Management method and management system for power supply system of wearable intelligent equipment

Abstract

The invention provides a management method and a management system for a power supply system of wearable intelligent equipment, and relates to the technical field of wearable intelligent equipment, and the method comprises the steps: building a task weight model, obtaining a task reliability score, building a task value model, judging the emergency degree of task information, building a four-quadrant classifier, and building a component-task incidence matrix. Dividing a plurality of levels of power supply pools, and distributing power supply pool weights according to the task reliability scores; and establishing a task-component joint optimization model, adjusting parameters of the task weight model, and outputting optimal sensitive parameters. According to the invention, through a multi-attribute weight distribution mechanism based on an entropy weight method, the capacity distribution of the core power supply pool, the dynamic power supply pool and the standby power supply pool is dynamically adjusted, the energy efficiency of the core power supply is optimized through a voltage climbing strategy, and the endurance time of equipment is prolonged; a dynamic load balancing protocol is triggered for a low-value and low-emergency task and is migrated to a low-power-consumption component to be executed, resource occupation is reduced, and invalid energy consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of wearable intelligent devices, and particularly to a management method and a management system for a power supply system of a wearable intelligent device. Background Art

[0002] Wearable intelligent devices are a type of highly integrated electronic devices that can be worn on the human body. Their design concepts emphasize portability, real-time interaction, and environmental perception capabilities. According to morphological and functional differences, they can be divided into body surface devices (such as smart watches and health bracelets), implantable devices (such as medical monitoring sensors), and augmented reality devices (such as AR glasses), etc. Integrated with biosensors (such as PPG heart rate sensors and EEG brain electrical monitoring), environmental sensors (such as temperature and humidity, barometers), and inertial measurement units (IMUs), real-time data processing is achieved through edge computing to support functions such as motion recognition and health warning. Dependent on miniaturized batteries (such as lithium polymer batteries) or hybrid energy supply systems (such as solar energy and thermoelectric conversion), it is necessary to balance battery life and performance under limited capacity. For example, a smart watch needs to maintain a battery life of several days under micro-watt-level power consumption, while AR devices face higher energy consumption challenges due to display and graphics rendering requirements. Natural interaction is achieved through flexible screens, tactile feedback, or voice interaction, and at the same time, it is necessary to adapt to the reliability and comfort in complex wearing scenarios (such as movement vibration and sweat erosion).

[0003] The power management system (PMS) of a wearable intelligent device is responsible for coordinating the energy distribution of multiple components (such as processors, sensors, and communication chips) within the device to optimize battery life and ensure the reliability of critical tasks. It covers battery management (such as charge and discharge cycle control), energy harvesting modules (such as photovoltaic, thermoelectric, and vibration energy conversion), and wireless charging technologies, and needs to adapt to energy input fluctuations in different scenarios.

[0004] Generally, the management method of the power supply system of a wearable intelligent device adopts first-come-first-served or simple preemptive scheduling, without adjusting the task execution strategy in combination with the real-time status of components. Moreover, traditional methods mostly rely on static parameter models and cannot dynamically adjust task weights or power distribution strategies through real-time feedback. The power pool weight distribution is long-term fixed, making it difficult to cope with sudden power fluctuations or task load changes, and the system robustness is insufficient. Summary of the Invention

[0005] The present invention provides a management method and a management system for a power supply system of a wearable intelligent device to solve the defects in the prior art.

[0006] On the one hand, the present invention provides a management method for a power supply system of a wearable intelligent device, including: Collecting task information of each load node and outputting task node attribute information; Combine historical task data to establish a task weight model based on the Sigmoid function and Weibull distribution; and input the task node attribute information into the task weight model to output the task reliability score; Construct a task value model based on the task reliability score to determine the urgency of task information; and construct a quadrant classifier based on the urgency of task information to output four types of task queues; Based on the four types of task queues, establish a component-task association matrix; extend the colored Petri net model, add task reliability constraints, and output component status labels and component reliability scores; According to the component reliability scores, divide multiple levels of power pools, and allocate weights to the power pools according to the task reliability scores, and output a power pool configuration table with task weight constraints; Perform reliability-aware scheduling based on the four types of task queues and the power pool configuration table, and output task scheduling instructions; Based on the task scheduling instructions and component status labels, establish a task-component joint optimization model to adjust the parameters of the task weight model, and output the optimal sensitive parameters; Adjust the power pool configuration table according to the optimal sensitive parameters, and output the power pool adjustment strategy.

[0007] According to a management method for a power system of a wearable intelligent device provided by the present invention, the steps of establishing a task weight model include: Extract historical task features from historical task data and use the historical task features as a training set; According to the historical task features, calculate the historical success rate and average delay deviation of the tasks, and combine with the Weibull distribution parameters to output a task failure probability distribution table; Learn the parameters of the Sigmoid function by the gradient descent method; and estimate the Weibull distribution parameters by MLE to obtain the task weight model.

