【启发式算法】白鲸优化算法【附python实现代码】
写在前面：
首先感谢兄弟们的订阅，让我有创作的动力，在创作过程我会尽最大能力，保证作品的质量，如果有问题，可以私信我，让我们携手共进，共创辉煌。
路虽远，行则将至；事虽难，做则必成。只要有愚公移山的志气、滴水穿石的毅力，脚踏实地，埋头苦干，积跬步以至千里，就一定能够把宏伟目标变为美好现实。
1.介绍
白鲸优化算法(Beluga whale optimization, BWO)是2022年提出的一种元启发式优化算法，其灵感来源于白鲸的生活行为。白鲸以成年鲸的纯白色而闻名，是高度群居的动物，它们可以成群聚集，有2到25个成员，平均有10个成员。与其他元启发式方法类似，BWO包含探索阶段和开发阶段，此外该算法还模拟了生物界中存在的鲸落现象。
论文：
Zhong C, Li G, Meng Z. Beluga whale optimization: A novel nature-inspired metaheuristic algorithm[J]. Knowledge-Based Systems, 2022, 109215.
2.算法流程
BWO流程图：
2.1 种群初始化
白鲸优化算法是基于种群的机制，将每条白鲸假设为一个候选解，并在优化过程中进行不断更新，则种群初始化位置为：
2.2 探索阶段
2.3 开发阶段
开发阶段的灵感来自于白鲸的捕食行为，白鲸可以根据附近白鲸合作觅食和位置移动。假设可以通过 Levy 飞行策略捕捉猎物，其公式为：
2.4 鲸鱼坠落
为了模拟每次迭代中鲸鱼坠落行为，从种群中选择一个鲸鱼坠落的概率，来模拟群体中的变化。白鲸要么转移到其它地方，要么被击落并掉入深海。为保证种群数量不变，通过利用白鲸位置和鲸鱼坠落步长更新鲸鱼位置，公式如下：
式中，鲸鱼坠落的概率从初始迭代的 0.1 下降到最后一次迭代的 0.05 ，说明当白鲸在优化过程中更接近食物源时，白鲸的就会危险降低。
3.代码
BWO代码：
import numpy as np
import math
from copy import deepcopy
import matplotlib.pyplot as plt
'''
白鲸优化算法
参数
n：白鲸种群大小
tmax：最大迭代次数
lb：变量下限
ub：变量上限
nd：变量维数
fobj：目标函数
'''
# 白鲸优化算法
def bwo(n, tmax, lb, ub, nd, fobj):
# 记录每轮更新最优解
best_fval = []
# 位置矩阵
x = lb + np.random.random((n, nd)) * (ub - lb)
# 适应值矩阵
fobj_list = [fobj(x[i, ...]) for i in range(n)]
fx = [f() for f in fobj_list]
# 适应值拷贝矩阵
temp_fx = deepcopy(fx)
# 最优解和最优解值
xposbest = x[np.argmin(fx)]
fvalbest = min(fx)
best_fval.append(fvalbest)
# 循环更新
for t in range(1, tmax + 1):
# 位置拷贝矩阵
temp_x = deepcopy(x)
# 鲸落概率
wf = 0.1 - 0.05 * (t / tmax)
# 计算每只白鲸的平衡因子bf
b0 = np.random.random(n)
for b0_index in range(len(b0)):
while b0[b0_index] == 0:
b0[b0_index] = np.random.random()
bf = b0 * (1 - t / (2 * tmax))
# 更新每一只白鲸位置
for i in range(n):
# 探索阶段
if bf[i] > 0.5:
r1 = np.random.random()
r2 = np.random.random()
while r1 == 0:
r1 = np.random.random()
while r2 == 0:
r2 = np.random.random()
# 选择随机白鲸
r = np.random.randint(0, n)
while r == i:
r = np.random.randint(0, n)
pj = np.arange(nd)
np.random.shuffle(pj)
if nd <= n / 5:
temp_x[i, pj[0]] = x[i, pj[0]] + (x[r, pj[0]] - x[i, pj[0]]) * (1 + r1) * np.sin(2 * np.pi * r2)
