
IEEE/CAA Journal of Automatica Sinica
Citation: | Ameer Hamza Khan, Xinwei Cao, Shuai Li, Vasilios N. Katsikis and Liefa Liao, "BAS-ADAM: An ADAM Based Approach to Improve the Performance of Beetle Antennae Search Optimizer," IEEE/CAA J. Autom. Sinica, vol. 7, no. 2, pp. 461-471, Mar. 2020. doi: 10.1109/JAS.2020.1003048 |
OPTIMIZATION plays an integral part in the efficient operation of almost all real-world systems [1]–[3]. Additionally, with the rise of machine learning in recent years, the development of numerically efficient and robust optimization algorithms has been a topic of great research interest [4]–[8]. Conventional optimization techniques use gradient-based methods to search for the optimum value [9]–[12]. Although these algorithms have proven to be quite stable [13], [14], however, they require the symbolic or numerical computation of the gradient direction. Most of the practical optimization problems are highly nonlinear with multimodal objective functions and non-convex constraints. Numerically evaluating the gradient for such function can be a computationally intensive task [15], [16]. Additionally, the computation of gradient imposes several conditions about continuity and differentiability of the objective function [17]. Such conditions do not hold for the vast majority of the systems, e.g., integer programming [18]. In fact, for the vast majority of the optimization problem, specifically in the control system [19], an accurate model of the system might be unknown in advance and require real-time estimation of parameters [20], [21].
The large majority of the gradient-based optimization is well-suited for computing systems with high computation power. However, their implementation on low-end embedded systems is challenging due to their limited computation power. To overcome such challenges, a new class of optimization algorithm called metaheuristic optimization has gained the attention of researchers [22]–[24]. These algorithms are mostly inspired by natural phenomena and do not require the computation of the gradient of the objective function. These algorithms have been shown in the literature to possess excellent convergence properties and high numerical efficiency. Even for problems in control systems where the model is unknown apriori [25], such algorithms have been employed for the parameter estimation, and tuning of controller gains [26]. One of the significant advantages offered by metaheuristic optimization is a relaxation on the conditions of continuity and differentiability. These algorithms can be effectively employed for solving discrete optimization problems. Additionally, because of their low complexity, they can be efficiently implemented on embedded systems with limited computational power. Given the advantages of metaheuristic algorithms, they have found their applications in several real-world systems [27]–[32].
Almost all the metaheuristic optimization algorithm proposed in [33]–[38] have found their inspiration in natural processes. Most of these algorithms have been inspired by the behavior of the animals, whereas others found their inspiration from the biological and chemical systems [39]. For example, biological evolution have given rise to a complete class of metaheuristic algorithm [40]–[42], e.g., genetic algorithms (GAs) [43], [44]. Other commonly used metaheuristic algorithms inspired by the behavior of living organisms include particle swarm optimization (PSO) [45], [46]. PSO was inspired by the swarming behavior commonly observed in birds and insects. Swarming behavior [47] has been of particular interest to researchers because it was observed that large groups of the birds and insects are able to coordinate by just following a set of straightforward rules. Although each member of the swarm is only aware of its limited surrounding environment, still their combined effort can solve a large-scale, complicated task. This behavior is termed as swarm intelligence [48], [49]. Swarm behavior can be considered as an optimization process in which a group of birds works together to maximize the survivability of the whole swarm. Swarm intelligence has inspired several other algorithms, e.g., ant colony optimization (ACO) [50], which is based on the social behavior of ants. An ant colony can contain millions of ants, but their social behavior and swarm intelligence help them efficiently search for food, thus maximizing the productivity of the colony. Similarly, the artificial fish swarm algorithm [51] is based on the swarming behavior of fishes and other aquatic lives. Although several nature-inspired algorithm metaheuristic algorithms have been presented in literature, few of them are mentioned here: Beetle antennae search (BAS) [52], [53], Cuckoo search [54], [55], invasive weed optimization (IWO) [56], honey bee algorithm (HBA) [57], [58], and firefly algorithms (FAs) [59]. Additionally, the optimization algorithm inspired by the swarming behavior of animals has also been proposed in recent years, e.g., grey wolf optimizer (GWO) [60] and its variations [61], [62]. These algorithms are inspired by the hierarchical structure and hunting mechanisms adopted by the wolfs. Other optimization algorithms are based on the sinusoidal function [63], [64], and use them to converge the initial population of random particles to the real solution. These metaheuristic algorithms have been shown to have good performance in real-world practical scenarios. However, their application in machine learning is still limited. The machine learning models are usually very large-scale, having a large number of parameters. The training of these models require searching very high-dimensional spaces, and conventional metaheuristic algorithm does not scale well to such high-dimensions [65].
As already mentioned in the abstract, in this paper, we consider BAS algorithm [52], [53]. Food foraging behavior observed in beetles inspired BAS. Beetle behavior is of particular research interest because unlike other insects, beetles usually do not work in a swarm and have the ability to search for food individually. Beetles can search for food using its antennae and sensitive sense of smell. Besides, the beetle antennae have several small fibers, and by sensing the difference of smell at each of the fiber, the beetle can develop a map of smell intensity of surrounding. This map helps in finding the direction of maximum smell change. Inspired by this concept, an estimation model can be developed, which helps in approximating the gradient direction. The estimation of an approximated gradient is a key feature of BAS, which distinguishes it from another metaheuristic algorithm. Since its introduction, BAS has found its application in several real-world systems [21], [66]–[75]. The working of the original BAS can be described like this; at each iteration, the value of the objective function is computed at each antennae fiber location, a vector is drawn from the fiber with the lowest value toward the fiber with the highest value, the vector represents the direction of the approximated gradient. The approximated gradient is then used to update the location of the beetle. The working of the original BAS is shown in Fig. 1(a).
