Download presentation
Presentation is loading. Please wait.
1
Central South University
A Simple Encoding Mechanism in Evolutionary Algorithms and Its Applications Yong Wang School of Automation Central South University
2
Evolutionary Algorithms and Encoding
Outline Evolutionary Algorithms and Encoding I: A Simple Encoding Mechanism and Its Application to Wind Farm Layout II: A Simple Encoding Mechanism and Its Application to Multi-UAV Enabled Mobile Edge Computing III: A Simple Encoding Mechanism and Its Application to a UAV-Assisted IoT Data Collection System IV: A Simple Encoding Mechanism and Its Application to Blockchain Conclusion 2
3
Evolutionary Algorithms and Encoding
Outline Evolutionary Algorithms and Encoding I: A Simple Encoding Mechanism and Its Application to Wind Farm Layout II: A Simple Encoding Mechanism and Its Application to Multi-UAV Enabled Mobile Edge Computing III: A Simple Encoding Mechanism and Its Application to a UAV-Assisted IoT Data Collection System IV: A Simple Encoding Mechanism and Its Application to Blockchain Conclusion 3
4
Evolutionary Algorithms (1/2)
What are evolutionary algorithms (EAs)? EAs are intelligent optimization and search techniques inspired by nature Why evolutionary algorithms (EAs)? The framework of evolutionary algorithms (EAs) x f(x) the first individual the second individual Population Selection the NPth individual x y f(x,y) Replacement Parent Set The optimal solution! The optimal solution! Crossover Mutation Offspring Set Is it the optimal solution? 4
5
Evolutionary Algorithms (2/2)
Main paradigms Genetic algorithm (GA) Evolution strategy (ES) Evolutionary programming (EP) Ant colony optimization (ACO) Differential evolution (DE) Particle swarm optimization (PSO) … the most popular paradigms in current studies 5
6
Encoding (1/6) In EAs, a population includes several individuals.
What does each individual represent? A solution How to design an individual Encoding 6
7
Encoding (2/6) Binary encoding Representative EAs: GA
Representative applications: feature selection, knapsack problems, etc. string position (locus) string (chromosome) 1 12 24 An individual character, feature, (gene) feature value (allele) 16 bits & 16 binary digits 7
8
Encoding (3/6) An issue in binary encoding — decoding An example 8
9
Encoding (4/6) Real-coded encoding
(-1.2, 2.3) Representative EAs: ES, EP, PSO, DE, etc Representative applications: real-value optimization problems An individual 9
10
Encoding (5/6) Order encoding Representative EAs: ACO
Representative applications: TSP, scheduling problems 2 5 1 4 7 1 4 3 2 1234 An individual 10
11
Encoding (6/6) Mixed encoding
(-1.2, 2.3, 1, 0, 0, 1) real-coded encoding + binary encoding (-1.2, 2.3, 1, 2, 3, 4) real-coded encoding + order encoding Representative applications: automotive body design, task scheduling and allocation, etc. A very challenging issue in current study An individual An individual 11
12
Evolutionary Algorithms and Encoding
Outline Evolutionary Algorithms and Encoding I: A Simple Encoding Mechanism and Its Application to Wind Farm Layout II: A Simple Encoding Mechanism and Its Application to Multi-UAV Enabled Mobile Edge Computing III: A Simple Encoding Mechanism and Its Application to a UAV-Assisted IoT Data Collection System IV: A Simple Encoding Mechanism and Its Application to Blockchain Conclusion 12
13
Wind Farm Layout (1/2) 13
14
Wind Farm Layout (2/2) Model
Wake Effect Objective: find the optimal layout of all the wind turbines to maximize the power output of a wind farm Constraints: a wind turbine should lie within the wind farm; the distance between wind turbine i and any other wind turbine should not be shorter than 5R 14
15
Related Work (1/2) Greedy methods
Only one layout is considered and only one wind turbine in the layout is moved by some strategies similar to random search or local search in each iteration. The evaluation of the layout can be sped up by utilizing the caching technique. Drawback: poor global search ability 15
16
Related Work (1/2) EAs Each individual usually represents an entire layout and each dimension of an individual denotes a coordinate of a wind turbine. After implementing evolutionary operators on the individuals, maybe many wind turbines rather than just one wind turbine are updated. It is hard to use the caching technique to accelerate the evaluation of a layout under this condition. Drawback: a great deal of computational time 16
17
A Simple Encoding Mechanism (1/2)
Motivation It is clear that the location of a wind turbine is determined by a 2-D coordinate system (i.e., x-axis and y-axis) and each dimension of all the wind turbines has the same search region. Y. Wang, H. Liu, H. Long, Z. Zhang, and S. Yang. Differential evolution with a new encoding mechanism for optimizing wind farm layout. IEEE Transactions on Industrial Informatics, vol. 14, no. 3, pp , 17
18
A Simple Encoding Mechanism (2/2)