temp_x[i, pj[1]] = x[i, pj[1]] + (x[r, pj[1]] - x[i, pj[1]]) * (1 + r1) * np.cos(2 * np.pi * r2)
else:
for j in range(int(nd / 2)):
temp_x[i, 2 * j] = x[i, 2 * j] + (x[r, pj[1]] - x[i, 2 * j]) * (1 + r1) * np.sin(2 * np.pi * r2)
temp_x[i, 2 * j + 1] = x[i, 2 * j + 1] + (x[r, pj[1]] - x[i, 2 * j + 1]) * (1 + r1) * np.cos(
2 * np.pi * r2)
# 开发阶段
else:
r3 = np.random.random()
r4 = np.random.random()
while r3 == 0:
r3 = np.random.random()
while r4 == 0:
r4 = np.random.random()
c1 = 2 * r4 * (1 - (t / tmax))
# 随机白鲸
r = np.random.randint(0, n)
while r == i:
r = np.random.randint(0, n)
beta = 3 / 2
delta = ((math.gamma(1 + beta) * np.sin((np.pi * beta) / 2))
/ (math.gamma((1 + beta) / 2) * beta * (2 ** ((beta - 1) / 2)))) ** (1 / beta)
u = np.random.randn(nd)
v = np.random.randn(nd)
lf = 0.05 * ((u * delta) / (abs(v) ** (1 / beta)))
temp_x[i, ...] = r3 * xposbest - r4 * x[i, ...] + c1 * lf * (x[r, ...] - x[i, ...])
# 处理超出边界值
flaglb = np.asarray(temp_x[i, ...] < lb).astype(np.int8)
flagub = np.asarray(temp_x[i, ...] > ub).astype(np.int8)
flag = flaglb + flagub
for k in range(nd):
if flag[k] == 0:
flag[k] = 1
else:
flag[k] = 0
temp_x[i, ...] = temp_x[i, ...] * flag + ub * flagub + lb * flaglb
result = fobj(temp_x[i, ...])
temp_fx[i] = result()
if temp_fx[i] < fx[i]:
x[i, ...] = temp_x[i, ...]
fx[i] = temp_fx[i]
# 鲸落
for i in range(n):
if bf[i] <= wf:
r5 = np.random.random()
r6 = np.random.random()
r7 = np.random.random()
while r5 == 0:
r5 = np.random.random()
while r6 == 0:
r6 = np.random.random()
while r7 == 0:
r7 = np.random.random()
c2 = 2 * wf * n
# 随机白鲸
r = np.random.randint(0, n)
xstep = (ub - lb) * (np.exp((-c2 * t) / tmax))
temp_x[i, ...] = r5 * x[i, ...] - r6 * x[r, ...] + r7 * xstep
# 处理超出边界值
flaglb = np.asarray(temp_x[i, ...] < lb).astype(np.int8)
flagub = np.asarray(temp_x[i, ...] > ub).astype(np.int8)
flag = flaglb + flagub
for m in range(nd):
if flag[m] == 0:
flag[m] = 1
else:
flag[m] = 0
temp_x[i, ...] = temp_x[i, ...] * flag + ub * flagub + lb * flaglb
result = fobj(temp_x[i, ...])
temp_fx[i] = result()
if temp_fx[i] < fx[i]:
x[i, ...] = temp_x[i, ...]