One of the key limitations of the BAS algorithm is the selection of a suitable value for initial step size and proper step size annealing. Careful tuning of these parameters is required to achieve a fast convergence rate and stable performance. The previous works on BAS [52], [53] use exponential decay and power-law annealing. However, these methods are not adaptive to the shape of the objective function, i.e., the step size change at the same rate. This behavior can results in slow convergence if the objective function has a narrow valley. From optimization literature, we know that for such objective functions, the gradient-based method, especially stochastic gradient descent (SGD), shows poor performance since the estimated point can bounce back and forth between walls of the valley during iterations [76]. It increases the number of iteration required to reach the optimal point. To improve the convergence rate, several modifications to the SGD algorithm have been proposed in the literature, which adaptively adjusts the step size and direction. On such popular algorithm is adaptive moment estimation (ADAM) [9]. The ADAM algorithm updates the value of step size based on all the past values of gradient and square of gradients. The idea of ADAM is borrowed from momentum SGD [76] but show more robustness and stability near-optimal points. Although several other rules to update step size have been proposed in the literature, e.g., RMSProp, AdaGrad [77], and AdaDelta [78] have been introduced in the literature. However, ADAM combines the features of several of these algorithms, for example, it computes the first and second-order estimates of gradient and also applies the bias-correction on values of these estimates. For this reason, we choose the ADAM algorithm to adaptively update the step size during runtime instead of manually adjusting the parameters in the beginning. The working of the BAS-ADAM is shown in Fig. 1(b). It can be seen that it shows faster convergence as compared to the original BAS.
To demonstrate the fast convergence, efficiency, and effectiveness of the proposed BAS-ADAM algorithm, we performed several sets of experiments and compared its performance with other benchmark metaheuristic algorithm. In the literature on metaheuristic algorithms, the performance of algorithms is verified using low dimensional benchmark functions [57]. In this paper, we took the rigor of the testing process one step further and used the BAS-ADAM to train a hidden layer neural network with nonlinear activation functions. Additionally, we used two other benchmark optimization problems, which were similar to other works in metaheuristic literature. We solved these benchmark optimization problems using the PSO, a state of the art metaheuristic algorithm, original BAS, and BAS-ADAM. It is shown in results that the BAS-ADAM show superior performance as compared to original BAS and PSO in solving the benchmark optimization problems, including the training of neural networks. The main contributions of this paper are highlighted as follows:
1) The proposed algorithm incorporates the concept of gradient estimation into the metaheuristic optimization framework to intelligently decide a search direction. Traditional metaheuristic algorithms are mostly gradient-free and use random search direction.
2) The algorithm adaptively adjusts the step-size according to the value of the estimated gradient. The traditional metaheuristic optimization algorithm primarily uses pre-defined rules to calculate the step-size and require manual tuning of hyper-parameters.
3) The proposed ADAM update-rules help in avoiding the slow convergence behavior in case of a narrow valley in the objective function. Therefore, the proposed update-rules show smoother convergence behavior as compared to traditional algorithms.
The rest of the paper is organized as follows: Section II presents the details of the BAS-ADAM algorithm, Section III presents the benchmark optimization problems we used in this paper, Section IV presents the experimental results, and Section V concludes the paper.
In this section, we will describe the technical details and mathematical formulation of the BAS-ADAM algorithm and outline its steps.
Here we will describe the BAS algorithm [52] and present the gradient estimation model. Consider the optimization problem:
maxx f(x) | (1) |
where
xb1=xt+ηt→b1xb2=xt+ηt→b2⋮xbm=xt+ηt→bm | (2) |
where
After creating the set
F={f(x):x∈X}={f(xb1),f(xb2),...,f(xbm)}. |
Now we will formulate the strategy to estimate the gradient direction using the element of setting
Fl={f:f⊂F∧f∈minkF}Fh={f:f⊂F∧f∈maxkF} | (3) |
where notation
Xl={x:x⊂X∧f(x)∈Fl}Xh={x:x⊂X∧f(x)∈Fh}. | (4) |
For a beetle analogy,
xl=∑x∈Xlxkxh=∑x∈Xhxk. | (5) |
The vectors
˜∇t=xh−xl | (6) |
where
Based on the value of estimated gradient
xnew=xt+δt(˜∇t) | (7) |
where
δt(˜∇t)=δ0ˆmt√ˆvt | (8) |
where
mt=β1mt−1+(1−β1)sign(f(xr)−f(xl))˜∇tvt=β2vt−1+(1−β2)˜∇2t. | (9) |
The factor of
ˆmt=mt1−βt1ˆvt=vt1−βt2. | (10) |
By putting the values of
xt+1={xnew,if f(xnew)>fbestxt,otherwise. | (11) |
The presented BAS-ADAM algorithm is given in Algorithm 1; it can be summarized as follow:
Algorithm 1: BAS-ADAM algorithm
Step 1: Begin with an random starting point
Step 2: Generate a set of
Step 3: Calculate the set for location of antennae fibres,
Step 4: Calculate the set of the objective function values
Step 5: Calculate the sets
Step 6: Calculate the set
Step 7: Calculate the centroid
Step 8:
Step 9: Calculate the first and second moments of estimated gradient using (9).
Step 10: Calculate the unbiased moments using (10).
Step 11: Determine candidate for update point
Step 12: The value of
Step 13: If
Now we will calculate the computational complexity of the BAS-ADAM. As demonstrated in [21] that the original BAS algorithm has a complexity of
This section presents the validation methodology used to test the efficiency, convergence, and efficacy of the proposed BAS-ADAM algorithm. We selected a set of three benchmark problems. The first is to search for the optimum value of Michalewicz function [79]. The second is a linear regression optimization problem, and the 3rd problem involves training of a single hidden layer neural network.
Algorithm 1: BAS-ADAM optimizer
Input: An objective function
Output: Optimal solution
while
Generate a set of
Calculate the set,
Calculate the set,
Select
Using
Calculate the centroids
Estimate the gradient direction,
Calculate the first moment
Correct the biasness in the calculated moments using (10).
