Multi-core processor dynamic regulation and control method and system based on load and temperature sensing

Abstract

The invention discloses a multi-core processor dynamic regulation and control method and system based on load and temperature sensing, and relates to the technical field of temperature control scheduling management, and the method comprises the steps: collecting and preprocessing load and temperature data in real time, and building a thermal sensing model to predict the temperature change; constructing a self-adaptive heuristic scheduling algorithm, and adjusting task distribution according to real-time data of the core load and the temperature; according to the load and temperature states, a dynamic voltage frequency adjustment strategy is adopted, and the frequency and the voltage are automatically adjusted when the load and the temperature are too high; and when the load and the temperature exceed set threshold values, a scheduling algorithm is optimized by adopting a self-learning mechanism, and a scheduling strategy and a temperature control scheme are automatically adjusted. According to the method, a more flexible and accurate dynamic regulation and control scheme is provided, the stability and high efficiency of the processor in a high-load and high-temperature environment can be ensured, performance reduction and hardware damage are avoided, and the obvious effects of saving energy and improving system stability are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of temperature control scheduling management, in particular to a multi-core processor dynamic regulation method and system based on load and temperature perception. BACKGROUND

[0002] With the development of computing technology, the performance and power consumption of processors have become increasingly prominent, especially in multi-core processors. How to effectively manage power consumption and heat without affecting system performance has become an important research direction. Multi-core processors face more challenges in load distribution and temperature control due to the increase in the number of cores. To solve these problems, modern multi-core processors usually combine dynamic voltage and frequency scaling (DVFS) technology and load balancing technology to reduce power consumption while optimizing performance. DVFS technology dynamically adjusts the working frequency and voltage of the processor core, which can control power consumption and heat according to the current task load and temperature state, and has become a common means to solve the problem of processor power consumption. On the other hand, the introduction of load perception and temperature perception models enables multi-core processors to more accurately adjust the working state, thereby providing more optimization space in task allocation, scheduling, and voltage frequency adjustment. In recent years, intelligent scheduling systems combining load and temperature perception have received widespread attention, and many research and products have begun to integrate load and temperature control scheduling algorithms based on real-time data collection, trying to further improve the stability and efficiency of processors under complex workloads.

[0003] However, existing multi-core processor dynamic regulation methods mainly have the following deficiencies: First, most existing methods use static strategies or single temperature control strategies for task scheduling, lacking adaptive adjustment mechanisms according to real-time changes in load and temperature. For processors under high load and high temperature conditions, existing methods often rely on simple frequency reduction strategies, which can lead to performance loss, especially when handling multiple tasks or high loads, and cannot dynamically balance the relationship between performance and power consumption. Second, although dynamic voltage and frequency scaling (DVFS) can effectively control power consumption, the adjustment range is limited in extreme load or high temperature environments, and the load and temperature differences between different cores are not fully considered, leading to local overheating or load imbalance problems. Existing scheduling algorithms often fail to achieve coordinated optimization of load and temperature, resulting in suboptimal heat management and task allocation, which in turn affects the overall performance of multi-core processors. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a multi-core processor dynamic regulation method based on load and temperature perception to solve the problems of inaccurate load and temperature perception, inflexible scheduling strategy, and insufficient balance between power consumption and performance in the prior art.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In the first aspect, the present application provides a multi-core processor dynamic regulation method based on load and temperature sensing, which comprises collecting and preprocessing load and temperature data in real time, establishing a thermal sensing model to predict temperature changes;

[0008] An adaptive heuristic scheduling algorithm is constructed to adjust task allocation according to real-time data of core load and temperature;

[0009] According to the load and temperature state, a dynamic voltage frequency adjustment strategy is adopted to automatically adjust the frequency and voltage when the load is too high and the temperature is too high;

[0010] When the load and temperature exceed the set threshold, a self-learning mechanism is used to optimize the scheduling algorithm to automatically adjust the scheduling strategy and temperature control scheme.

[0011] As a preferred scheme of the multi-core processor dynamic regulation method based on load and temperature sensing, the preprocessing of load and temperature data includes real-time collection of load data of each core through a monitoring tool hardware interface, smoothing through weighted average, removing noise and ensuring data stability;

[0012] Temperature data of each core is collected in real time by a temperature sensor and smoothed using a weighted average method;

[0013] Load data and temperature data are standardized. Load data is standardized by ratio to the maximum load of the system, and temperature data is standardized by ratio to the maximum temperature of the system.

[0014] As a preferred scheme of the multi-core processor dynamic regulation method based on load and temperature sensing, the establishment of the thermal sensing model includes modeling the temperature change of each core through the heat conduction equation; the change of temperature is composed of power consumption, core heat capacity and heat conduction coefficient; the change of core temperature is calculated at each time step through the discretization equation;

[0015] Based on historical load, power consumption and temperature data, the coefficients are trained through regression analysis to predict the temperature change of each core according to real-time power consumption and load;

[0016] Set the upper limit of temperature and the warning temperature threshold, when the temperature approaches or exceeds the warning value, trigger the self-learning mechanism to optimize the scheduling algorithm for scheduling; combine real-time load and temperature data to dynamically adjust the frequency, voltage and task allocation strategy of the core.