[0008] According to a management method for a power system of a wearable intelligent device provided by the present invention, the steps of extending the colored Petri net model include: Collect the list of hardware components of the device; Create an independent place for each hardware component in the list of hardware components and mark it using color marking rules to output a component operation status table; Formulate state transition rules according to the task requests in the four types of task queues, the component operation status table, and the power consumption monitoring data; Configure arc connection rules according to the component operation status table and state transition rules, and output the colored Petri net model.

[0009] According to a management method for a power system of a wearable intelligent device provided by the present invention, the steps of constructing a task value model include: Filter valid tasks from the task reliability score, extract the P_score column as the feature vector, and output the reliability feature vector; Obtain task delay tolerance from task node attribute information, calculate the urgency coefficient, and output the urgency coefficient vector; According to the attribute information of the task node, the average energy consumption of all tasks is calculated, and the energy consumption weight is calculated based on the average energy consumption, and the energy consumption weight vector is output; The reliability feature vector, the emergency coefficient vector and the energy consumption weight vector are combined into a three-dimensional feature matrix by column, and the task value feature matrix is output; According to the task value feature matrix, the random forest regression model is used to predict the task value.

[0010] According to a management method for a wearable smart device power system provided by the present invention, the steps of constructing a four-quadrant classifier include: Divide the task information into four task urgency quadrants according to the urgency of the task information; According to the task value model, the task urgency quadrant is divided into two dimensions; The compensation mechanism is triggered according to the result of the two-dimensional partitioning, and its resource usage is reduced through dynamic voltage and frequency adjustment.

[0011] According to a wearable smart device power system management method provided by the present invention, the step of allocating power pool weights includes: Based on the task reliability score and task value, the entropy weight method is used to calculate the power pool weight allocation coefficient; The power pool includes the core power pool, dynamic power pool and backup power pool; a minimum weight threshold is set for the core power pool, the weight of the dynamic power pool is allocated according to the power pool weight allocation coefficient, the remaining capacity is allocated to the backup power pool, and a power pool configuration table is output.

[0012] According to a management method for a wearable smart device power system provided by the present invention, the steps of performing reliability-aware scheduling include: Real-time monitoring of component status tags and power bank balance; The task urgency quadrants include: the first quadrant is for high-value and high-urgency tasks, and the third quadrant is for low-value and high-urgency tasks. Preemptive scheduling is used for high-value and high-urgency tasks, and they are bound to the core power pool for power supply. Enable flexible scheduling for low-value, high-urgency tasks, allow downgraded component execution, and output scheduling decisions.

[0013] According to a wearable smart device power system management method provided by the present invention, the step of adjusting the parameters of the task weight model includes: According to the current power pool weights and component status tags, multi-objectives are transformed into single-objective optimization through weighted summation to generate a Pareto front solution set; According to the Pareto front solution set, the parameters of the task weight model are adjusted item by item, the change range of the model parameters is observed to generate sensitive parameters; the sensitive parameters are adjusted according to the feedback of real-time performance indicators, and the optimal sensitive parameters are generated through incremental optimization.

[0014] According to a management method for a power supply system of a wearable intelligent device provided by the present invention, the power pool adjustment strategy includes: Based on the power pool weight distribution coefficient, the power pool capacity is redistributed, a voltage climbing strategy is executed on the core power pool, and a core power pool capacity adjustment instruction is output; the task urgency quadrant further includes: low-value and low-urgency tasks. If the backup power pool is enabled, a dynamic load balancing protocol is triggered to migrate the low-value and low-urgency tasks to low-power components for execution.

[0015] The present invention also provides a management system for a power supply system of a wearable intelligent device, including: a task information acquisition module that collects task information of each load node of the device in real time and outputs task node attribute information; A task reliability scoring module, which is used to output a task reliability score by establishing a task weight model based on the Sigmoid function and the Weibull distribution based on historical data and task node attribute information; A task value degree evaluation module, which is used to construct a task value degree model according to the task reliability score to determine the urgency of task information; A task classification module, which is used to construct a four-quadrant classifier according to the task information urgency and output four types of task queues; A component status monitoring module, which is used to output component status tags and component reliability scores through a colored Petri net model according to the four types of task queues and power consumption data; A power pool management module, which is used to divide multiple levels of power pools according to the component reliability scores and component status tags, and allocate power pool weights according to the task reliability scores, and output a power pool configuration table; A scheduling engine, which is used to formulate a task scheduling strategy according to the four types of task queues in combination with the power pool configuration table and output a task scheduling instruction; A parameter optimization module, which is used to adjust the parameters of the task weight model according to the task scheduling instruction and the component status tag and output the optimal sensitive parameters; A system policy adjustment module, which is used to adjust the power pool configuration table according to the optimal sensitive parameters and output a power pool adjustment strategy.