Calculate a condidate for update-value
if
else
end
t
end
Michalewicz function [79] is
M(x)=−d∑i=1sin(xi)sin(ix2ipi)2m | (12) |
where
x∗=minx M(x) ≡ maxx −M(x). | (13) |
We used
For Michalewicz function, the objective function was fixed, and the optimum value is already known in the literature. However, our second problem involves solving a linear regression optimization problem in which the objective function is defined using randomly generated data points. Since linear regression is essentially the least-square optimization problem; therefore, the objective function is convex. The linear regression optimization problem can be defined like this: given a vector of
y=θnxn+θn−1xn−1+⋯+θ1x1+θ0=n∑i=0θixi | (14) |
where
y=θTx |
where
If we are provided with a set of
R(θ)=[y1−θTx1, y2−θTx2, ..., ym−θTxm]T |
based on this vector, we can define a cost function based on square-sum of the residual values as
C(θ)=||R(θ)||2=m∑i=0(yi−θTxi)2. |
Therefore, linear regression can be written as the following optimization problem
θ∗=minθ C(θ) ≡ maxθ (−C(θ)). | (15) |
During experiments, we used
As the third benchmark problem, we choose the training of a hidden layer neural network with nonlinear activation functions. Most of the metaheuristic algorithms show poor performance when it comes to the training of a machine learning model because these models are highly nonlinear and contain a large number of parameters. Without the information about the gradient direction, searching a high dimensional space becomes very challenging. Since our metaheuristic algorithm involves a rough estimation of the gradient direction, it is expected to show better performance for high dimensional search spaces. Additionally, we used the ADAM algorithm to adjust step size adaptively, therefore the proposed algorithm also does an excellent job in avoiding the local minimum.
It is known from machine learning literature that hidden layer neural networks are excellent in end to end learning [80], i.e., they can model an unknown process by just observing the input-output data points. They can model an arbitrary nonlinear process without needing an apriori mathematical model. Therefore we used the point cloud of Fig. 3 same as linear regression. However, this time, instead of using a linear model of (14), we used a neural network. It is expected that the neural network will provide a better fit to the point cloud and therefore produce a lower value for the cost function. The architecture of the neural network we used in this paper has two inputs, ten hidden neurons, and one output. Now we will derive the objective function for this optimization problem. Let
o1=ϕ(xTW1+b1)y=o1W2+b2 | (16) |
where
W1=[w(1)11w(1)12w(1)13…w(1)1kw(1)21w(1)22w(1)13…w(1)2k⋮⋮⋮⋱⋮w(1)n1w(1)n2w(1)13…w(1)nk]W2=[w(2)11w(2)21w(2)31…w(2)k1]T |
and,
b1=[b(1)1b(1)2b(1)3…b(1)k],b2=[b(2)1]. |
Now let us consider a set of
C(W1,W2,b1,b2)=m∑i=0(yi−ϕ(xT1W1+b1)W2−b2)2. |
Using the cost function
θ∗=minθ C(θ) ≡ maxθ(−C(θ)). | (17) |
In our experiments we used a neural network with 10 hidden neurons, 2 inputs and 1 output. There are a total of 41 trainable parameters in such a neural network. In summary, the problem 3 have these final values for the optimization problem:
Now we will present and discuss the experimental results for the validation problems presented in Sections III-A, III-B and III-C. Additionally, we compared the performance of the proposed BAS-ADAM algorithm with the PSO and original BAS algorithm. For the PSO algorithm, we used its implementation provided by MATLAB’s global optimization toolbox [81], whereas for BAS-ADAM and original BAS, we wrote the code in MATLAB.
For Michalewicz function, we used
The performance of the BAS-ADAM, along with PSO and original BAS, is shown in Fig. 4. The Fig. 4(a) shows the decrease in value of residual error with increase in iterations. The plot of error residual clearly shows that the BAS-ADAM shows the best convergence performance as compared to the other two optimization algorithms. The optimal point and minimum value reached by the BAS-ADAM comes quite close to the one given in the literature. BAS-ADAM is able to reach the theoretical minimum value of
The second validation problem involves solving the linear regression optimization problem given in (15). It is a convex optimization problem with a single global optimum and no local optimum. We tuned the parameters for BAS and BAS-ADAM until we reached the best performance.
The experimental results are shown in Fig. 5. The plot in Fig. 5(a) shows the decrease in value of error residual with iterations. The plot shows that BAS-ADAM shows the final rate of convergence, followed by PSO and original BAS. However, the final reached by original BAS,
Among the three validation problems, the training of a hidden layer neural network is the most challenging one. As mentioned in Section III-C, we used a neural network with ten hidden neurons to model the point cloud in Fig. 3. The topology of the neural network includes two inputs, one fully connected hidden layer with ten neurons and one output.
Fig. 6 shows the experimental results for training the neural network using the three optimization algorithms. The plot of error residuals is shown in Fig. 6(a). According to the plot, the BAS-ADAM shows the best performance, followed by BAS and PSO. The BAS-ADAM is able to reach the objective function value of around 6.5, i.e.,
In this paper, we presented a robust nature-inspired metaheuristic algorithm, called BAS-ADAM. The proposed algorithm is inspired by the BAS algorithm and improves its performance for the objective function, having very steep valleys. We achieve this by using the adaptive moment estimation (ADAM) technique widely applied in gradient-based optimization algorithms. The proposed algorithm estimated a gradient direction at each iteration using the antennae fibers of the beetle. The estimated gradient is then used as input to the ADAM algorithm, which computes its first and second moments. These moments are then used to adjust the step size for the next iterations. This technique offers a significant advantage as compared to original BAS because it automatically adapts the step size for each dimension according to the shape of the objective function. Therefore the BAS-ADAM shows much smoother convergence performance near the optimum points, as shown in the experiments. For original BAS, the step size is calculated using exponential decay rule and independent of the shape of the objective function. It is demonstrated through extensive experiments that the proposed algorithm offers fast and robust convergence as compared to other metaheuristic optimization algorithms. We selected three benchmark optimization problems to test the performance of the proposed algorithm. We repeated the same experiments with original BAS and PSO optimizers to present the comparative results. The experimental results show that, on average, BAS-ADAM takes fewer iterations and able to reach better optimum values as compared to original BAS and PSO. Additionally, we trained a hidden layer neural network, which shows the potential of the proposed algorithm in real-time machine learning applications.