[0017] As a preferred scheme of the load and temperature-aware multi-core processor dynamic regulation method, wherein: the adaptive heuristic scheduling algorithm comprises that a population is composed of multiple individuals, each individual represents a task scheduling scheme; there are n cores and m tasks, each individual is represented by an mxn matrix, wherein the elements of the matrix represent whether the task is assigned to the core;

[0018] A fitness function is constructed to evaluate the priority of each core based on the load balancing, temperature control effect and frequency adjustment of the task scheduling scheme; the fitness of the individual is calculated according to the load and temperature of the scheduled task;

[0019] An adaptive selection mechanism is introduced, the selection probability of the individual is calculated according to the current fitness and historical fitness, and a diversity protection method is used to assist adjustment; different crossover strategies are formulated according to the types of compute-intensive tasks and I / O-intensive tasks, and domain-specific crossover is performed;

[0020] The guided mutation is used to determine the tasks of the mutation according to the task allocation of the cores with lower load and lower temperature in the current population; the local search mutation combines a local optimization algorithm to further optimize the quality of the individual after guided mutation; an elite reservation and partial replacement strategy is used to screen the individuals after mutation.

[0021] As a preferred scheme of the load and temperature-aware multi-core processor dynamic regulation method, wherein: the dynamic voltage and frequency adjustment strategy comprises: real-time monitoring the load and temperature of each core, and collecting data using a sensor monitoring module; the load data L i and the temperature data T i respectively represent the real-time load and temperature of the ith core; according to the real-time collected load and temperature data, the priority of each core is calculated, and an evaluation function P i is used to represent the task priority of each core, and an adaptive heuristic scheduling algorithm is used to output an optimized scheduling strategy, and the scheduling rules specifically include:

[0022] The cores with lower load and lower temperature predicted by the thermal-aware model are preferentially selected to avoid cores with too high load and temperature; when the load and temperature of a core are predicted to exceed the set threshold, the task is automatically migrated to a core with lower load and temperature;

[0023] When the load L i of a core is detected to exceed the set threshold L max , the frequency f i of the core is automatically reduced; when the temperature T i exceeds the set threshold T max , the frequency f i and the voltage V i; when the temperature rises, the frequency f of the core is automatically adjusted i Dynamic control is performed; when the core temperature T i increases, the frequency f i will be reduced in proportion: the voltage V i changes according to frequency adjustment; when the load and temperature exceed the threshold value at the same time, first reduce the frequency and voltage to reduce the burden, and adjust the task scheduling strategy according to the real-time load and temperature.

[0024] As a preferred scheme of the load and temperature aware multi-core processor dynamic regulation method of the application, wherein: the self-learning mechanism is adopted to optimize the scheduling algorithm, including that in the self-learning mechanism, each state-action pair (s t ,a t ) is initialized to 0, indicating that there is no prior knowledge at the initial time for all scheduling and temperature control schemes;

[0025] The heuristic rule output by the adaptive heuristic scheduling algorithm based on task scheduling initializes the scheduling strategy, and automatically allocates tasks to cores with low load and low temperature;

[0026] Set the learning parameters, and the system needs to perceive the current state s t at each time t, which specifically includes:

[0027] The core load L i (t) represents the load of core i at time t; the core temperature T i (t) represents the temperature of core i at time t; the state vector s t is given to all core load and temperature information;

[0028] The state s t is taken as the input of the Q-learning algorithm, and the appropriate action is selected through the reinforcement learning algorithm;

[0029] The priority of each core is calculated using the aforementioned evaluation function P i ; the state vector s t is calculated in real time by the evaluation function P i to calculate the task priority of each core;

[0030] According to the results of the evaluation function, the emergency task scheduling is performed when the set threshold value is exceeded.

[0031] As a preferred scheme of the load and temperature aware multi-core processor dynamic regulation method of the application, wherein: the automatic adjustment of the scheduling strategy and the temperature control scheme includes adjusting the scheduling strategy according to real-time data feedback; at each time step, the algorithm selects a suitable scheduling action according to the current load and temperature information;

[0032] The Q value is updated by a reward signal, iteration is performed until an optimal scheduling strategy and temperature control scheme are found; the scheduling strategy is adjusted based on Q-learning update, in the training process of Q-learning, the scheduling strategy is automatically adjusted according to the updated Q value; at each new time step, the optimal action is selected according to the current state to realize load balancing and temperature control.

[0033] In the second aspect, the application provides a load and temperature sensing based multi-core processor dynamic regulation system, comprising a sensor monitoring module, which collects real-time load and temperature data of cores and pre-processes the data;

[0034] A scheduling algorithm module dynamically adjusts task allocation according to real-time load and temperature data to realize load balancing and temperature control.

[0035] An adjustment module automatically adjusts the frequency and voltage of cores by using a dynamic voltage frequency adjustment strategy according to real-time load and temperature states to realize more efficient temperature control.

[0036] A self-learning optimization module automatically optimizes the scheduling strategy and temperature control scheme by using real-time feedback and historical data of the system.