[0016] The present invention provides a management method and management system for a power supply system of a wearable intelligent device. By combining the Sigmoid function with the Weibull distribution to construct a task weight model, and integrating parameters such as historical success rate and delay deviation, a quantitative evaluation of task reliability is achieved. Tasks are dynamically divided into high / low priority queues according to urgency and value through a four-quadrant classifier, and targeted preemptive or flexible scheduling strategies are adopted to ensure that high-value tasks are given priority to be powered by the core power pool, and low-value tasks reduce energy consumption by downgrading or migrating components, thereby significantly improving system response efficiency and user experience. Through a multi-attribute weight allocation mechanism based on the entropy weight method, combined with real-time power monitoring and component status labels, the capacity allocation of the core, dynamic and backup power pools is dynamically adjusted, and the core power energy efficiency is optimized through a voltage climbing strategy to extend the device's battery life. The coordinated scheduling of the backup power pool and low-power components effectively avoids resource waste and enhances the system's fault tolerance. A color Petri net extended model tracks component status in real time, combined with random forest regression to predict task value, forming a closed "monitoring-decision-making-optimization" loop. Pareto front solutions and incremental optimization are used to adjust task weight model parameters, dynamically balancing energy consumption, latency, and reliability constraints, improving system robustness in complex scenarios. Low-value, low-urgency tasks are subject to dynamic load balancing protocols, migrating execution to low-power components. Flexible scheduling reduces resource utilization and inefficient energy consumption. The state transition rules and arc connection configuration of the color Petri net enable efficient linkage between hardware components and tasks, optimizing overall energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a step diagram of a method for managing a power system of a wearable smart device provided by an embodiment of the present invention; Figure 2 The present invention provides a schematic diagram of a management system for a power supply system of a wearable smart device. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0020] Embodiment 1: The following combines Figure 1 - Figure 2 to describe a management method and a management system for a power supply system of a wearable intelligent device of the present invention.

[0021] As Figure 1 shown, a management method for a power supply system of a wearable intelligent device provided by an embodiment of the present invention includes: Collect the task information of each load node, including task type, execution period, data volume, priority, and hardware component dependency relationship, and output a task node attribute information table containing dynamic parameters. Task information refers to various characteristics and attributes related to each task in the wearable intelligent device, and this information can comprehensively describe the running status and requirements of the task.

[0022] Combined with historical task data, establish a task weight model based on the Sigmoid function and the Weibull distribution. And input the task node attribute information into the task weight model to output a task reliability score. The Sigmoid function is a common S-shaped function, and its output value is between 0 and 1, with good differentiability and non-linear characteristics. Here, the Sigmoid function is used to model the weight of the task, which can map the characteristics of the task to a suitable weight interval, thereby reflecting the importance and reliability of the task. The Weibull distribution is a continuous probability distribution, which is widely used in reliability engineering and life data analysis to describe the failure probability distribution of things.

[0023] The steps of establishing the task weight model include: Extract 12-dimensional historical task features such as task type, execution success rate, delay time, and resource occupancy rate from the historical task data, and construct a training set containing time series features. The 12-dimensional historical task features refer to 12 feature dimensions related to the task extracted from the historical task data, and these feature dimensions can comprehensively reflect the running status and performance of the task. For example, the execution success rate represents the ratio of the number of successful task executions to the total number of executions, the delay time is the difference between the actual execution time and the expected execution time of the task, and the resource occupancy rate reflects the proportion of system resources occupied by the task during execution.

[0024] Calculate the historical success rate and average delay deviation of tasks based on historical task characteristics. Then, correct the historical data deviation based on exponential smoothing, and the correction method is expressed as: ; ; In the formula, represents the corrected average delay deviation, represents the delay deviation of the i-th task, and α is the smoothing factor, which can be adaptively adjusted according to the length M of the historical data.

[0025] Exponential smoothing is a time series prediction method. By assigning decreasing weights to historical data, it can smooth the random fluctuations in the data, thus more accurately reflecting the trend and pattern of the data. Combining with the Weibull distribution parameters, output the task failure probability distribution table.

[0026] Iteratively update the parameters of the Sigmoid function through the Adam optimizer with an adaptive learning rate to minimize the prediction error cross-entropy loss. The Adam optimizer is an optimization algorithm with an adaptive learning rate, which combines the advantages of the momentum optimization and RMSprop optimization algorithms. It can adaptively adjust the learning rate according to the gradient of the parameters, thus accelerating the convergence speed of the model and improving the optimization effect of the model. Use the maximum likelihood estimation (MLE) to solve the Weibull distribution parameters, and constrain the model complexity through the KL divergence to obtain a task weight model with enhanced robustness. The KL divergence is an index to measure the difference between two probability distributions, which is used to constrain the model complexity and prevent the model from overfitting. The task weight calculation is expressed as: ; ; In the formula, w task represents the task weight, and w task ∈(0,1). z represents the linear combination of task characteristics, where x j is the 12-dimensional historical task characteristics (such as execution success rate, delay time, resource occupancy rate, etc.), w j is the weight dynamically adjusted by the Adam optimizer, and b is the bias term.

[0027] The method for estimating the Weibull distribution parameters is expressed as: ; In the formula, and are the estimated Weibull distribution parameters output, x i represents the historical execution time of the i-th task, λ is the scale parameter, representing the average failure time of the task. k is the shape parameter, reflecting the failure mode (k < 1 is a decreasing failure rate, k = 1 is a constant failure rate, k > 1 is an increasing failure rate). Indicates the task failure probability.