Potential future research direction includes analyzing the performance of other step-size update rules, e.g., momentum, AdaGrad, and AdaDelta, as compared to ADAM. Additionally, extending the concept of gradient estimation to other metaheuristic optimization algorithms such as PSO, differential evolution, and different swarm-based algorithms will also constitute an important research direction. Similarly, formulating a strategy to improve the robustness of gradient estimation while using a small number of particles will also be an interesting topic to be explored.
[1] |
H. Q. Wang, P. X. Liu, X. J. Xie, X. P. Liu, T. Hayat, and F. E. Alsaadi, “Adaptive fuzzy asymptotical tracking control of nonlinear systems with unmodeled dynamics and quantized actuator,” Inf. Sci., DOI: 10.1016/j.ins.2018.04.011
|
[2] |
C. G. Yang, J. Luo, C. Liu, M. Li, and S. L. Dai, “Haptics electromyography perception and learning enhanced intelligence for teleoperated robot,” IEEE Trans. Autom. Sci. Eng., vol. 16, no. 4, pp. 1512–1521, Oct. 2019. doi: 10.1109/TASE.2018.2874454
|
[3] |
X. Luo, H. Wu, H. Q. Yuan, and M. C. Zhou, “Temporal pattern-aware QoS prediction via biased non-negative latent factorization of tensors,” IEEE Trans. Cybern., DOI: 10.1109/TCYB.2019.2903736
|
[4] |
L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large-scale machine learning,” SIAM Rev., vol. 60, no. 2, pp. 223–311, May 2018. doi: 10.1137/16M1080173
|
[5] |
S. Theodoridis, Machine Learning: A Bayesian and Optimization Perspective. San Diego: Academic Press, 2015.
|
[6] |
T. Liu, B. Tian, Y. F. Ai, L. Li, D. P. Cao, and F. Y. Wang, “Parallel reinforcement learning: a framework and case study,” IEEE/CAA J. Autom. Sinica, vol. 5, no. 4, pp. 827–835, Jul. 2018. doi: 10.1109/JAS.2018.7511144
|
[7] |
H. Q. Wang, S. W. Liu, and X. B. Yang, “Adaptive neural control for non-strict-feedback nonlinear systems with input delay,” Inf. Sci., vol. 514, pp. 605–616, Apr. 2020. doi: 10.1016/j.ins.2019.09.043
|
[8] |
X. Luo, Z. G. Liu, S. Li, M. S. Shang, and Z. D. Wang, “A fast non-negative latent factor model based on generalized momentum method,” IEEE Trans. Syst., Man, Cybern.: Syst., DOI: 10.1109/TSMC.2018.2875452
|
[9] |
D. P. Kingma and J. Ba, “Adam: a method for stochastic optimization,” arXiv: 1412.6980, 2014.
|
[10] |
S. Ruder, “An overview of gradient descent optimization algorithms,” arXiv: 1609.04747, 2016.
|
[11] |
Y. Shi and Y. N. Zhang, “Solving future equation systems using integral-type error function and using twice ZNN formula with disturbances suppressed,” J. Franklin Ins., vol. 356, no. 4, pp. 2130–2152, Mar. 2019. doi: 10.1016/j.jfranklin.2018.11.026
|
[12] |
L. Xiao, S. Li, F. J. Lin, Z. G. Tan, and A. H. Khan, “Zeroing neural dynamics for control design: comprehensive analysis on stability, robustness, and convergence speed,” IEEE Trans. Ind. Informatics, vol. 15, no. 5, pp. 2605–2616, May 2019. doi: 10.1109/TII.2018.2867169
|
[13] |
Y. N. Zhang, Z. Y. Qi, B. B. Qiu, M. Yang, and M. L. Xiao, “Zeroing neural dynamics and models for various time-varying problems solving with ZLSF models as minimization-type and Euler-type special cases[research frontier],” IEEE Comput. Intell. Mag., vol. 14, no. 3, pp. 52–60, Aug. 2019. doi: 10.1109/MCI.2019.2919397
|
[14] |
Y. N. Zhang, Z. Y. Qi, M. Yang, J. J. Guo, and H. C. Huang, “Step-width theoretics and numerics of four-point general DTZN model for future minimization using jury stability criterion,” Neurocomputing, vol. 357, pp. 231–239, Sept. 2019. doi: 10.1016/j.neucom.2019.04.054
|
[15] |
R. Johnson and T. Zhang, “Accelerating stochastic gradient descent using predictive variance reduction,” in Proc. 26th Int. Conf. Neural Information Processing Systems, Red Hook, USA, 2013, pp. 315–323.
|
[16] |
X. Luo, D. X. Wang, M. C. Zhou, and H. Q. Yuan, "Latent factor-based recommenders relying on extended stochastic gradient descent algorithms," IEEE Trans. Syst., Man, Cybern.: Syst., DOI: 10.1109/TSMC.2018.2884191
|
[17] |
J. D. Lee, M. Simchowitz, M. I. Jordan, and B. Recht, “Gradient descent only converges to minimizers,” in Proc. 29th Annu. Conf. Learning Theory, 2016, pp. 1246–1257.