[0037] In the third aspect, the application provides a computer device comprising a memory and a processor, and the memory stores a computer program, wherein the computer program is executed by the processor to realize any step of the load and temperature sensing based multi-core processor dynamic regulation method according to the first aspect of the application.

[0038] In the fourth aspect, the application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by the processor to realize any step of the load and temperature sensing based multi-core processor dynamic regulation method according to the first aspect of the application.

[0039] The application has the beneficial effects that: by collecting and preprocessing the load and temperature data in real time, combining with the heat perception model to accurately predict temperature changes, the task scheduling and power consumption management of the multi-core processor can be effectively optimized. The method uses an adaptive heuristic scheduling algorithm, which can dynamically adjust task allocation according to the load and temperature of the core, avoiding task concentration on high-load or high-temperature cores, thereby improving the overall efficiency of the system. First, real-time monitoring of load and temperature changes can dynamically adjust the frequency and voltage according to the system state, avoiding excessive power consumption and heat accumulation, effectively reducing energy consumption and temperature. Second, using a self-learning mechanism, the scheduling strategy can be optimized according to actual feedback during operation, gradually improving the intelligent level of the system, adapting to different workloads and environmental conditions. In addition, the method optimizes load and temperature control in coordination, avoiding the limitations of independent temperature and load adjustment in traditional technologies, improving the accuracy of task scheduling and the effect of power consumption management. Compared with the prior art, the application provides a more flexible and accurate dynamic control scheme, which can ensure the stability and efficiency of the processor under high load and high temperature environment, avoid performance degradation and hardware damage, and has significant energy saving and system stability improvement effect. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating labor.

[0041] Figure 1 Flow chart of the multi-core processor dynamic regulation method based on load and temperature perception. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0043] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0044] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0045] Reference Figure 1 For an embodiment of the present application, the embodiment provides a load and temperature-aware dynamic multi-core processor regulation method, comprising the following steps:

[0046] S1: Real-time collection and preprocessing of load and temperature data, establishment of thermal-aware model to predict temperature changes.

[0047] Further, the preprocessing of load and temperature data includes real-time collection of load data of each core through monitoring tool hardware interface, smoothing through weighted average, removal of noise and ensuring data smoothness.

[0048] Real-time collection of temperature data of each core through temperature sensor, smoothing of temperature data using weighted average method.

[0049] Standardization processing of load data and temperature data; load data is standardized by ratio to system maximum load, and temperature data is standardized by ratio to system maximum temperature.

[0050] The establishment of thermal-aware model includes modeling of temperature changes of each core through heat conduction equation; temperature changes are composed of power consumption, core heat capacity and heat conduction coefficient; through discretization equation, core temperature changes are calculated within each time step.

[0051] Based on historical load, power consumption and temperature data, coefficients are trained through regression analysis, and temperature change of each core is predicted according to real-time power consumption and load.

[0052] Setting temperature upper limit and warning temperature threshold, when temperature approaches or exceeds warning value, triggering self-learning mechanism to optimize scheduling algorithm for scheduling; combining real-time load and temperature data, dynamically adjusting core frequency, voltage and task allocation strategy.

[0053] Real-time monitoring: Real-time acquisition of load conditions of each core, represented as core working intensity (e.g. CPU occupancy rate).

[0054] Load data: For each core i, load data L i (t) represents the load condition of the i-th core at time t. Load data is usually obtained through the scheduler of the operating system or a custom load monitoring tool, represented as CPU usage rate.

[0055] Smoothing filter: In order to remove noise in load data, data is smoothed. The specific way is to use weighted average method:

[0056] Smoothed load: The smoothed load L i (t) will be based on the load value L i(t-1) and the current load L i (t) is calculated by weighted average, with a smoothing factor of α. This means that the load data at the current time will jointly affect the final smoothing result with the load value at the previous time. The specific calculation method is:

[0057] L i (t) = α·L i (t-1) + (1-α)·L i (t)

[0058] Where α is the smoothing factor, controlling the degree of influence of historical load data on the current load value.

[0059] Collect core temperature data, temperature monitoring: the temperature T i (t) of each core is obtained through the built-in sensor. If there is noise in the data, the temperature data is also smoothed.

[0060] Smoothed temperature: temperature data T i (t) will be based on the temperature value T i (t-1) at the previous time and the temperature value T i (t) at the current time. The weighted average is calculated with a smoothing factor of β, and the formula is:

[0061] T i (t) = β·T i (t-1) + (1-β)·T i (t)

[0062] Where β is the temperature smoothing factor, used to adjust the weighted proportion of the previous and subsequent temperature data.

[0063] Load and temperature normalization: In order to ensure the consistency of the subsequent calculation process, the load and temperature are normalized to make them change within a unified range:

[0064] Normalized load: load value L i (t) will be normalized according to the maximum load Load_max(t) in the system:

[0065]

[0066] Normalized temperature: temperature value T i (t) will be normalized according to the maximum temperature Temp_max(t) in the system:

[0067]

[0068] Establish a heat perception model. Heat conduction model: the power consumption P i(t) will affect its temperature. According to the heat conduction equation, the temperature change T i (t) can be described by its power consumption, heat capacity, and heat conduction coefficient.

