Embedded multi-task scheduling optimization method for wearable electroencephalogram sensor

Abstract

The invention discloses an embedded multi-task scheduling optimization method for a wearable electroencephalogram sensor, and the method comprises the steps: S1, carrying out the priority sorting of all tasks in a task set according to the multi-task and multi-target scheduling demands of the wearable electroencephalogram sensor; s2, generating an initialized group S0 and an external archive solution set lambda; step S3, executing Pareto local search operation on the initialized group S0 to quantify a solution set; s4, whether the maximum number of iterations is reached or not is judged, and if yes, the step S6 is executed; s5, executing disturbance operation on the current solution set, updating the external archived solution set lambda, and returning to S4; and S6, outputting the external archive solution set lambda as a final optimization result. According to the embedded multi-task scheduling optimization method for the wearable electroencephalogram sensor, the memory sensing multi-target task scheduler based on the iterative local search algorithm is designed in order to solve the limitation of limited onboard computing resources and energy consumption in wearable medical equipment.

Description

Technical Field

[0001] This invention relates to the field of embedded multi-core processor technology for wearable EEG acquisition devices, and in particular to an embedded multi-task scheduling optimization method for wearable EEG sensors. Background Technology

[0002] Depression is a prevalent mental health disorder characterized by persistent sadness, loss of interest or pleasure, and feelings of worthlessness or guilt. It affects individuals of all ages and backgrounds, has negative social impacts, and contributes significantly to disability-adjusted life years (DALYs) globally. Approximately 284 million people worldwide suffer from anxiety disorders, and 264 million from depression. Depression's high disabling nature poses a significant challenge to global public health systems. Globally, general practitioners correctly identify depression only 47.3% of cases based on interviews and clinical scales. This indicates a severe lack of objective and universally applicable diagnostic tools for detecting and diagnosing depressive disorders, leading to difficulties in timely diagnosis and intervention for individuals with depression.

[0003] With advancements in affective computing, cognitive computing, computational psychophysiology, and artificial intelligence, researchers have widely adopted electroencephalography (EEG) signals as a key tool for detecting potential objective physiological indicators of depression. EEG signals, as a tool for recording real-time changes in brain activity, have been shown to be closely linked to depression, particularly prefrontal cortex signals. Existing technologies collectively underscore the importance and effectiveness of combining wearable EEG devices with advanced machine learning and deep learning techniques for depression detection. This integration further drives the need to deploy optimized learning models on resource-constrained computing platforms such as microcontroller units, thus fostering the emergence of TinyML, tailored for edge computing scenarios. TinyML enables machine learning models to run on resource-constrained portable wearable devices. However, executing quantized ML / DL methods in real-time on heavily constrained portable wearable devices must address latency and energy efficiency issues. Their computationally and memory-intensive characteristics pose a significant challenge to the practical deployment of wearable devices. In other words, resource-constrained wearable devices face difficulties in supporting the execution of computationally and memory-intensive ML / DL methods, which poses a challenge to wearable devices based on embedded multi-core processors that require high energy efficiency, low latency, high throughput, and the efficiency and accuracy of maintaining models.

[0004] Existing research primarily focuses on epilepsy detection, with limited exploration of low-power embedded multi-core processor hardware-algorithm co-optimization for depression detection and classification. Typically, the classification process for depression EEG signal data involves data preprocessing, feature extraction, and model classification. However, directly executing these processes on embedded multi-core processors for multitasking presents significant technical hurdles. For example, reducing maximum completion time often requires increasing resources through parallelization to improve task throughput per unit time, but this leads to increased energy consumption. Over-optimizing maximum completion time can reduce throughput due to resource fragmentation. Simultaneously, maintaining high throughput still requires high resource load operation, further exacerbating energy consumption pressures. Therefore, the complexity of these application tasks highlights the need for automated optimization of scheduling and mapping configuration tasks on computationally robust embedded systems. In fact, mapping and scheduling a set of dependent tasks, in its simplest form without resource usage constraints or batch processing, is a classic NP-hard problem. To meet the latency, energy consumption, and throughput requirements of hardware platforms during inference, finding scalable and efficient methods to map this complex computational process to embedded multi-core processors is becoming increasingly important.