|
[18] |
W. N. Chen, J. Zhang, H. S. H. Chung, W. L. Zhong, W. G. Wu, and Y. H. Shi, “A novel set-based particle swarm optimization method for discrete optimization problems,” IEEE Trans. Evol. Comput., vol. 14, no. 2, pp. 278–300, Apr. 2010. doi: 10.1109/TEVC.2009.2030331
|
[19] |
S. Ling, H. Q. Wang, and P. X. Liu, “Adaptive fuzzy dynamic surface control of flexible-joint robot systems with input saturation,” IEEE/CAA J. Autom. Sinica, vol. 6, no. 1, pp. 97–107, Jan. 2019. doi: 10.1109/JAS.2019.1911330
|
[20] |
H. Nguyen-Xuan, G. Liu, C. a. Thai-Hoang, and T. Nguyen-Thoi, “An edge-based smoothed finite element method (ES-FEM) with stabilized discrete shear gap technique for analysis of reissner-mindlin plates,” Computer Methods in Applied Mechanics and Engineering, vol. 199, no. 9–12, pp. 471–489, Jan. 2010. doi: 10.1016/j.cma.2009.09.001
|
[21] |
A. H. Khan, S. Li, and X. Luo, “Obstacle avoidance and tracking control of redundant robotic manipulator: an RNN based metaheuristic approach,” IEEE Trans. Ind. Informatics, DOI: 10.1109/TII.2019.2941916
|
[22] |
Y. Zhou, L. J. Kong, Z. Y. Wu, S. P. Liu, Y. Q. Cai, and Y. Liu, “Ensemble of multi-objective metaheuristic algorithms for multi-objective unconstrained binary quadratic programming problem,” Appl. Soft Comput., vol. 81, pp. 105485, Aug. 2019. doi: 10.1016/j.asoc.2019.105485
|
[23] |
J. A. Parejo, A. Ruiz-Cortes, S. Lozano, and P. Fernandez, “Metaheuristic optimization frameworks: a survey and benchmarking,” Soft Comput., vol. 16, no. 3, pp. 527–561, Mar. 2012. doi: 10.1007/s00500-011-0754-8
|
[24] |
C. Blum, A. Roli, and M. Sampels, Hybrid Metaheuristics: An Emerging Approach to Optimization. Heidelberg, Germany: Springer, 2008.
|
[25] |
H. Q. Wang, P. X. Liu, X. D. Zhao, and X. P. Liu, “Adaptive fuzzy finite-time control of nonlinear systems with actuator faults,” IEEE Trans. Cybern., DOI: 10.1109/TCYB.2019.2902868
|
[26] |
O. Roeva and T. Slavov, “Pid controller tuning based on metaheuristic algorithms for bioprocess control,” Biotechnol. Biotechnol. Equip., vol. 26, no. 5, pp. 3267–3277, Apr. 2012. doi: 10.5504/BBEQ.2012.0065
|
[27] |
A. Song, W. N. Chen, T. L. Gu, H. Q. Yuan, S. Kwong, and J. Zhang, “Distributed virtual network embedding system with historical archives and set-based particle swarm optimization,” IEEE Trans. Syst., Man, Cybern.: Syst., DOI: 10.1109/TSMC.2018.2884523
|
[28] |
C. G. Yang, G. Z. Peng, Y. N. Li, R. X. Cui, L. Cheng, and Z. J. Li, “Neural networks enhanced adaptive admittance control of optimized robot-environment interaction,” IEEE Trans. Cybern., vol. 49, no. 7, pp. 2568–2579, Jul. 2019. doi: 10.1109/TCYB.2018.2828654
|
[29] |
L. Cheng, W. C. Liu, C. G. Yang, T. W. Huang, Z. G. Hou, and M. Tan, “A neural-network-based controller for piezoelectric-actuated stick-slip devices,” IEEE Trans. Ind. Electron., vol. 65, no. 3, pp. 2598–2607, Mar. 2018. doi: 10.1109/TIE.2017.2740826
|
[30] |
H. Y. Liu, L. Cheng, M. Tan, and Z. G. Hou, “Containment control of general linear multi-agent systems with multiple dynamic leaders: a fast sliding mode based approach,” IEEE/CAA J. Autom. Sinica, vol. 1, no. 2, pp. 134–140, Apr. 2014. doi: 10.1109/JAS.2014.7004542
|
[31] |
H. Y. Liu, M. C. Zhou, and Q. Liu, “An embedded feature selection method for imbalanced data classification,” IEEE/CAA J. Autom. Sinica, vol. 6, no. 3, pp. 703–715, May 2019. doi: 10.1109/JAS.2019.1911447
|
[32] |
X. Luo, Z. D. Wang, and M. S. Shang, “An instance-frequency-weighted regularization scheme for non-negative latent factor analysis on high-dimensional and sparse data,” IEEE Trans. Syst., Man, Cybern.: Syst., DOI: 10.1109/TSMC.2019.2930525
|
[33] |
X. S. Yang, Nature-Inspired Metaheuristic Algorithms, Beckington, UK: Luniver Press, 2008.
|
[34] |
J. Krause, J. Cordeiro, R. S. Parpinelli, and H. S. Lopes, “A survey of swarm algorithms applied to discrete optimization problems,” in Swarm Intelligence and Bio-Inspired Computation, X. S. Yang, Z. H. Cui, R. B. Xiao, A. H. Gandomi, and M. Karamanoglu, Eds. London, UK: Elsevier, 2013, pp. 169–191.