[0069] Further, a thermal-aware model is established to predict the temperature change. The temperature change of each core is modeled by the heat conduction equation, considering the power consumption, heat capacity of the core, and heat conduction coefficient. This is to accurately simulate the temperature change of each core, and based on historical load, power consumption, and temperature data, the model coefficients are trained using regression analysis, so that under the condition of real-time load and power consumption change, the temperature change of each core can be accurately predicted. The prediction of temperature provides an important basis for subsequent temperature control strategy and scheduling optimization.

[0070] The heat capacity and heat conduction, the temperature change of the core is determined by the power consumption P i (t) of the core, the heat capacity C i of the core, and the heat conduction coefficient R i . This relationship can be represented by the heat equation as:

[0071]

[0072] Where σ i is the heat loss due to heat loss, C i is the heat capacity of the core, which represents the ability of the core to store heat. Temperature prediction and temperature threshold setting, temperature prediction: predict the temperature change of each core according to historical power consumption and load data. Specifically, the temperature change ΔT i (t) can be represented by the weighted average of the power consumption P i (t) of the core and the load L i (t):

[0073] ΔT i (t) = α·P i (t) + β·L i (t)

[0074] Here, α and β are coefficients fitted by historical data.

[0075] It should be noted that the load and temperature data of each core are collected in real time, and the data are collected through a hardware interface. The goal of this step is to obtain real-time working status for subsequent analysis. The load and temperature data are smoothed by a weighted average method to remove noise in the data and make the data more stable and reliable. Then, the data are standardized: the load data are standardized by the ratio to the maximum load of the system, and the temperature data are standardized by the ratio to the maximum temperature of the system. This step ensures that the data of different cores can be compared on the same scale, thereby providing unified basic data for subsequent scheduling optimization.

[0076] S2: Constructing an adaptive heuristic scheduling algorithm, adjusting task allocation according to real-time data of core load and temperature.

[0077] Further, the adaptive heuristic scheduling algorithm comprises a population consisting of multiple individuals, each individual representing a task scheduling scheme; there are n cores and m tasks, and each individual is represented by an m x n matrix, wherein the elements of the matrix represent whether the task is allocated to the core.

[0078] First, initialize the population, which consists of multiple individuals, each individual representing a task scheduling scheme. The size of the population is pop_size.

[0079] A joint evaluation function is used to evaluate the fitness of each individual. This function combines factors such as load, temperature, frequency, and voltage to evaluate the priority of each core, thereby determining the effectiveness of task scheduling. The evaluation function formula is:

[0080] P i i = w1·L i i + w2·T i

[0081] where: P i i is the priority of core i; L i i is the load of core i; T i i is the temperature of core i; w1 and w2 are the weight coefficients of load and temperature, respectively.

[0082] The fitness of each individual is calculated based on the load and temperature of the tasks it schedules. The higher the fitness, the more balanced the load and the better the temperature control.

[0083] Construct the fitness function to evaluate the priority of each core based on load balancing, temperature control effect, and frequency adjustment of the task scheduling scheme; the fitness of the individual is calculated based on the load and temperature of the scheduled tasks;

[0084] An adaptive selection mechanism is introduced, where the selection probability of an individual is calculated based on the current fitness and the weighted historical fitness, and a diversity protection method is used to assist in adjusting the selection probability. Different crossover strategies are developed for computationally intensive tasks and I / O intensive tasks, and domain-specific crossover is performed.

[0085] In traditional selection mechanisms, individuals with higher fitness are more likely to be selected, which can lead to a decrease in population diversity and cause the algorithm to fall into a local optimal solution. To address this issue, an adaptive selection mechanism is introduced, which dynamically adjusts the selection probability by considering the historical performance of individuals and population diversity. The historical performance of each individual reflects its fitness in the past few generations. The historical fitness average is used to evaluate the performance of an individual, which is represented by the formula:

[0086]

[0087] where f i (t) is the fitness of individual i in generation t, and k is the number of generations evaluated. Individuals with better historical performance will be selected preferentially.

[0088] To protect the diversity of the population and avoid premature convergence, a diversity index D is introduced to evaluate the diversity of the population. When the diversity is low, the selection pressure is reduced, allowing more exploration; when the diversity is high, the selection pressure is increased, accelerating convergence.

[0089] The selection pressure S is dynamically adjusted based on the diversity of the population:

[0090]

[0091] where α is the adjustment factor and D is the diversity measure. The selection pressure increases as the diversity decreases, thus driving the search towards the optimal solution.

[0092] The roulette wheel selection and tournament selection are combined to perform parent selection. The selection probability of an individual is proportional to its fitness and historical performance, ensuring that excellent individuals can be passed on to the next generation with high probability.

