Page 150 - ITU KALEIDOSCOPE, ATLANTA 2019

2019 ITU Kaleidoscope Academic Conference

5.3 Self-repairing artificial fish swarm algorithm objective functions Y 1 , as shown in Equation (17):

Artificial fish swarm algorithm is a new bionic and swarm ( A Õ Õ 1 ! )

B
intelligence algorithm to search for an optimized solution by Y 1 = Max ∗ x ij ∗ m i (17)
modeling four daily behavior including preying, swarming, i=0 j=0 p ij
following and random behavior. Since only one commodity in Compared to the Ideal point method and Hierarchical method,
each type of sensing device could be selected, we proposed we choose the Linear-weighting method that uses λ 1 and λ 2
the representation scheme of artificial fish to compress the as weights of two objective functions based on the elderly
solution space and reduce the search region, as shown in desires. As shown in Equation (18), λ 1 and λ 2 are adjusted
Equation (16). More specifically, the value of x i shows that based on the inclination of the elderly, while λ 1 +λ 2 =1. Also,
the x i th commodity for ith type of sensing devices is selected. w 1 and w 2 are the weighting factors of the objective functions
Y 1 and Y 2 . Finally, Max{Y 1 } is the optimal value of Y 1 and
Max{Y 2 }is the optimal value ofY 2 under the same constraints.
X f = {x 1 , x 2 , x 3 ,..., x i ,..., x m } (16)
Y = Max{λ 1 w 1 Y 1 + λ 2 w 2 Y 2 } (18)
Definition 4 The artificial fish X f is located near the
constraint boundary of the problem, if X f is a feasible
solution and it would become infeasible when we find a 1
w 1 = (19)
commodity k in type i and assign k to x i , where i ∈ Max{Y 1 }
{1,2,3...A}, k ∈ {1,2,3...B} and the Performance/Price of 1
kth commodity is bigger than x i ’s in type i sensing devices. w 2 = (20)
Max{Y 2 }
The research [22] proves that an optimized solution and
constraint boundary of the knapsack problem are usually 6. EXPERIMENTS AND RESULTS ANALYSIS
symbiotic. Thus, we proposed a self-repairing strategy In order to verify the accuracy and validity of the proposed
to repair artificial fish return to the constraint boundary. APGS, we formalized 40 care demands and 200 smart
For any infeasible solution X f , there are two self-repairing services for geriatric care, including hypertension, heart
strategies, where Operation (1) on x i with the least value of diseases and diabetes. Firstly, we need to verify the
Performance/Price and Operation (2) on x i with the biggest scientificity and validity of selected services for geriatric care
value of Price: in experiment 1; then, the global optimization capability and
Operation (1): Replacing x i with k, if Performance/Price of convergence rate of SAFSA should be verified in experiment
kth commodity is biggest in the same type of sensing devices 2; finally, we built the evaluation indicator system of SDSP
whose Price is less than the Price of x i , where i ∈ {1,2,3...A} to verify the scientificity and validity of SDSP of the smart
and k ∈ {1,2,3...B}. home, based on evaluations of smart home researchers.
Operation (2): Replacing x i with k, as the Price of kth In the first experiment, we listed care demands for geriatric
commodity is least in the same type of sensing devices whose experts to select needed care demands for 100 sample elderly,
Performance/Price is bigger than x i , where i ∈ {1,2,3...A} suffering from hypertension, diabetes and heart diseases.
and k ∈ {1,2,3...B}. Weight parameters in SDMM are set as: ϕ 1 =ϕ 2 =0.5, γ 1 =0.7
and γ 2 =0.3. Then, we recommended smart services, whose
Proof 1 If the infeasible artificial fish X f becomes a feasible mapping similarity are bigger than 0.8, for geriatric experts to
solution after the last Operation (1), we can conclude that manually select suitable smart services to meet care demands.
Ñ
Ñ
the Price of kth sensing device is less than x i . Assume that Ultimately, we use Precision( C C D ), Recall( C D D ) and
the artificial fish X f is not near the constraint boundary. F1( 2∗Precision∗Recall ) to verify the validity of selected
Precision+Recall
According to Definition 4, there is a sensing device whose services, where C is the service set calculated by proposed
Performance/Price is bigger than k and Price is less than SDMM and D is the service set selected by experts.
x i . However, due to the description of Operation (1), the We compared the performance of mapping similarity method
value of Value/Price of k is biggest in the candidate sensing (MS), keyword mapping method (KM) and variable precision
devices whose Price is less than Price of x i . Therefore, the rough set method (VPRS). As shown in Figure-3(a), the
assumption is not true and X f is near the constraint boundary precision is more than 94.4%, which indicates that selected
after Operation (1) of self-repairing behavior. smart services can excellently cover care demands of the
sample elderly. Compared to KM and VPRS methods, our
approach has higher accuracy and coverage, since precision,
As the above Proof 1 reveals, the artificial fish X f
become feasible solution after Operation (1). Additionally, recall and F1 of the proposed MS are almost the biggest.
Operation (2) can be similarly proved with the same method. The similar function and description between smart services
Therefore, we properly invoke self-repairing behavior to are important reasons causing 4.5% of the mismatch. In
Figure-3(b), the performance of MS for elderly suffering
search for the optimal solution, when the artificial fish X f
become infeasible after four basic behaviors. from diabetes is better than hypertension and heart diseases.
Since two objective functions of the multi-objective knapsack In particular, the performance of MS for a single geriatric
problem have opposite targets, we make a conversion for the disease is usually better than two or more geriatric diseases.

– 130 –