[0028] Construct a task value model based on the task reliability score to determine the urgency of task information. The steps for constructing the task value model include: Select valid tasks from the task reliability score, extract the P_score column as the feature vector, and output the reliability feature vector. Specifically, the isolation forest algorithm is used to eliminate abnormal tasks, and the P_score column (a normalized index combining success rate and failure probability) is extracted as the core feature vector. The P_score column refers to a key index column in the task reliability score, which combines the success rate and failure probability of the task and is normalized to a value between 0 and 1, intuitively reflecting the reliability of the task.

[0029] Obtain the task delay tolerance from the task node attribute information, calculate the emergency coefficient, and output the emergency coefficient vector. The task delay tolerance refers to the maximum delay time that the task can tolerate. Different tasks have different tolerances for delays. Tasks with high real-time requirements have a lower tolerance for delays, while some non-real-time tasks have a relatively higher tolerance for delays. The emergency coefficient is an index calculated based on the task delay tolerance, used to measure the urgency of the task. The higher the emergency coefficient, the more urgent the task. The calculation method of the emergency coefficient is expressed as: ; In the formula, C ur represents the emergency coefficient, and C ur ∈[0,1], and the larger the value, the more urgent the task. τ max represents the maximum delay time allowed for the task (defined by the task type, such as for real-time monitoring tasks τ max = 100ms). τ rem represents the remaining available time (the longest time the task can be delayed under the current battery level).

[0030] According to the task node attribute information, calculate the average energy consumption of all tasks, and dynamically adjust the energy consumption weight based on the remaining battery power to prevent energy consumption imbalance under low battery levels. The energy consumption weight refers to the weight assigned according to the energy consumption of the task, used to balance the energy consumption requirements of the task during resource allocation. By calculating the average energy consumption of all tasks and dynamically adjusting the energy consumption weight according to the remaining battery power, the problem of energy consumption imbalance under low battery levels can be effectively prevented, thereby extending the battery life of the wearable intelligent device.

[0031] Merge the reliability feature vector, the emergency coefficient vector, and the energy consumption weight vector into a three-dimensional feature matrix by column, and output the task value feature matrix, which can comprehensively reflect the reliability, urgency, and energy consumption requirements of the task, providing basic data for the construction of the task value model.

[0032] According to the task value feature matrix, use the random forest regression model to output the task value in the range of 0-1, and calibrate the scoring scale through Mahalanobis distance. Mahalanobis distance is a metric for measuring the distance between data points, which takes into account the covariance matrix of the data and can more accurately reflect the similarity and difference between data points. The method of calibrating the scoring scale through Mahalanobis distance is expressed as: ; ; In the formula, D is the Mahalanobis distance value, v represents the task feature vector, including reliability score, emergency coefficient, and energy consumption weight. μ represents the mean vector of the feature vector, ∑ is the covariance matrix, and T is the transpose operation of the matrix. V value represents the normalized task value, which is mapped to the interval [0,1]. max(D) is the maximum value of the Mahalanobis distance value.

[0033] Construct a Cartesian coordinate system based on task urgency and value, divide four task urgency quadrants, and introduce a compensation mechanism to optimize resource allocation and output four types of task queues.

[0034] The steps to construct a quadrant classifier include: Divide the task information into four task urgency quadrants according to the task information urgency. Define the first quadrant as high-value and high-urgency tasks (H / H), the second quadrant as high-value and low-urgency tasks (H / L), the third quadrant as low-value and high-urgency tasks (L / H), and the fourth quadrant as low-value and low-urgency tasks (L / L). High-value and high-urgency tasks (H / H) refer to tasks with relatively high task value and urgency. Such tasks are usually crucial for the operation of the system and require priority resource allocation and execution as soon as possible. High-value and low-urgency tasks (H / L) refer to tasks with high task value but low urgency. Although such tasks are important, they can be executed appropriately later. Low-value and high-urgency tasks (L / H) refer to tasks with low task value but high urgency. Although such tasks are urgent, they contribute less to the overall value of the system. Low-value and low-urgency tasks (L / L) refer to tasks with relatively low task value and urgency. Such tasks can usually be postponed or executed when resources permit.

[0035] Based on the output value of the task value model and the real-time urgency threshold (Eth), a two-dimensional division is carried out, specifically including: E≥Eth and V≥Vth → H / H class; E<Eth and V≥Vth → H / L class; E≥Eth and V<Vth → L / H class; E<Eth and V<Vth → L / L class. Trigger a compensation mechanism for L / H class tasks, and reduce their resource occupancy through dynamic voltage and frequency scaling (DVFS). Dynamic voltage and frequency scaling (DVFS) reduces the energy consumption of the system by dynamically adjusting the voltage and frequency of the processor. For low-value high-urgency tasks (L / H), DVFS is used to reduce their resource occupancy, thereby releasing resources for other more important tasks and improving the overall performance and resource utilization rate of the system.