[0093] In traditional single-point crossover, the genes of the parents are exchanged at a single point, which may not effectively explore the global solution space. Therefore, multi-point crossover is used, which exchanges genes at multiple crossover points. Multi-point crossover can increase the diversity of the population and avoid local optimal solutions. Suppose the genes of parent 1 and parent 2 are:

[0094] Parent 1: (x1, x2, …, x k , …, x m )

[0095] Parent 2: (y1, y2, …, y ky,..., y m )

[0096] Select multiple cross points (for example, cross points k1 and k2), and then exchange the parent genes before and after these points to obtain two child individuals:

[0097]

[0098] For task scheduling problems, considering the dependency relationship between tasks and cores, domain-specific crossover is adopted. For example, compute-intensive tasks and I / O-intensive tasks should adopt different crossover strategies. For each task, the characteristics of the task and its impact on core load, temperature should be considered when crossing. For example, compute-intensive tasks are more suitable for allocation to cores with lower load, while I / O-intensive tasks can be preferentially allocated to cores with lower temperature. When crossing, these constraints should be considered to ensure reasonable task allocation.

[0099] Using guided mutation, traditional mutation methods may destroy the effective solution structure by randomly changing individual genes. To improve this, guided mutation is adopted. In guided mutation, the mutation operation is not just random, but is based on the task allocation of the cores with lower load or lower temperature in the current population to select the mutation task. For example, select the task on the core with lower load for mutation, and migrate it to the core with higher load but lower temperature. Guided mutation can be more in line with system requirements and avoid unnecessary exploration. Local search mutation combines local optimization algorithms (such as hill climbing, simulated annealing, etc.) to further optimize the mutated individual. The goal of local search mutation is to fine-tune the individual and further improve the scheduling effect

[0100] It should be noted that guided mutation is used to determine the task of variation according to the task allocation of the cores with lower load and lower temperature in the current population; local search mutation combines local optimization algorithms to further optimize the quality of the solution by fine-tuning the individual after guided mutation; and an elite preservation and partial replacement strategy is used to screen the individual after mutation.

[0101] S3: According to the load and temperature state, a dynamic voltage and frequency adjustment strategy is adopted to automatically adjust the frequency and voltage when the load is too high and the temperature is too high.

[0102] Further, the dynamic voltage and frequency adjustment strategy includes real-time monitoring of the load and temperature of each core, using a sensor monitoring module to collect data; the load data L i and temperature data T i represent the real-time load and temperature of the i-th core, respectively; according to the real-time collected load and temperature data, the priority of each core is calculated, and an evaluation function P iThe task priority of each core is represented, and an adaptive heuristic scheduling algorithm is used to output an optimized scheduling strategy. The scheduling rules specifically include:

[0103] The core with lower load and lower temperature predicted by the thermal-aware model is preferentially selected to avoid the core with too high load and temperature. When the load and temperature of the core are predicted to exceed the set threshold, the task is automatically migrated to the core with lower load and temperature.

[0104] Temperature threshold setting: In order to avoid system failure caused by too high temperature, the temperature threshold is set as:

[0105] Upper temperature T max and warning temperature T warn : T max = 85℃, T warn = 75℃

[0106] Load-aware scheduling algorithm design, load balancing strategy, shortest job first (SJF): The shortest job first strategy is used for task scheduling. That is, tasks are preferentially allocated to cores with lower load. Specifically, the core i with the lightest load is selected:

[0107]

[0108] Where N is the total number of cores, and the task is allocated to the core with the smallest load.

[0109] Dynamic task allocation, temperature and load joint consideration: In order to ensure that task allocation is not only based on load, but also considers temperature, a weighted load and temperature value is used to determine task allocation:

[0110]

[0111] Where λ is the weight coefficient, used to balance the influence of load and temperature, to ensure system load balancing and temperature not exceeding the threshold.

[0112] Adaptive frequency and voltage control (DVFS), frequency adjustment strategy, frequency regulation: When the temperature of the core is higher, its frequency is reduced to reduce power consumption and slow down temperature rise. The frequency f i (t) will be adjusted according to the temperature T i (t) of the core:

[0113] f i (t) = f max - γ·T i (t) norm

[0114] Where f maxis the maximum frequency, γ is the temperature adjustment factor of the frequency. Voltage adjustment algorithm, voltage and frequency linkage: voltage V i (t) and frequency f i (t) is proportional. When the frequency decreases, the voltage also decreases accordingly:

[0115]

[0116] where V max is the maximum voltage, voltage adjustment and frequency adjustment linkage, to ensure that power consumption is reduced.

[0117] The scheduling strategy of combining load and temperature into a unified optimization framework. The scheduling of tasks not only considers the load, but also considers the core temperature. Under this framework, task allocation and frequency control are dynamically adjusted:

[0118]

[0119] where the Cost function considers the impact of load and temperature on scheduling, ensuring that each core is neither overloaded nor overheated.

[0120] When the load L i of the core is detected to exceed the set threshold L max , the frequency f i of the core is automatically reduced; when the temperature T i exceeds the set threshold T max , the frequency f i and voltage V i of the core are reduced; when the temperature rises, the frequency f i of the core is automatically adjusted for dynamic control; when the core temperature T i rises, the frequency f i will be reduced in proportion: the voltage V i will change according to the frequency adjustment; when the load and temperature exceed the threshold at the same time, the frequency and voltage are first reduced to reduce the burden, and the task scheduling strategy is adjusted according to the real-time load and temperature.