|
[35] |
H. Shayanfar and F. S. Gharehchopogh, “Farmland fertility: a new metaheuristic algorithm for solving continuous optimization problems,” Appl. Soft Comput., vol. 71, pp. 728–746, Oct. 2018. doi: 10.1016/j.asoc.2018.07.033
|
[36] |
E. Wari and W. H. Zhu, “A survey on metaheuristics for optimization in food manufacturing industry,” Appl. Soft Comput., vol. 46, pp. 328–343, Sep. 2016. doi: 10.1016/j.asoc.2016.04.034
|
[37] |
X. Luo, M. C. Zhou, S. Li, Z. H. You, Y. N. Xia, and Q. S. Zhu, “A nonnegative latent factor model for large-scale sparse matrices in recommender systems via alternating direction method,” IEEE Trans. Neural Networks Learn. Syst., vol. 27, no. 3, pp. 579–592, Mar. 2016. doi: 10.1109/TNNLS.2015.2415257
|
[38] |
X. Luo, M. C. Zhou, Y. N. Xia, and Q. S. Zhu, “An efficient non-negative matrix-factorization-based approach to collaborative filtering for recommender systems,” IEEE Trans. Ind. Informatics, vol. 10, no. 2, pp. 1273–1284, May 2014. doi: 10.1109/TII.2014.2308433
|
[39] |
J. Q. Li, H. Y. Sang, Q. K. Pan, P. Y. Duan, and K. Z. Gao, “Solving multi-area environmental/economic dispatch by Pareto-based chemical-reaction optimization algorithm,” IEEE/CAA J. Autom. Sinica, vol. 6, no. 5, pp. 1240–1250, Sep. 2019. doi: 10.1109/JAS.2017.7510454
|
[40] |
P. J. Angeline, G. M. Saunders, and J. B. Pollack, “An evolutionary algorithm that constructs recurrent neural networks,” IEEE Trans. Neural Networks, vol. 5, no. 1, pp. 54–65, Jan. 1994. doi: 10.1109/72.265960
|
[41] |
W. S. Gao and C. Shao, “Iterative dynamic diversity evolutionary algorithm for constrained optimization,” Acta Autom. Sinica, vol. 40, no. 11, pp. 2469–2479, Nov. 2014. doi: 10.1016/S1874-1029(14)60398-0
|
[42] |
S. Gupta, K. Deep, A. A. Heidari, H. Moayedi, and H. L. Chen, “Harmonized salp chain-built optimization,” Eng. Comput., DOI: 10.1007/s00366-019-00871-5
|
[43] |
D. E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning. Reading, USA: Addion Wesley, 1989.
|
[44] |
F. P. Such, V. Madhavan, E. Conti, J. Lehman, K. O. Stanley, and J. Clune, “Deep neuroevolution: genetic algorithms are a competitive alternative for training deep neural networks for reinforcement learning,” arXiv: 1712.06567, 2017.
|
[45] |
J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proc. IEEE Int. Conf. Neural Networks, Perth, Australia, 2001, pp. 1942–1948.
|
[46] |
J. J. Wang and T. Kumbasar, “Parameter optimization of interval type-2 fuzzy neural networks based on PSO and BBBC methods,” IEEE/CAA J. Autom. Sinica, vol. 6, no. 1, pp. 247–257, Jan. 2019. doi: 10.1109/JAS.2019.1911348
|
[47] |
H. Shi, L. Wang, and T. G. Chu, “Swarming behavior of multi-agent systems,” J. Control Theory Appl., vol. 2, no. 4, pp. 313–318, Nov. 2004. doi: 10.1007/s11768-004-0034-6
|
[48] |
M. G. Hinchey, R. Sterritt, and C. Rouff, “Swarms and swarm intelligence,” Computer, vol. 40, no. 4, pp. 111–113, Apr. 2007. doi: 10.1109/MC.2007.144
|
[49] |
X. Feng, Y. B. Wang, H. Q. Yu, and F. Luo, “A novel intelligence algorithm based on the social group optimization behaviors,” IEEE Trans. Syst.,Man,Cybern.:Syst., vol. 48, no. 1, pp. 65–76, Jan. 2018. doi: 10.1109/TSMC.2016.2586973
|
[50] |
M. Dorigo and G. Di Caro, “Ant colony optimization: a new meta-heuristic,” in Proc. Congr. Evolutionary Computation, Washington, USA, 1999, pp. 1470–1477.
|
[51] |
M. Neshat, G. Sepidnam, M. Sargolzaei, and A. N. Toosi, “Artificial fish swarm algorithm: a survey of the state-of-the-art, hybridization, combinatorial and indicative applications,” Artif. Intell. Rev., vol. 42, no. 4, pp. 965–997, Dec. 2014. doi: 10.1007/s10462-012-9342-2
|
[52] |
X. Y. Jiang and S. Li, “Bas: beetle antennae search algorithm for optimization problems,” arXiv: 1710.10724, 2017.
|
[53] |
X. Y. Jiang and S. Li, “Beetle antennae search without parameter tuning (BAS-WPT) for multi-objective optimization,” arXiv: 1711.02395, 2017.
|
[54] |
X. S. Yang and S. Deb, “Engineering optimisation by cuckoo search,” arXiv: 1005.2908, 2010.
|
[55] |
J. Zhao, S. X. Liu, M. C. Zhou, X. W. Guo, and L. Qi, “Modified cuckoo search algorithm to solve economic power dispatch optimization problems,” IEEE/CAA J. Autom. Sinica, vol. 5, no. 4, pp. 794–806, Jul. 2018. doi: 10.1109/JAS.2018.7511138
|
[56] |
A. R. Mehrabian and C. Lucas, “A novel numerical optimization algorithm inspired from weed colonization,” Ecological Informatics, vol. 1, no. 4, pp. 355–366, Dec. 2006. doi: 10.1016/j.ecoinf.2006.07.003
|
[57] |
S. Nakrani and C. Tovey, “On honey bees and dynamic server allocation in internet hosting centers,” Adapt. Behav., vol. 12, no. 3–4, pp. 223–240, Dec. 2004. doi: 10.1177/105971230401200308
|
[58] |
D. Karaboga and B. Akay, “A survey: algorithms simulating bee swarm intelligence,” Artif. Intell. Rev., vol. 31, no. 1–4, pp. 61–85, Jun. 2009. doi: 10.1007/s10462-009-9127-4
|
[59] |
X. S. Yang, "Firefly algorithms for multimodal optimization," in Proc. 5th Int. Symp. Stochastic Algorithms: Foundations and Applications, Sapporo, Japan, 2009, pp. 169–178.
