjustment factor; The carrying capacity analysis module detects the matrix calculation task processing rate of each computing unit under standard power consumption based on the real-time task volume allocation result and the task volume prediction result, predicts the total processing volume of the matrix calculation tasks in the target period, and determines whether the idle amount of the task carrying capacity in the target period can accommodate the predicted matrix calculation task volume, thereby obtaining the task carrying analysis result; Based on the task load analysis results, if the predicted computing task volume cannot be accommodated, the computing unit optimization module will adjust the power consumption of the computing unit in real time according to the excess task volume, optimize the task processing rate of the computing unit, and obtain the computing unit power consumption adjustment result.

2. The deep learning-based network matrix system according to claim 1, characterized in that: The method for calculating the task carrying capacity ratio of each computing unit is: Based on the distributed computing units connected to the network, the idle task carrying capacity of each computing unit is extracted to obtain the task carrying capacity information; Based on the task carrying capacity information, the formula is: Calculate the task carrying capacity ratio of each computing unit , get the task carrying capacity ratio information, where Indicates The task carrying capacity ratio of each computing unit, It is The idle task carrying capacity of computing units, is the total number of computational units, No. The idle task carrying capacity of each computing unit.

3. The deep learning-based network matrix system according to claim 2, characterized in that: The steps for obtaining the real-time task allocation result are: Based on the matrix calculation tasks to be processed and the task carrying capacity ratio information, the formula is: Calculate the amount of matrix calculation tasks assigned to each computing unit ,in, is the total matrix calculation task, For the The task carrying capacity ratio of each computing unit, For the The amount of matrix calculation tasks assigned to the computing units, is the dynamic adjustment coefficient, is the amount of redundant tasks, is the load balancing factor; Based on the matrix calculation task amount assigned to each computing unit , the matrix calculation tasks are allocated to each computing unit through the network to obtain the real-time task allocation results.

4. The deep learning-based network matrix system according to claim 1, characterized in that: The method for predicting the total processing capacity of the matrix calculation task in the target time period is: Based on the real-time task allocation result, extract the task processing rate of each computing unit under standard power consumption conditions to obtain task processing rate information; Based on the task processing rate information, combined with the time interval from the current time to the target time period , through the formula: Calculate the total amount of tasks expected to be processed before the target period , get the expected task processing data, where For the The task processing rate of each computing unit, is the time interval from the current time to the target time period, is the total amount of tasks expected to be processed before the target period, is the adjustment factor, is the balance coefficient, is the attenuation factor, is the total number of computational units.

5. The deep learning-based network matrix system according to claim 4, characterized in that: The steps for obtaining the task load analysis result are as follows: Based on the estimated task processing volume data, combined with the matrix calculation task volume data, the remaining task volume is calculated to obtain the remaining task volume information; Based on the remaining task amount information and in combination with the total task carrying capacity of the computing unit, the idle amount of the task carrying capacity in the target period is calculated to obtain the idle task carrying capacity information; Based on the task carrying capacity idle information and task volume prediction results, the formula is: Calculate Deviation Task Volume ,like If is positive, then the idle capacity of the task carrying capacity in the target period cannot accommodate the predicted matrix calculation task volume. If it is negative or zero, it means that it can accommodate and obtain the task load analysis result, where Calculate the amount of work for the matrix of predicted future target periods, is the deviation task amount, It is the idle amount of task carrying capacity in the target period.

6. The network matrix system based on deep learning according to claim 1, characterized in that: The steps of obtaining the calculation unit power consumption adjustment result are as follows: Based on the task load analysis result, if the predicted computing task volume cannot be accommodated, the additional task volume that each unit needs to process is calculated according to the deviation task volume; Based on the additional amount of tasks that each unit needs to process and the time interval from the current time to the target time period, calculate the task processing rate that each unit needs to increase; Based on the task processing rate that each unit needs to increase , through the formula: Calculate adjusted power consumption , adjust the computing unit immediately, optimize the task processing rate of the computing unit, and obtain the power consumption adjustment result of the computing unit, where, is the standard power consumption, is the power consumption adjustment factor, is the nonlinear response coefficient, For each computing unit, the task processing rate needs to be increased. is the adjusted power consumption.

Images