[0121] It should be noted that in order to deal with the situation of excessive load or excessive temperature of the core, a dynamic voltage and frequency adjustment strategy is adopted. This strategy monitors the load and temperature data of each core in real time, and calculates the priority of each core according to these data. When the load or temperature of a core is detected to exceed the set threshold, the system will reduce the power consumption and temperature by adjusting the frequency and voltage. The purpose of this is to protect the core from excessive load and excessive temperature, prolong the life of the core and avoid overheating.

[0122] S4: When the load and temperature exceed the set threshold, a self-learning mechanism is used to optimize the scheduling algorithm, automatically adjusting the scheduling strategy and temperature control scheme.

[0123] Further, the self-learning mechanism optimization scheduling algorithm includes that, in the self-learning mechanism, each state action pair (s t ,a t ) is initialized to 0, indicating that there is no prior knowledge at the initial time for all scheduling and temperature control schemes.

[0124] The heuristic rule output by the adaptive heuristic scheduling algorithm based on task scheduling initializes the scheduling strategy and automatically allocates tasks to cores with low load and low temperature.

[0125] Set learning parameters, and the system needs to perceive the current state s t at each time t, specifically including:

[0126] The core load L i (t) represents the load of the core i at time t; the core temperature T i (t) represents the temperature of the core i at time t; the state vector s t is given to the load and temperature information of all cores.

[0127] The state s t is taken as the input of the Q-learning algorithm, and a suitable action is selected through the reinforcement learning algorithm;

[0128] The priority of each core is calculated using the aforementioned evaluation function P i ; the state vector s t is calculated in real time through the evaluation function P i to calculate the task priority of each core.

[0129] According to the results of the evaluation function, emergency task scheduling is performed when the set threshold is exceeded.

[0130] Heuristic model initialization: heuristic rules based on task scheduling (such as the load and temperature evaluation function P i =w1·L i +w2·T i ) initialize the scheduling strategy. Through these heuristic rules, the initial scheduling strategy automatically allocates tasks to cores with low load and low temperature.

[0131] Set learning parameters: learning rate η, which determines the step size of Q value update; discount factor λ, which is used to balance immediate reward and future reward; exploration rate ε, which is given a higher exploration rate at the initial stage and gradually decreases later.

[0132] State definition and perception The system needs to perceive the current state s t at each time t, including:

[0133] The state vector s tContains all the core load and temperature information:

[0134] s t = [L1(t), T1(t), L2(t), T2(t), …, L n (t), T n (t)]

[0135] This state will be input to the Q-learning algorithm, which will choose the appropriate action.

[0136] Adaptive scheduling and temperature control based on real-time data requires dynamic adjustment of scheduling strategies and temperature control schemes based on the following conditions.

[0137] Load and temperature evaluation: Use the aforementioned evaluation function P i = w1·L i + w2·T i to calculate the priority of each core. Cores with low load and low temperature will be prioritized for task allocation. The real-time state vector s t will be used to calculate the task priority of each core through this evaluation function.

[0138] Task allocation: based on the results of the evaluation function, tasks are allocated to cores with lower load and temperature to avoid overloading or overheating some cores.

[0139] Frequency and voltage adjustment (DVFS): when the load or temperature of a core exceeds a certain threshold, the system automatically performs dynamic voltage frequency adjustment (DVFS):

[0140] Frequency adjustment: when the load or temperature is too high, the frequency of the core is automatically reduced to reduce the computational burden and reduce heat generation.

[0141] Voltage adjustment: when the frequency is reduced, the voltage will also be reduced to reduce power consumption.

[0142] The automatic adjustment of scheduling strategies and temperature control schemes includes adjusting the scheduling strategy based on real-time data feedback; at each time step, the algorithm selects an appropriate scheduling action based on the current load and temperature information.

[0143] Update Q values through reward signals, iterate until the optimal scheduling strategy and temperature control scheme are found; adjust the scheduling strategy based on Q-learning updates during the Q-learning training process; at each new time step, select the optimal action based on the current state to achieve load balancing and temperature control.

[0144] Self-learning mechanism optimizes scheduling algorithm When the load and temperature exceed the set threshold, the self-learning mechanism starts working, optimizing the scheduling strategy and temperature control scheme.

[0145] Online learning: Using the Q-learning algorithm in reinforcement learning, the scheduling strategy is adjusted according to real-time data feedback. At each time step, the algorithm selects an appropriate scheduling action (task allocation or frequency adjustment, etc.) based on the current load and temperature information.

[0146] Q-value update: The system updates the Q-value through the reward signal. The reward r t Calculated by load, temperature and power consumption, etc.:

[0147]

[0148] Where L i (t) is the load of core i, T i (t) is the temperature of core i, E(t) is the energy consumption of the system. Then, the Q-value is updated by the following formula:

[0149]

[0150] This process is repeated until the optimal scheduling strategy and temperature control scheme are found.

[0151] Strategy adjustment based on Q-learning update: During the training process of Q-learning, the system will automatically adjust the scheduling strategy according to the updated Q-value. At each new time step, the system will select the optimal action based on the current state to achieve load balancing and temperature control.