Advantages A population represents a layout. Under this condition, if only one wind turbine (i.e., one individual) is moved in each iteration, the caching technique can be applied. Each individual can be updated by evolutionary operators. In this manner, the global search ability can be strengthened. The population size does not need to be predefined since it is equal to the number of wind turbines. Therefore, the advantages of greedy methods and EAs can be combined effectively by this encoding mechanism. 18
19
Evolutionary Algorithms and Encoding
Outline Evolutionary Algorithms and Encoding I: A Simple Encoding Mechanism and Its Application to Wind Farm Layout II: A Simple Encoding Mechanism and Its Application to Multi-UAV Enabled Mobile Edge Computing III: A Simple Encoding Mechanism and Its Application to a UAV-Assisted IoT Data Collection System IV: A Simple Encoding Mechanism and Its Application to Blockchain Conclusion 19
20
Multi-UAV Enabled Mobile Edge Computing
Task 1: deployment of UAVs the number and locations of UAVs Task 2: task scheduling whether a task is executed locally or on a UAV and how many resources should be allocated 20
21
Model Objective: jointly optimize the deployment of UAVs and the task scheduling to minimize the system energy consumption 21
22
Challenges We need to optimize the number of UAVs (N), the location of UAVs, and the offloading decisions and the resource allocation for M mobile users. It is evident that the number of decision variables increases with the increase of M and/or N. It includes an integer decision variable, continuous decision variables, and binary decision variables. Thus, it is an optimization problem with mixed decision variables. Actually, the deployment of UAVs and the task scheduling are closely coupled with each other. 22
23
A Simple Encoding Mechanism (1/3)
Motivation Each UAV has two decision variables (i.e., Xj and Yj in the x-axis and y-axis), and all elements in {X1,…,XN} have the same upper and lower bounds, as well as all elements in {Y1,…,YN}. Y. Wang, Z.-Y. Ru, K. Wang, and P.-Q. Huang. Joint deployment and task scheduling optimization for large-scale mobile users in multi-UAV enabled mobile edge computing. IEEE Transactions on Cybernetics, minor revision. 23
24
A Simple Encoding Mechanism (2/3)
Advantages Each individual has a fixed length during the evolution, rather than a variable length. The length of each individual is equal to two, which means the deployment of UAVs is optimized in a very low-dimensional search space, i.e., two. 24
25
A Simple Encoding Mechanism (3/3)
How to adjust the population size After analysing this system, to achieve the minimum energy consumption, we should give a priority to the number of UAVs under the condition that all tasks can be completed. Based on this property, we first determine the maximum number of UAVs, and gradually reduce the number of if all tasks can be completed. In principle, the number of UAVs is adaptively adjusted. 25
26
Evolutionary Algorithms and Encoding
Outline Evolutionary Algorithms and Encoding I: A Simple Encoding Mechanism and Its Application to Wind Farm Layout II: A Simple Encoding Mechanism and Its Application to Multi-UAV Enabled Mobile Edge Computing III: A Simple Encoding Mechanism and Its Application to a UAV-Assisted IoT Data Collection System IV: A Simple Encoding Mechanism and Its Application to Blockchain Conclusion 26
27
A UAV-Assisted IoT Data Collection System
A UAV is employed as a data collection platform for a group of ground IoT devices 27
28
Model Objective: optimize the UAV deployment, including the number and locations of stop points of the UAV to minimize the energy consumption of the system which is composed of the energy consumption of the UAV and all IoT devices. 28
29
Challenges When using evolutionary algorithms to solve this deployment problem, each individual usually represents an entire deployment. Due to the fact that the number of stop points is unknown a priori, the length of each individual in the population should be varied during the optimization process. As a result, the commonly used mutation and crossover operators cannot be used directly. 29
30
Related Work (1/2) The approaches with special mutation and crossover operators Each individual in the population represents an entire deployment, which consists of the locations of a set of stop points. Then, the population represents NP entire deployments. The special mutation and crossover operators need to be adopted. Drawback: the dimension of the search space may change from generation to generation for each individual, thus causing confused search. 30
31
Related Work (2/2) The approaches with additional variables
The individual and the population are also encoded into an entire deployment and NP entire deployments, respectively. The length of each individual in the population is 3kmax, consisting of kmax additional variables (i.e., Ci,1,…,Ci,kmax) and 2kmax original variables (i.e., Xi,1, Yi,1,…,Xi,kmax,Yi,kmax). Drawback: the length of each individual will drastically increase due to the introduction of additional variables, thus leading to a high-dimensional search space. 31