[0036] According to the four types of task queues, a component-task association matrix is established. By establishing a component-task association matrix, the requirements of each task for hardware components can be clearly understood. Expand the colored Petri net model, add task reliability constraints, and output component status labels and component reliability scores. The Petri net model is an extended Petri net model that can more flexibly describe the dynamic behavior and complex relationships of the system by introducing color markings to represent different types of elements or states. By expanding the colored Petri net model and adding task reliability constraints, the interaction process between tasks and hardware components in wearable smart devices can be more accurately simulated and analyzed, while considering the impact of task reliability on system operation. The steps for expanding the colored Petri net model include: Collect the hardware component list of the device. The hardware component list is the set of all hardware components in the wearable smart device, including processors, sensors, memories, communication modules, etc.

[0037] Create an independent place for each hardware component in the hardware component list and mark it using color marking rules, and output the component operation status table. A place is a basic element in the Petri net model, used to represent resources or states in the system. The marking color set includes: Running state (green): Normal power supply and utilization rate > 70%, indicating that the hardware component is in a normal working state and its resource utilization rate is relatively high; Energy-saving state (yellow): The supply voltage is reduced to 0.9V, indicating that the hardware component is in an energy-saving mode, reducing energy consumption by reducing the supply voltage; Fault state (red): Continuous error rate > 5%, indicating that the hardware component has a fault and a high error rate, and needs to be repaired or switched to a standby component; Output the component operation status table containing timestamps, recording the last 3 state transition events.

[0038] Calculate the component reliability through the Weibull distribution, dynamically update the parameters: battery attenuation rate and usage intensity coefficient, and monitor the state transitions in real time (e.g., green → pink when the task is triggered, pink → red when the battery runs out). Component reliability refers to the probability that a hardware component works properly within a certain period of time. For example, when the task is triggered, it changes from the running state (green) to the standby state (pink), and when the battery runs out, it changes from the standby state (pink) to the failure state (red).

[0039] Based on the task requests in the four types of task queues, the component operation status table, and the power monitoring data, formulate state transition rules. The state transition rules include: When a task request arrives and the component is in the energy-saving state, trigger the warm-up protocol (e.g., increase the voltage to 1.1V 10ms in advance), and update the arc connection weights. If the component fails continuously three times, automatically switch to the standby component and trigger the fault propagation blocking protocol. The warm-up protocol is a mechanism used to improve the performance of hardware components. By increasing the voltage in advance, etc., the hardware component can quickly switch from the energy-saving state to the normal working state. The arc connection weight is an important parameter in the Petri net model, indicating the connection strength or transition probability between places. By updating the arc connection weights, the dynamic behavior of the system can be adjusted.

[0040] Add a task reliability threshold in the Petri net transition. For example: Only when the reliability score of the core component is greater than or equal to 0.9, is it allowed to execute critical tasks. The task reliability threshold is a mechanism used to ensure the reliability of task execution. By setting the reliability threshold, it can be ensured that only when the reliability of the hardware component meets the requirements, is it allowed to execute critical tasks, thereby improving the overall reliability of the system.

[0041] Finally, according to the component operation status table and the state transition rules, configure the arc connection rules, and output the colored Petri net model for simulating and analyzing the dynamic behavior and resource management process of the wearable intelligent device.

[0042] According to the component reliability score, divide multiple levels of power pools, and allocate power pool weights according to the task reliability score, and output the power pool configuration table with task weight constraints. The steps for allocating power pool weights include: Based on the task reliability score and task value, use the entropy weight method to calculate the power pool weight distribution coefficient. The entropy weight method is a multi-attribute decision-making method based on information entropy. By calculating the information entropy of each attribute, the weight of each attribute is determined. The specific method is expressed as: ; ; ; Where \(E_a\) is the information entropy under the \(a\)-th index (reliability, value), and the larger the value, the lower the index discrimination. \(r\) ia represents the score of the \(i\)-th task under the \(a\)-th index (reliability, value). \(p\) ia represents the normalized proportion of index \(a\), \(w\) j represents the weight assignment value of the \(j\)-th power supply pool, \(k\) represents the index of the power supply pool, \(E\) j represents the information entropy of the \(j\)-th power supply pool, \(E\) k represents the information entropy of the \(k\)-th power supply pool, reflecting the discrimination of the corresponding index of this power supply pool (the larger the entropy, the lower the discrimination), and the information entropy formula can also be used for calculation.

[0043] The power supply pool includes a core power supply pool, a dynamic power supply pool, and a standby power supply pool. A fixed 30% capacity is allocated to the core power supply pool, and a guaranteed threshold (such as the voltage is not lower than 0.85V) is set, which is dedicated to executing tasks in the first quadrant. The weight of the dynamic power supply pool is allocated with the remaining 65% capacity according to the power supply pool weight distribution coefficient, supporting dynamic voltage climbing (5% increase each time, with an interval of 200ms), and the remaining 5% capacity is assigned to the standby power supply pool, which is only enabled in the low battery mode and gives priority to ensuring safety components such as vital sign monitoring. Finally, a power supply pool configuration table is output by combining the three power supply pools.