|
[60] |
S. Mirjalili, S. M. Mirjalili, and A. Lewis, “Grey wolf optimizer,” Adv. Eng. Softw., vol. 69, pp. 46–61, Mar. 2014. doi: 10.1016/j.advengsoft.2013.12.007
|
[61] |
S. Gupta and K. Deep, “A novel random walk grey wolf optimizer,” Swarm Evol. Comput., vol. 44, pp. 101–112, Feb. 2019. doi: 10.1016/j.swevo.2018.01.001
|
[62] |
S. Gupta and K. Deep, “An opposition-based chaotic grey wolf optimizer for global optimisation tasks,” J. Exp. Theor. Artif. Intell., vol. 31, no. 5, pp. 751–779, 2019. doi: 10.1080/0952813X.2018.1554712
|
[63] |
S. Gupta and K. Deep, “Improved sine cosine algorithm with crossover scheme for global optimization,” Knowl.-Based Syst., vol. 165, pp. 374–406, Feb. 2019. doi: 10.1016/j.knosys.2018.12.008
|
[64] |
S. Gupta and K. Deep, “A hybrid self-adaptive sine cosine algorithm with opposition based learning,” Expert Syst. Appl., vol. 119, pp. 210–230, Apr. 2019. doi: 10.1016/j.eswa.2018.10.050
|
[65] |
M. S. Shang, X. Luo, Z. G. Liu, J. Chen, Y. Yuan, and M. C. Zhou, “Randomized latent factor model for high-dimensional and sparse matrices from industrial applications,” IEEE/CAA J. Autom. Sinica, vol. 6, no. 1, pp. 131–141, Jan. 2019. doi: 10.1109/JAS.2018.7511189
|
[66] |
Z. Y. Zhu, Z. Y. Zhang, W. S. Man, X. Q. Tong, J. Z. Qiu, and F. F. Li, “A new beetle antennae search algorithm for multi-objective energy management in microgrid,” in Proc. 13th IEEE Conf. Industrial Electronics and Applications, Wuhan, China, 2018, pp. 1599–1603.
|
[67] |
X. Y. Yin and Y. Ma, “Aggregation service function chain mapping plan based on beetle antennae search algorithm,” in Proc. 2nd Int. Conf. Telecommunications and Communication Engineering, Beijing, China, 2018, pp. 225–230.
|
[68] |
X. M. Lin, Y. F. Liu, and Y. L. Wang, “Design and research of DC motor speed control system based on improved BAS,” in Proc. Chinese Automation Congr., Xi’an, China, 2018, pp. 3701–3705.
|
[69] |
Y. T. Sun, J. F. Zhang, G. C. Li, Y. H. Wang, J. B. Sun, and C. Jiang, “Optimized neural network using beetle antennae search for predicting the unconfined compressive strength of jet grouting coalcretes,” Int. J. Numer. Anal. Methods Geomech., vol. 43, no. 4, pp. 801–813, Mar. 2019. doi: 10.1002/nag.2891
|
[70] |
Q. Wu, X. D. Shen, Y. Z. Jin, Z. Y. Chen, S. Li, A. H. Khan, and D. C. Chen, “Intelligent beetle antennae search for UAV sensing and avoidance of obstacles,” Sensors, vol. 19, no. 8, pp. 1758, Apr. 2019. doi: 10.3390/s19081758
|
[71] |
M. J. Lin and Q. H. Li, “A hybrid optimization method of beetle antennae search algorithm and particle swarm optimization,” DEStech Trans. Engineering and Technology Research, 2018.
|
[72] |
Q. Wu, H. Lin, Y. Z. Jin, Z. Y. Chen, S. Li, and D. C. Chen, “A new fallback beetle antennae search algorithm for path planning of mobile robots with collision-free capability,” Soft Comput., vol. 24, no. 3, pp. 2369–2380, Feb. 2020. doi: 10.1007/s00500-019-04067-3
|
[73] |
Y. Q. Fan, J. P. Shao, and G. T. Sun, “Optimized PID controller based on beetle antennae search algorithm for electro-hydraulic position servo control system,” Sensors, vol. 19, no. 12, pp. 2727, Jun. 2019. doi: 10.3390/s19122727
|
[74] |
S. Xie, X. M. Chu, M. Zheng, and C. G. Liu, “Ship predictive collision avoidance method based on an improved beetle antennae search algorithm,” Ocean Eng., vol. 192, pp. 106542, Nov. 2019. doi: 10.1016/j.oceaneng.2019.106542
|
[75] |
J. H. Yang and Z. R. Peng, “Beetle-swarm evolution competitive algorithm for bridge sensor optimal placement in SHM,” IEEE Sens. J., DOI: 10.1109/JSEN.2019.2934996.
|
[76] |
N. Qian, “On the momentum term in gradient descent learning algorithms,” Neural Networks, vol. 12, no. 1, pp. 145–151, Jan. 1999. doi: 10.1016/S0893-6080(98)00116-6
|
[77] |
J. Duchi, E. Hazan, and Y. Singer, “Adaptive subgradient methods for online learning and stochastic optimization,” J. Machine Learning Research, vol. 12, pp. 2121–2159, Jun. 2011.
|
[78] |
M. D. Zeiler, “ADADELTA: an adaptive learning rate method,” arXiv: 1212.5701, 2012.
|
[79] |
A. P. Engelbrecht, “Fitness function evaluations: a fair stopping condition?” in Proc. IEEE Symp. Swarm Intelligence, Orlando, USA, 2014, pp. 1–8.
|
[80] |
I. Sutskever, O. Vinyals, and Q. V. Le, “Sequence to sequence learning with neural networks,” arXiv: 1409.3215, 2014.
|
[81] |
Mathworks, MATLAB: Global Optimization Toolbox 2018b. The Mathworks Inc., 2018.