32
A Simple Encoding Mechanism (1/4)
Motivation Although the number of stop points that the UAV needs to visit is variable during the evolution, the dimension of the location of each stop point is fixed. Since the flying height H of the UAV is a constant in this case, the dimension of the location of each stop point can be considered as two (i.e., the x-axis and y-axis). Moreover, the values of the locations of all stop points in the x-axis and y-axis have the same upper and lower bounds. 32
33
A Simple Encoding Mechanism (2/4)
The location of each stop point is encoded into an individual. The whole population represents an entire deployment. The population size is equal to the number of stop points. P.-Q. Huang, Y. Wang, K. Wang, and K. Yang. Differential evolution with a variable population size for deployment optimization in a UAV-assisted IoT data collection system. IEEE Transactions on Emerging Topic in Computational Intelligence, submitted. 33
34
A Simple Encoding Mechanism (3/4)
Advantages The population size is equal to the number of stop points, which is optimized during the evolution. Therefore, this parameter has been eliminated. The length of each individual in the proposed encoding mechanism is always fixed. In this manner, the special mutation and crossover operators are not required. Compared with the approaches with additional variables, the length of each individual in the population is equal to two. Consequently, the dimension of the search space is remarkably reduced. 34
35
A Simple Encoding Mechanism (4/4)
How to adjust the population size We design a strategy to adjust the population size according to the performance improvement. By this strategy, the number of stop points can be increased, reduced, or kept unchanged adaptively. 35
36
Evolutionary Algorithms and Encoding
Outline Evolutionary Algorithms and Encoding I: A Simple Encoding Mechanism and Its Application to Wind Farm Layout II: A Simple Encoding Mechanism and Its Application to Multi-UAV Enabled Mobile Edge Computing III: A Simple Encoding Mechanism and Its Application to a UAV-Assisted IoT Data Collection System IV: A Simple Encoding Mechanism and Its Application to Blockchain Conclusion 36
37
Blockchain Edge computing assists the mobile blockchain network to offload mining tasks 37
38
Model Objective: jointly optimize offloading decision, and communication and computation resources to maximize the profit of miners 38
39
Challenges The offloading decision is a binary variable and the resource allocations are continuous variables. Thus, it is an optimization problem with mixed variables. Although the search space is large, the complex constraints make the feasible region small. 39
40
Related Work (1/2) ineffective values Communication resources
Communication resources Offloading decision x1 x2 x3 x4 f1 f2 f3 f4 p1 p2 p3 p4 Computation resources An individual 1 f1 f2 f3 f4 p1 p2 p3 p4 ineffective values 40
41
Related Work (2/2) Remove redundant information
Although we can remove redundant resource allocation information of miners who do not offload, due to the channel disturbance, it is very easy to generate infeasible offspring if we utilize traditional evolutionary operators. 41
42
A Simple Encoding Mechanism (1/4)
Motivation Each individual represents the resource allocations of one miner who chooses to offload. Thus, it provides a natural way to eliminate redundant information for miners who choose not to offload and to avoid to optimize mixed variables. 42
43
A Simple Encoding Mechanism (2/4)
The computation resource and communication resource of one miner who chooses to participate in mining are encoded into an individual. An individual (resource allocations of one miner) fi1 pi1 fi2 pi2 … f1n pin A population (an entire allocation scheme) 43
44
A Simple Encoding Mechanism (3/4)
Advantages The encoding mechanism can remove redundant information and reduce the size of the search space. It transforms the mixed variable optimization problem into a continuous optimization problem A population represents a resource allocation scheme of all miners. Under this condition, if only the resource allocation of one miner is updated in each iteration, it is more likely to generate a feasible offspring compared with traditional evolutionary operators. 44
45
A Simple Encoding Mechanism (4/4)
How to adjust the population size Based on our analysis, in the early stage, the revenue can be increased by increasing the number of miners under constraints. In the later stage, it is necessary to reduce the number of miners to maintain a higher value of the revenue. We design a probabilistic strategy to adjust population size. The number of miners participating in mining can be increased, decreased or maintained adaptively. 45
46
Thank you very much for your attention!
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.