[0005] To address the aforementioned issues, this invention addresses the complex process of managing multiple tasks from EEG preprocessing to modular classification models on embedded multi-core processors. A novel memory-aware multi-objective task scheduler based on an iterative local search algorithm is introduced. This scheduler integrates Pareto local search mechanisms with multi-objective optimization strategies, significantly improving system task scheduling efficiency while maintaining satisfactory classification performance of ML / DL models. Furthermore, based on the designed device, collected data, proposed algorithm, and optimized model, this invention develops a wearable EEG sensor-assisted depression diagnosis system driven by an embedded microprocessor. This system shows great potential as a universally effective tool for assisting depression screening. Summary of the Invention

[0006] The purpose of this invention is to provide an embedded multi-task scheduling optimization method for wearable EEG sensors. To address the limitations of limited onboard computing resources and energy consumption in wearable medical devices, a memory-aware multi-objective task scheduler based on an iterative local search algorithm is designed. This method can be deployed in a low-energy, high-efficiency EEG processing task scheduling system for wearable medical devices that are sensitive to computing resources and energy.

[0007] This invention provides an embedded multi-task scheduling optimization method for wearable EEG sensors, comprising the following steps:

[0008] Step S1: To meet the multi-task, multi-objective scheduling requirements of wearable EEG signal sensors, obtain the task scheduling table, processor parameters, and DVFS level; prioritize each task in the task set; initialize the start time, execution time, completion time, memory usage, and DVFS level of all subtasks on the processor; and establish a memory-constrained multi-core processor multi-objective task scheduling planning model.

[0009] Step S2: Generate an initial population S0, calculate the objective function value of each individual in S0, and construct an external archive solution set Λ;

[0010] Step S3: Perform a Pareto local search operation on the initial population S0 to evolve the solution set, update the external archive solution set Λ, and save the current optimal solution.

[0011] Step S4: Determine if the maximum number of iterations has been reached: if so, proceed to S6;

[0012] Step S5: Perform a perturbation operation on the current solution set, and perform a Pareto local search operation on the perturbed solution set. Update the external archive solution set Λ and return to S4.

[0013] Step S6: Output the external archive solution set Λ as the final optimization result.

[0014] Preferably, in step S3, the Pareto local search operation includes calculating the non-dominated relations of individuals in the population, filtering the solution set of the Pareto front, and updating the Pareto non-dominated solutions of the external archive Λ.

[0015] Preferably, in step S5, the perturbation operation includes changing the solution structure by randomly adjusting the task encoding, resetting the DVFS level or processor allocation scheme of the subtask, and keeping the perturbation amplitude within a preset mutation threshold.

[0016] Preferably, in step S2, the update mechanism of the external archive solution set Λ includes comparing the new solution with the non-dominant relationship in the archive, eliminating the solution dominated by the new solution, adding the new solution to the archive when it is not dominated by any solution in the archive, and periodically cleaning up duplicate solutions in the archive to control the size.

[0017] Preferably, in step S2, the objective function value is calculated based on a weighted combination of the task completion time span, total processor power consumption, peak memory usage, and DVFS level switching frequency.

[0018] Preferably, in step S1, the priority ranking includes any strategy such as earliest deadline priority, longest execution time priority, and task dependency depth priority.

[0019] Therefore, this invention employs the aforementioned embedded multi-task scheduling optimization method for wearable EEG sensors. Based on the Pareto optimality principle, the priority of tasks is dynamically adjusted to further optimize the maximum completion time, throughput, and total energy consumption of a wearable EEG sensor-assisted depression diagnosis system with an embedded multi-core processor, while maintaining the classification accuracy of the system. This provides a universal and effective solution for developing a low-power, real-time depression diagnosis system using wearable EEG devices.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of an embedded multi-task scheduling optimization method for wearable EEG sensors according to the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0024] The terms "first," "second," and similar words used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0025] Example 1

[0026] like Figure 1 As shown, the present invention provides an embedded multi-task scheduling optimization method for wearable EEG sensors, comprising the following steps:

[0027] Step S1: To meet the multi-task, multi-objective scheduling requirements of wearable EEG signal sensors, obtain the task scheduling table, processor parameters, and DVFS level; prioritize each task in the task set; initialize the start time, execution time, completion time, memory usage, and DVFS level of all subtasks on the processor; and establish a memory-constrained multi-core processor multi-objective task scheduling planning model.

[0028] In step S1, the priority ranking includes any strategy such as earliest deadline first, longest execution time first, and task dependency depth first.

[0029] Step S2: Generate an initial population S0, calculate the objective function value of each individual in S0, and construct an external archive solution set Λ;

[0030] In step S2, the update mechanism of the external archive solution set Λ includes comparing the new solution with the non-dominant relationship in the archive, eliminating the solution dominated by the new solution, adding the new solution to the archive when it is not dominated by any solution in the archive, and periodically cleaning up duplicate solutions in the archive to control the size.

[0031] In step S2, the objective function value is calculated based on a weighted combination of the task completion time span, total processor power consumption, peak memory usage, and DVFS level switching frequency.