[0044] Perform reliability-aware scheduling according to the four types of task queues and the power supply pool configuration table, and output task scheduling instructions. The steps for performing reliability-aware scheduling include: Real-time monitor the component status tags and the remaining capacity of the power supply pool.

[0045] For high-value and high-urgency tasks, use preemptive scheduling, bind the core power supply pool for power supply, and increase the CPU frequency, while freezing the tasks in the third quadrant. Preemptive scheduling is a priority-driven scheduling strategy, where high-priority tasks can preempt the resources of low-priority tasks, which can avoid resource conflicts and improve the overall performance of the system.

[0046] For low-value and high-urgency tasks, enable elastic scheduling, allow degraded components to execute, turn off the image enhancement function, and only retain the basic data processing function. Finally, combine high-value and high-urgency tasks and low-value and high-urgency tasks, and output a scheduling decision. Enabling elastic scheduling for low-value and high-urgency tasks, allowing degraded components to execute, turning off the image enhancement function, and only retaining the basic data processing function can reduce resource consumption and improve the resource utilization rate of the system while ensuring the urgency of tasks.

[0047] According to the task scheduling instructions and the component status tags, establish a task-component joint optimization model, adjust the parameters of the task weight model, and output the optimal sensitive parameters. The steps for adjusting the parameters of the task weight model include: ​According to the current power pool weights and component status tags, multi-objectives are transformed into a single-objective optimization through weighted summation to generate a Pareto front solution set. A multi-objective function (delay, energy consumption, reliability loss) is constructed and transformed into a single objective through weighting.

[0048] The NSGA-II algorithm is used to generate the Pareto front solution set, and the optimal solution that satisfies the voltage constraint is selected as the scheduling instruction. The Pareto front solution set refers to the set of solutions in multi-objective optimization where it is impossible to improve one objective without deteriorating other objectives. The way to generate the Pareto front solution set using the NSGA-II algorithm is expressed as: ; ; In the formula, f1, f2, f3 are multi-objective functions (delay, energy consumption, reliability loss). x * is the Pareto optimal solution set, that is, the optimal combination of adjusting task weight parameters.

[0049] According to the Pareto front solution set, the parameters of the task weight model are adjusted item by item, the change range of the model parameters is observed, and sensitive parameters are generated. The sensitive parameters are adjusted according to the feedback of real-time performance indicators, and the optimal sensitive parameters are generated through incremental optimization. The sensitive parameters are screened by the Morris method (such as task switching overhead, DVFS response delay), the parameters are sorted according to sensitivity, and the parameters with an optimization influence degree > 0.3 are adjusted. The calculation method of sensitive parameters is expressed as: ; In the formula, Δ l is the sensitive parameter, and the larger the value, the more significant the influence of the parameter on the objective function. q is the number of perturbation times, usually q = 10. δ l is the parameter perturbation step size, and x l is the parameter of the l-th task weight model.

[0050] According to the optimal sensitive parameters, the power pool configuration table is adjusted, and the power pool adjustment strategy is output. The power pool adjustment strategy includes: Reallocate the power pool capacity based on the power pool weight distribution coefficient, execute the voltage ramp-up strategy for the core power pool, and output the core power pool capacity adjustment instruction. When it is detected that the load of the core power pool > 85%, trigger the capacity reallocation protocol, temporarily allocate 10% of the capacity of the dynamic power pool to the core pool, and reduce the sampling rate of non-critical components (such as from 100 Hz to 50 Hz). Dynamically adjust the voltage of the core power pool (DVFS), with the maximum ramp-up slope limited to 0.1 V / ms to prevent circuit overshoot. DVFS can optimize performance and power consumption under different workloads by dynamically adjusting voltage and frequency. For example, when the device is in a low-load state, the voltage and frequency can be reduced to reduce power consumption; while in a high-load state, the voltage and frequency can be increased to improve performance.

[0051] If the backup power pool is enabled, trigger the dynamic load balancing protocol to migrate low-value and low-urgency tasks to low-power components for execution, thus achieving the dynamic load balancing of the system. That is, when a component enters the fault state, the automatic backup power pool is enabled, and the state is restored through the differential synchronization protocol (only transmitting changed data), with the restoration time < 50 ms. The dynamic load balancing protocol is a mechanism for balancing the system load. By migrating tasks from high-load components to low-load components, the overall performance and resource utilization rate of the system can be improved. The differential synchronization protocol is a data synchronization mechanism that only transmits the changed parts of the data instead of the entire data, thus reducing the data transmission volume and synchronization time.