|
[1] | Jiajun Cheng, Haonan Chen, Zhirui Xue, Yulong Huang, Yonggang Zhang. An Online Exploratory Maximum Likelihood Estimation Approach to Adaptive Kalman Filtering[J]. IEEE/CAA Journal of Automatica Sinica, 2025, 12(1): 228-254. doi: 10.1109/JAS.2024.125001 |
[2] | Sheng Qi, Rui Wang, Tao Zhang, Weixiong Huang, Fan Yu, Ling Wang. Enhancing Evolutionary Algorithms With Pattern Mining for Sparse Large-Scale Multi-Objective Optimization Problems[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(8): 1786-1801. doi: 10.1109/JAS.2024.124548 |
[3] | Fei Ming, Wenyin Gong, Yaochu Jin. Even Search in a Promising Region for Constrained Multi-Objective Optimization[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(2): 474-486. doi: 10.1109/JAS.2023.123792 |
[4] | Jing Zhou, Jun Shang, Tongwen Chen. Cybersecurity Landscape on Remote State Estimation: A Comprehensive Review[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(4): 851-865. doi: 10.1109/JAS.2024.124257 |
[5] | Kailong Liu, Qiao Peng, Yuhang Liu, Naxin Cui, Chenghui Zhang. Explainable Neural Network for Sensitivity Analysis of Lithium-ion Battery Smart Production[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(9): 1944-1953. doi: 10.1109/JAS.2024.124539 |
[6] | Meng Wang, Jiawan Zhang, Jiayi Ma, Xiaojie Guo. Cas-FNE: Cascaded Face Normal Estimation[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(12): 2423-2434. doi: 10.1109/JAS.2024.124899 |
[7] | Xinli Shi, Xiangping Xu, Guanghui Wen, Jinde Cao. Fixed-Time Gradient Flows for Solving Constrained Optimization: A Unified Approach[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(8): 1849-1864. doi: 10.1109/JAS.2023.124089 |
[8] | Chi Ma, Dianbiao Dong. Finite-time Prescribed Performance Time-Varying Formation Control for Second-Order Multi-Agent Systems With Non-Strict Feedback Based on a Neural Network Observer[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(4): 1039-1050. doi: 10.1109/JAS.2023.123615 |
[9] | Dingxin He, HaoPing Wang, Yang Tian, Yida Guo. A Fractional-Order Ultra-Local Model-Based Adaptive Neural Network Sliding Mode Control of n-DOF Upper-Limb Exoskeleton With Input Deadzone[J]. IEEE/CAA Journal of Automatica Sinica, 2024, 11(3): 760-781. doi: 10.1109/JAS.2023.123882 |
[10] | Zhenyu Lei, Shangce Gao, Zhiming Zhang, Haichuan Yang, Haotian Li. A Chaotic Local Search-Based Particle Swarm Optimizer for Large-Scale Complex Wind Farm Layout Optimization[J]. IEEE/CAA Journal of Automatica Sinica, 2023, 10(5): 1168-1180. doi: 10.1109/JAS.2023.123387 |
[11] | Zhuoxuan Li, Iakov Korovin, Xinli Shi, Sergey Gorbachev, Nadezhda Gorbacheva, Wei Huang, Jinde Cao. A Data-Driven Rutting Depth Short-Time Prediction Model With Metaheuristic Optimization for Asphalt Pavements Based on RIOHTrack[J]. IEEE/CAA Journal of Automatica Sinica, 2023, 10(10): 1918-1932. doi: 10.1109/JAS.2023.123192 |
[12] | Zhijia Zhao, Jian Zhang, Shouyan Chen, Wei He, Keum-Shik Hong. Neural-Network-Based Adaptive Finite-Time Control for a Two-Degree-of-Freedom Helicopter System With an Event-Triggering Mechanism[J]. IEEE/CAA Journal of Automatica Sinica, 2023, 10(8): 1754-1765. doi: 10.1109/JAS.2023.123453 |
[13] | Yu Cao, Jian Huang. Neural-Network-Based Nonlinear Model Predictive Tracking Control of a Pneumatic Muscle Actuator-Driven Exoskeleton[J]. IEEE/CAA Journal of Automatica Sinica, 2020, 7(6): 1478-1488. doi: 10.1109/JAS.2020.1003351 |
[14] | Yan Ma, Xiuwen Zhou, Bingsi Li, Hong Chen. Fractional Modeling and SOC Estimation of Lithium-ion Battery[J]. IEEE/CAA Journal of Automatica Sinica, 2016, 3(3): 281-287. |
[15] | Xiaowei Feng, Xiangyu Kong, Hongguang Ma. Coupled Cross-correlation Neural Network Algorithm for Principal Singular Triplet Extraction of a Cross-covariance Matrix[J]. IEEE/CAA Journal of Automatica Sinica, 2016, 3(2): 147-156. |
[16] | Xiaojun Tang, Jie Chen. Direct Trajectory Optimization and Costate Estimation of Infinite-horizon Optimal Control Problems Using Collocation at the Flipped Legendre-Gauss-Radau Points[J]. IEEE/CAA Journal of Automatica Sinica, 2016, 3(2): 174-183. |
[17] | Xiaoming Sun, Shuzhi Sam Ge. Adaptive Neural Region Tracking Control of Multi-fully Actuated Ocean Surface Vessels[J]. IEEE/CAA Journal of Automatica Sinica, 2014, 1(1): 77-83. |
[18] | Sen Wang, Ling Chen, Dongbing Gu, Huosheng Hu. Cooperative Localization of AUVs Using Moving Horizon Estimation[J]. IEEE/CAA Journal of Automatica Sinica, 2014, 1(1): 68-76. |
[19] | Yanqiong Zhang, Youcheng Lou, Yiguang Hong. An Approximate Gradient Algorithm for Constrained Distributed Convex Optimization[J]. IEEE/CAA Journal of Automatica Sinica, 2014, 1(1): 61-67. |
[20] | Qiming Zhao, Hao Xu, Sarangapani Jagannathan. Near Optimal Output Feedback Control of Nonlinear Discrete-time Systems Based on Reinforcement Neural Network Learning[J]. IEEE/CAA Journal of Automatica Sinica, 2014, 1(4): 372-384. |