[0032] Step S3: Perform a Pareto local search operation on the initial population S0 to evolve the solution set, update the external archive solution set Λ, and save the current optimal solution.

[0033] In step S3, the Pareto local search operation includes calculating the non-dominated relations of individuals in the population, filtering the solution set of the Pareto front, and updating the Pareto non-dominated solutions of the external archive Λ.

[0034] Step S4: Determine if the maximum number of iterations has been reached: if so, proceed to S6;

[0035] Step S5: Perform a perturbation operation on the current solution set, and perform a Pareto local search operation on the perturbed solution set. Update the external archive solution set Λ and return to S4.

[0036] In step S5, the perturbation operation includes changing the solution structure by randomly adjusting the task encoding, resetting the DVFS level or processor allocation scheme of the subtask, and keeping the perturbation amplitude within a preset mutation threshold.

[0037] Step S6: Output the external archive solution set Λ as the final optimization result.

[0038] A memory-aware multi-objective task scheduler based on an iterative local search algorithm, called MAMILS, was developed and optimized for onboard execution of AI models with low computational and space complexity.

[0039] By using the proposed MAMILS algorithm to optimize key performance indicators such as minimizing maximum completion time, total energy consumption, and task throughput in a wearable three-lead EEG sensor system, significant optimization of system task scheduling efficiency was achieved, while maintaining satisfactory classification performance of the ML / DL model.

[0040] Based on the designed device, collected data, proposed algorithm, and optimized model, a wearable EEG-assisted depression diagnosis system driven by an embedded microprocessor was developed. It features low power consumption and real-time processing, and has great potential as a universally effective tool for assisting depression screening.

[0041] Therefore, this invention adopts the above-mentioned embedded multi-task scheduling optimization method for wearable EEG sensors. To solve the limitations of limited onboard computing resources and energy consumption in wearable medical devices, a memory-aware multi-objective task scheduler based on an iterative local search algorithm is designed. This scheduler can be deployed in a low-energy, high-efficiency EEG processing task scheduling system for wearable medical devices that are sensitive to computing resources and energy.

[0042] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An embedded multi-task scheduling optimization method for wearable EEG sensors, characterized in that, Includes the following steps: Step S1: To meet the multi-task, multi-objective scheduling requirements of wearable EEG signal sensors, obtain the task scheduling table, processor parameters, and DVFS level; prioritize each task in the task set; initialize the start time, execution time, completion time, memory usage, and DVFS level of all subtasks on the processor; and establish a memory-constrained multi-core processor multi-objective task scheduling planning model. Step S2: Generate an initial population S0, calculate the objective function value of each individual in S0, and construct an external archive solution set Λ; Step S3: Perform a Pareto local search operation on the initial population S0 to evolve the solution set, update the external archive solution set Λ, and save the current optimal solution. Step S4: Determine if the maximum number of iterations has been reached: if so, proceed to S6; Step S5: Perform a perturbation operation on the current solution set, and perform a Pareto local search operation on the perturbed solution set. Update the external archive solution set Λ and return to S4. Step S6: Output the external archive solution set Λ as the final optimization result.

2. The embedded multi-task scheduling optimization method for wearable EEG sensors according to claim 1, characterized in that, In step S3, the Pareto local search operation includes calculating the non-dominated relations of individuals in the population, filtering the solution set of the Pareto front, and updating the Pareto non-dominated solutions of the external archive Λ.

3. The embedded multi-task scheduling optimization method for wearable EEG sensors according to claim 1, characterized in that, In step S5, the perturbation operation includes changing the solution structure by randomly adjusting the task encoding, resetting the DVFS level or processor allocation scheme of the subtask, and keeping the perturbation amplitude within a preset mutation threshold.

4. The embedded multi-task scheduling optimization method for wearable EEG sensors according to claim 1, characterized in that, In step S2, the update mechanism of the external archive solution set Λ includes comparing the new solution with the non-dominant relationship in the archive, eliminating the solution dominated by the new solution, adding the new solution to the archive when it is not dominated by any solution in the archive, and periodically cleaning up duplicate solutions in the archive to control the size.

5. The embedded multi-task scheduling optimization method for wearable EEG sensors according to claim 1, characterized in that, In step S2, the objective function value is calculated based on a weighted combination of the task completion time span, total processor power consumption, peak memory usage, and DVFS level switching frequency.

6. The embedded multi-task scheduling optimization method for wearable EEG sensors according to claim 1, characterized in that, In step S1, the priority ranking includes any strategy such as earliest deadline first, longest execution time first, and task dependency depth first.

Images