[0052] In summary, the management method of the power supply system of the wearable intelligent device provided by the present invention constructs a task weight model by combining the Sigmoid function and the Weibull distribution, integrates parameters such as historical success rate and delay deviation, and realizes the quantitative evaluation of task reliability; the tasks are dynamically divided into high / low priority queues according to urgency and value through a quadrant classifier, and preemptive or elastic scheduling strategies are adopted accordingly to ensure that high-value tasks are preferentially powered by the core power supply pool, and low-value tasks reduce energy consumption through component degradation or migration, significantly improving the system response efficiency and user experience. Through the multi-attribute weight allocation mechanism based on the entropy weight method, combined with real-time power monitoring and component status tags, the capacity allocation of the core, dynamic and standby power supply pools is dynamically adjusted, and the core power efficiency is optimized through the voltage climbing strategy to extend the device battery life. The collaborative scheduling of the standby power supply pool and the low-power components effectively avoids resource waste and enhances the system fault tolerance. The component status is tracked in real time through the extended model of the colored Petri net, and the task value is predicted by combining the random forest regression to form a "monitoring - decision - optimization" closed loop; the task weight model parameters are dynamically balanced by the Pareto front solution set and incremental optimization to balance the constraints of energy consumption, delay and reliability, and improve the robustness of the system in complex scenarios. By triggering the dynamic load balancing protocol for low-value and low-urgency tasks and migrating them to low-power components for execution, combined with elastic scheduling to reduce resource occupancy and reduce ineffective energy consumption. Through the state transition rules and arc connection configurations of the colored Petri net, the efficient linkage between hardware components and tasks is realized, and the overall energy efficiency ratio is optimized.

[0053] As Figure 2 shown, the present invention also provides a management system for the power supply system of a wearable intelligent device, including: A task information acquisition module, configured to collect task information of each load node of the device in real time and output task node attribute information, including task identifiers, timestamps, geographical location tags, and associated component statuses. The task information includes task types, priorities, resource requirements, execution cycles, energy consumption characteristics, and environmental parameters, and the dynamic load status is obtained through multi-sensor fusion technology (such as accelerometers, gyroscopes, heart rate sensors).

[0054] A task reliability scoring module, configured to output a task reliability score based on historical data and task node attribute information by establishing a task weight model based on the Sigmoid function and the Weibull distribution. The Sigmoid function is used to map task characteristics to a credibility probability in the 0-1 interval, and the Weibull distribution models the randomness of task completion time, and the model parameters are optimized by combining the maximum likelihood estimation method. A sliding time window mechanism can be introduced to update the model periodically to adapt to load changes, and the scoring accuracy is verified through Monte Carlo simulation.

[0055] The task value evaluation module is used to construct a task value model based on the task reliability score to determine the urgency of task information. A multi-dimensional task value model is constructed according to the task reliability score, integrating task urgency (deadline sensitivity), energy consumption cost (energy consumption per unit task), and system impact factor (number of dependent components), and calculating the comprehensive task value score. The dynamic weight adjustment algorithm is adopted to adjust the model parameters in real time according to the user behavior pattern (such as motion state, health monitoring requirements), and output the task urgency classification label (such as urgent / high / medium / low).

[0056] The task classification module is used to construct a four-quadrant classifier according to the task information urgency and output four types of task queues.

[0057] The component status monitoring module is used to output the component status label and the component reliability score through the colored Petri net model according to the four types of task queues and the power data. The Markov chain prediction algorithm can also be introduced to evaluate the component reliability score, and combined with the fault tree analysis (FTA) to generate a risk warning signal to support preventive power distribution.

[0058] The power pool management module is used to divide multiple levels of power pools according to the component reliability score and the component status label, and assign power pool weights according to the task reliability score, and output the power pool configuration table.

[0059] The scheduling engine is used to formulate a task scheduling strategy according to the four types of task queues, combined with the power pool configuration table, and output the task scheduling instruction.

[0060] The parameter optimization module is used to adjust the parameters of the task weight model according to the task scheduling instruction and the component status label, and output the optimal sensitive parameters.

[0061] The system policy adjustment module is used to adjust the power pool configuration table according to the optimal sensitive parameters, dynamically adjust the power pool classification threshold, weight distribution coefficient and redundancy and fault tolerance strategy, and output the power pool adjustment strategy.

[0062] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0063] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A management method for a power supply system of a wearable intelligent device, characterized in that, Including: Collect the task information of each load node and output the task node attribute information; Combine the historical task data to establish a task weight model based on the Sigmoid function and the Weibull distribution; and input the task node attribute information into the task weight model to output the task reliability score; Construct a task value model according to the task reliability score to determine the urgency of the task information; and construct a four-quadrant classifier according to the task information urgency to output four types of task queues; According to the four types of task queues, establish a component-task association matrix; extend the colored Petri net model, add task reliability constraints, and output component status labels and component reliability scores; According to the component reliability scores, divide multiple levels of power pools, and allocate power pool weights according to the task reliability scores to output a power pool configuration table; Perform reliability-aware scheduling according to the four types of task queues and the power pool configuration table to output task scheduling instructions; According to the task scheduling instructions and the component status labels, establish a task-component joint optimization model to adjust the parameters of the task weight model and output the optimal sensitive parameters; According to the optimal sensitive parameters, adjust the power pool configuration table to output a power pool adjustment strategy.

2. The management method of a power supply system of a wearable intelligent device according to claim 1, wherein, The steps to establish the task weight model include: Extract the historical task features from the historical task data and use the historical task features as the training set; According to the historical task features, calculate the historical success rate and average delay deviation of the tasks, and combine with the Weibull distribution parameters to output a task failure probability distribution table; Learn the parameters of the Sigmoid function by the gradient descent method; and estimate the Weibull distribution parameters by MLE to obtain the task weight model.