[0152] Automatic adjustment of task allocation: If the load or temperature of a core is consistently high, the system will automatically transfer tasks from that core to other cores with lower load or lower temperature.

[0153] Dynamic adjustment of temperature control strategy: When the temperature of a core exceeds the set threshold, the system will dynamically adjust the frequency and voltage to reduce power consumption and heat generation, thereby optimizing the temperature control strategy.

[0154] Optimization of self-learning mechanism: Over time, the system will optimize the scheduling algorithm based on historical data and real-time feedback, adjusting the frequency and voltage adjustment ratio, and automatically adapting to scheduling requirements under different loads and temperatures.

[0155] Termination condition and policy output, learning termination condition: When the change of Q-value is less than a certain threshold, or the maximum number of training iterations is reached, the learning process is terminated.

[0156] Final scheduling strategy: Once the learning is completed, the system will get an optimal scheduling strategy that can automatically make the best task scheduling and temperature control decisions at any time according to the current load and temperature.

[0157] It should be noted that in order to achieve more intelligent scheduling optimization, the system introduces a self-learning mechanism. Through the Q-learning algorithm, the system can automatically adjust the scheduling strategy based on real-time load and temperature data without prior knowledge. The initial value of each state-action pair is zero, and as the system runs, the Q-learning algorithm updates the Q value based on the feedback reward signal to find the optimal scheduling strategy. In this way, the system not only can cope with different load and temperature scenarios, but also can adaptively adjust the strategy to optimize task allocation and temperature control. During operation, the system can optimize load balancing and temperature control through continuous adjustment of scheduling strategy and temperature control scheme based on Q-learning training updates. When the load or temperature reaches the set threshold, the system will migrate the load to the core with lower temperature or lower load through task scheduling, and reduce the burden of the core by adjusting the frequency and voltage. As the training progresses, the system continuously improves the scheduling strategy and eventually reaches the optimal state of load balancing and temperature control.

[0158] The embodiment also provides a multi-core processor dynamic regulation system based on load and temperature perception, which comprises a sensor monitoring module that collects real-time load and temperature data of the core and pre-processes the data.

[0159] A scheduling algorithm module dynamically adjusts task allocation based on real-time load and temperature data to achieve load balancing and temperature control.

[0160] An adjustment module automatically adjusts the frequency and voltage of the core using a dynamic voltage frequency adjustment strategy based on real-time load and temperature state to achieve more efficient temperature control.

[0161] A self-learning optimization module automatically optimizes the scheduling strategy and temperature control scheme using real-time feedback and historical data of the system.

[0162] The embodiment also provides a computer device suitable for the multi-core processor dynamic regulation method based on load and temperature perception, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the multi-core processor dynamic regulation method based on load and temperature perception proposed in the above embodiment.

[0163] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0164] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for dynamically regulating a multi-core processor based on load and temperature sensing as described in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0165] To sum up, the present application achieves the following effects: real-time acquisition and preprocessing of load and temperature data, establishment of a thermal sensing model to predict temperature changes; construction of an adaptive heuristic scheduling algorithm to adjust task allocation according to real-time data of core load and temperature; adoption of a dynamic voltage frequency adjustment strategy to automatically adjust the frequency and voltage when the load and temperature are too high according to the load and temperature states; and adoption of a self-learning mechanism to optimize the scheduling algorithm and automatically adjust the scheduling strategy and temperature control scheme when the load and temperature exceed the set threshold.

[0166] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for dynamic control of a multi-core processor based on load and temperature sensing, characterized in that: This includes real-time acquisition and preprocessing of load and temperature data, and the establishment of thermal sensing models to predict temperature changes; Build an adaptive heuristic scheduling algorithm to adjust task allocation based on real-time data of core load and temperature; Based on load and temperature conditions, a dynamic voltage and frequency adjustment strategy is adopted to automatically adjust the frequency and voltage when the load is too high or the temperature is too high. When the load and temperature exceed the set thresholds, a self-learning mechanism is used to optimize the scheduling algorithm and automatically adjust the scheduling strategy and temperature control scheme.

2. The multi-core processor dynamic control method based on load and temperature sensing as described in claim 1, characterized in that: The preprocessed load and temperature data includes: real-time acquisition of load data for each core through the hardware interface of the monitoring tool; smoothing through weighted averaging; noise removal; and ensuring data stability. Temperature data for each core is collected in real time using temperature sensors, and the temperature data is smoothed using a weighted average method. Standardize the load and temperature data; Load data is standardized by the ratio to the system's maximum load, and temperature data is standardized by the ratio to the system's maximum temperature.

3. The multi-core processor dynamic control method based on load and temperature sensing as described in claim 2, characterized in that: The establishment of the thermal sensing model includes modeling the temperature change of each core using the heat conduction equation; the temperature change is composed of power consumption, core heat capacity, and heat conduction coefficient; and the core temperature change is calculated in each time step using the discretized equation. Based on historical load, power consumption and temperature data, coefficients are trained through regression analysis, and the temperature change of each core is predicted according to real-time power consumption and load. Set an upper temperature limit and a warning temperature threshold. When the temperature approaches or exceeds the warning value, trigger a self-learning mechanism to optimize the scheduling algorithm for scheduling. Combine real-time load and temperature data to dynamically adjust the core frequency, voltage and task allocation strategy.

