IEEE Consumer Electronics Magazine - January/February 2022 - 75

propose to add an adaptive mutation and an
injection strategy to alleviate these problems.
Adaptive mutation solves the problem that
genes of excellent individuals with high adaptabilities
are destroyed and enters a random
search. If the mutation probability is too low,
then it is difficult to introduce new genes, which
makes the algorithm's later search stalled.17 The
injection strategy has a powerful supplementary
diversity mechanism, which alleviates the earlier
convergence in the algorithm and can make the
algorithm easier to avoid reaching the local optimal
solution.
GA for NP-Hard
The task scheduling problem of heterogeneous
multicore processors is an NP-hard problem.
How to quickly obtain the task scheduling
strategy of the multitask model is a key indicator
to evaluate the performance of the algorithm. Efficient
algorithms can reduce the time consumed
to solve the task scheduling strategy. The scheduling
strategy designed in this article improves
the GA to build the model. Selection, crossover,
and mutation in the GA are the key operations of
population iterative evolution. The final evolution
and optimization effects of the entire population
are related to the fitness function, crossover
probability, and mutation probability set in the
algorithm. Therefore, in order to solve the problem
of low efficiency and local optimal solution in
the GA, we add adaptive mutation and injection
strategies to improve the GA.
Tasks Prescheduling
To bind the execution relationship between
heterogeneous processing cores and various
types of tasks, we detect the computing capacities
of different cores by processing specified
tasks beforehand. We introduce an regulatory
factor g to adjust the proportion of tasks for testing,
where 0 < g < 1. Based on this, we calculate
the total completion time and find a minimum
one as the initial setting of the intelligence task
scheduling algorithm.
Steps of Intelligence Task Scheduling Strategy
We propose a swarm intelligence algorithm
that improves the GA using the following steps.
The flow chart is shown in Figure 2.
January/February 2022
Figure 2. Flow chart of GA modeling and solving.
Step 1: Initialize parameters of the intelligence
task scheduling strategy. Each individual in
the initial population denotes a scheduling
scheme. In this model, we combine the task
pre-scheduling and the dependencies
between tasks as follows:
(i) We first calculate the HðTiÞ, where
HðTiÞ denotes the height value of task
Ti based on the DAG diagram.
(ii) The tasks are allocated to GPU and CPU
based on the prescheduling strategy.
(iii) According to the HðTiÞ obtained in
step (i), the tasks randomly assigned
on each core are sorted in an increasing
order, and the sorting result is the
execution order of tasks on the processing
unit.
(iv) When the size of the initial population
does not meet the requirements, we
switch back to (ii); otherwise, we go to
step 2.
Step 2: We calculate the fitness function of all
individuals in the population, and all individuals
in the population are ranked based on
the fitness order. The selection of the fitness
function affects whether the algorithm can
find the optimal solution and the convergence
speed directly. We judge the quality
of the solution according to the value of the
fitness function.
We use Sca to denote the size of the current
population, and we use TtotalðSÞ to denote the
75
IEEE Consumer Electronics Magazine - January/February 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2022