3. The management method of a power supply system of a wearable intelligent device according to claim 2, characterized in that, The steps to extend the colored Petri net model include: Collect the list of hardware components of the device; Create an independent place for each hardware component in the hardware component list and mark it using the color marking rule to output a component operation status table; Formulate state transition rules according to the task requests in the four types of task queues, the component operation status table, and the power consumption monitoring data; Configure arc connection rules according to the component operation status table and the state transition rules to output the colored Petri net model.

4. The management method of a power supply system of a wearable intelligent device according to claim 1, characterized in that, The steps to construct the task value model include: Select valid tasks from the task reliability scores, extract the P_score column as the feature vector, and output the reliability feature vector; Obtain the task delay tolerance from the task node attribute information, calculate the emergency coefficient, and output the emergency coefficient vector; According to the task node attribute information, calculate the average energy consumption of all tasks, and calculate the energy consumption weight according to the average energy consumption to output the energy consumption weight vector; Combine the reliability feature vector, the emergency coefficient vector, and the energy consumption weight vector column by column into a three-dimensional feature matrix to output a task value feature matrix; According to the task value feature matrix, use a random forest regression model to predict the task value.

5. The management method of a power supply system of a wearable intelligent device according to claim 1, characterized in that, The steps to construct the four-quadrant classifier include: Divide the task information into four task urgency quadrants according to the urgency of the task information; According to the task value model, perform two-dimensional division on the task urgency quadrants; Trigger a compensation mechanism according to the result of the two-dimensional division, and reduce its resource occupancy through dynamic voltage and frequency adjustment.

6. The management method of a power supply system for a wearable intelligent device according to claim 5, wherein The steps of allocating the power pool weights include: Based on the task reliability score and the task value, calculate the power pool weight allocation coefficient using the entropy weight method; The power pool includes a core power pool, a dynamic power pool, and a standby power pool; set a guaranteed weight threshold for the core power pool, allocate the weight of the dynamic power pool according to the power pool weight allocation coefficient, and transfer the remaining capacity to the standby power pool, and output a power pool configuration table.

7. The management method of a power supply system for a wearable intelligent device according to claim 6, wherein, The steps of performing reliability-aware scheduling include: Real-time monitor the component status tags and the remaining power of the power pool; The task urgency quadrants include: the first quadrant is high-value and high-urgency tasks, and the third quadrant is low-value and high-urgency tasks; use preemptive scheduling for the high-value and high-urgency tasks, and bind the core power pool for power supply; Enable elastic scheduling for the low-value and high-urgency tasks, allow degraded components to execute, and output a scheduling decision.

8. The management method of a power supply system of a wearable intelligent device according to claim 7, characterized in that, The steps of adjusting the parameters of the task weight model include: According to the current power pool weights and the component status tags, convert multiple objectives into single-objective optimization through weighted summation to generate a Pareto front solution set; According to the Pareto front solution set, adjust the parameters of the task weight model item by item, observe the change range of the model parameters to generate sensitive parameters; adjust the sensitive parameters according to the real-time performance index feedback, and generate the optimal sensitive parameters through incremental optimization.

9. The management method of a power supply system of a wearable intelligent device according to claim 8, characterized in that, The power pool adjustment strategy includes: Reallocate the power pool capacity based on the power pool weight allocation coefficient, execute a voltage ramp-up strategy for the core power pool, and output a core power pool capacity adjustment instruction; the task urgency quadrants also include: low-value and low-urgency tasks. If the standby power pool is enabled, trigger a dynamic load balancing protocol to migrate the low-value and low-urgency tasks to low-power components for execution.

10. A management system for a power supply system of a wearable intelligent device, which adopts a management method for a power supply system of a wearable intelligent device as described in any one of claims 1 to 9, characterized in that, Include: A task information acquisition module that real-time acquires task information of each load node of the device and outputs task node attribute information; A task reliability scoring module for outputting a task reliability score by establishing a task weight model based on the Sigmoid function and the Weibull distribution based on historical data and the task node attribute information; A task value evaluation module for determining the task information urgency by constructing a task value model according to the task reliability score; A task classification module for constructing a four-quadrant classifier according to the task information urgency and outputting four types of task queues; A component status monitoring module for outputting component status tags and component reliability scores through a colored Petri net model according to the four types of task queues and power consumption data; A power pool management module for dividing multiple levels of power pools according to the component reliability scores and the component status tags, and allocating the weights of the power pools according to the task reliability scores, and outputting a power pool configuration table; A scheduling engine, which is used to formulate a task scheduling strategy according to the four types of task queues in combination with the power pool configuration table and output a task scheduling instruction; A parameter optimization module, which is used to adjust the parameters of the task weight model according to the task scheduling instruction and the component status label and output optimal sensitive parameters; A system policy adjustment module, which is used to adjust the power pool configuration table according to the optimal sensitive parameters and output a power pool adjustment policy.

Images