4. The multi-core processor dynamic control method based on load and temperature sensing as described in claim 3, characterized in that: The construction of the adaptive heuristic scheduling algorithm includes a population consisting of multiple individuals, each representing a task scheduling scheme; There are n cores and m tasks. Each individual is represented by an m×n matrix, where the elements of the matrix indicate whether a task is assigned to a core. Construct a fitness function to evaluate the priority of each core based on the load balancing, temperature control effect, and frequency adjustment of the task scheduling scheme; the fitness of an individual is calculated based on the load and temperature of the scheduled task. An adaptive selection mechanism is introduced, where the selection probability of an individual is calculated by weighting the current fitness and the historical fitness, and a diversity protection method is used to assist in the adjustment; different crossover strategies are formulated according to the type of computationally intensive and I / O intensive tasks to carry out domain-specific crossover. Guided mutation is used to determine the mutation task based on the core tasks with low load and low temperature in the current population; local search mutation combined with local optimization algorithm further optimizes the quality of the solution by fine-tuning the individuals after guided mutation; and elite retention and partial replacement strategies are used to screen the mutated individuals.

5. The multi-core processor dynamic control method based on load and temperature sensing as described in claim 4, characterized in that: The dynamic voltage and frequency adjustment strategy includes real-time monitoring of the load and temperature of each core, and data collection using a sensor monitoring module; load data L i and temperature data T i These represent the real-time load and temperature of the i-th core, respectively. Based on real-time collected load and temperature data, the priority of each core is calculated using the evaluation function P. i This indicates the task priority of each core, and an optimized scheduling strategy is output using an adaptive heuristic scheduling algorithm. The scheduling rules specifically include: Prioritize cores with lower load and temperature predicted by the thermal sensing model to avoid cores with excessive load and temperature; when the load and temperature prediction of a core exceed the set threshold, automatically migrate tasks to cores with lower load and temperature. When core load L is detected i Exceeding the set threshold L max At that time, the core frequency f is automatically reduced. i When temperature T i Exceeding the set threshold T max At that time, the core frequency f will be reduced. i and voltage V i When the temperature rises, the core frequency f is automatically adjusted. i Dynamic control is implemented; when the core temperature T i When the frequency f increases, i The voltage V will be reduced proportionally. i The frequency is adjusted accordingly; when both load and temperature exceed the threshold, the frequency and voltage are reduced first to alleviate the burden, and the task scheduling strategy is adjusted according to the real-time load and temperature.

6. The multi-core processor dynamic control method based on load and temperature sensing as described in claim 5, characterized in that: The self-learning mechanism-based optimization scheduling algorithm includes, in the self-learning mechanism, each state-action pair (s) t ,a t The initial value is 0, indicating that there is no prior knowledge of any scheduling and temperature control schemes at the beginning; The heuristic rules output by the adaptive heuristic scheduling algorithm based on task scheduling are used to initialize the scheduling strategy and automatically allocate tasks to cores with low load and low temperature. Set the learning parameters; at each time step t, the system needs to perceive the current state s. t Specifically, it includes: Core load L i (t) represents the load on core i at time t; core temperature T i (t) represents the temperature of core i at time t; is assigned to the state vector s. t All core load and temperature information; State s t As input to the Q-learning algorithm, the appropriate action is selected through the reinforcement learning algorithm; Using the aforementioned evaluation function P i To calculate the priority of each core; state vector s t Real-time evaluation function P i To calculate the task priority for each core; Based on the results of the evaluation function, emergency task scheduling is executed when the threshold is exceeded.

7. The multi-core processor dynamic control method based on load and temperature sensing as described in claim 6, characterized in that: The automatic adjustment scheduling strategy and temperature control scheme include adjusting the scheduling strategy based on real-time data feedback; at each time step, the algorithm selects a suitable scheduling action based on the current load and temperature information. The Q value is updated by reward signals, and the process is iterated until the optimal scheduling strategy and temperature control scheme are found. Based on Q-learning update strategy adjustment, the scheduling strategy is automatically adjusted according to the updated Q value during the Q-learning training process; at each new time step, the optimal action is selected according to the current state to achieve load balancing and temperature control.

8. A multi-core processor dynamic control system based on load and temperature sensing, based on the multi-core processor dynamic control method based on load and temperature sensing as described in any one of claims 1 to 7, characterized in that: This includes a sensor monitoring module that collects core load and temperature data in real time and preprocesses this data. The scheduling algorithm module dynamically adjusts task allocation based on real-time load and temperature data to achieve load balancing and temperature control. The adjustment module uses a dynamic voltage and frequency adjustment strategy to automatically adjust the core frequency and voltage based on real-time load and temperature conditions, in order to achieve more efficient temperature control. The self-learning optimization module automatically optimizes scheduling strategies and temperature control schemes using real-time feedback and historical data from the system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the multi-core processor dynamic control method based on load and temperature sensing as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the multi-core processor dynamic control method based on load and temperature sensing as described in any one of claims 1 to 7.

Images