fx[i] = temp_fx[i]
# 更改最优值
temp_xposbest = x[np.argmin(fx)]
temp_fvalbest = min(fx)
if temp_fvalbest < fvalbest:
fvalbest = temp_fvalbest
xposbest = temp_xposbest
best_fval.append(fvalbest)
return xposbest, fvalbest, best_fval
# mu函数
def mu_fun(x, a, k, m):
if x > a:
return k * ((x - a) ** m)
elif -a <= x <= a:
return 0
elif x < a:
return k * ((-x - a) ** m)
# 求解适应值
def get_function_result(fname, *args):
result = -1
if fname == "f1":
def f1():
return sum(args[0] ** 2)
result = f1
if fname == "f2":
def f2():
sum1 = 0
sum2 = 1
for i in range(len(args[0])):
sum1 += abs(args[0][i])
sum2 *= abs(args[0][i])
return sum1 + sum2
result = f2
elif fname == "f3":
def f3():
sum1 = 0
for i in range(len(args[0])):
sum1 += abs(args[0][i]) ** (i + 1)
return sum1
result = f3
elif fname == "f4":
def f4():
sum1 = 0
sum2 = 0
for i in range(len(args[0])):
sum1 += args[0][i]
for j in range(len(args[0])):
sum2 += sum1 ** 2
return sum2
result = f4
elif fname == "f5":
def f5():
return np.max(abs(args[0]))
result = f5
elif fname == "f6":
def f6():
sum1 = 0
for i in range(len(args[0]) - 1):
sum1 += (100 * ((args[0][i + 1] - (args[0][i] ** 2)) ** 2) + ((args[0][i] - 1) ** 2))
return sum1
result = f6
elif fname == "f7":
def f7():
sum1 = 0
for i in range(len(args[0])):
sum1 += ((args[0][i] + 0.5) ** 2)
return sum1
result = f7
elif fname == "f8":
def f8():
sum1 = 0
for i in range(len(args[0])):
sum1 += (i + 1) * (args[0][i] ** 4)
return sum1 + np.random.random()
result = f8
elif fname == "f9":
def f9():
sum1 = 0
sum2 = 0
for i in range(len(args[0])):
sum1 += args[0][i] ** 2
sum2 += 0.5 * (i + 1) * args[0][i]
return sum1 + sum2 ** 2 + sum2 ** 4
result = f9
elif fname == "f10":
def f10():
sum1 = 0
for i in range(len(args[0])):
sum1 += args[0][i] * np.sin(np.sqrt(abs(args[0][i])))
return -sum1
result = f10
elif fname == "f11":
def f11():
sum1 = 0
sum2 = 0
for i in range(len(args[0])):
sum1 += np.sin(args[0][i]) ** 2
sum2 += args[0][i] ** 2
return 1 + sum1 - np.exp(-sum2)
result = f11
elif fname == "f12":
def f12():
sum1 = 0
for i in range(len(args[0])):
sum1 += (args[0][i] ** 4) - 16 * (args[0][i] ** 2) + 5 * args[0][i]
return 0.5 * sum1
result = f12
elif fname == "f13":
def f13():
sum1 = 0
for i in range(len(args[0])):
sum1 += (args[0][i] ** 2) - 10 * np.cos(2 * np.pi * args[0][i]) + 10
return sum1
result = f13
elif fname == "f14":
def f14():
sum1 = 0
sum2 = 0
for i in range(len(args[0])):
sum1 += args[0][i] ** 2
sum2 += np.cos(2 * np.pi * args[0][i])
return -20 * np.exp(-0.2 * np.sqrt(sum1 / len(args[0]))) - np.exp(sum2 / len(args[0])) + 20 + np.e
result = f14
elif fname == "f15":
def f15():
sum1 = 0
sum2 = 1
for i in range(len(args[0])):
sum1 += args[0][i] ** 2
sum2 *= np.cos(args[0][i] / np.sqrt(i + 1))
return sum1 / 4000 - sum2 + 1
result = f15
elif fname == "f16":
def f16():
sum1 = 0
sum2 = 0
sum3 = 0
for i in range(len(args[0])):
sum1 += np.sin(args[0][i]) ** 2
sum2 += args[0][i] ** 2
sum3 += np.sin(np.sqrt(abs(args[0][i]))) ** 2
return (sum1 - np.exp(-sum2)) * np.exp(-sum3)
result = f16
elif fname == "f17":
def f17():
sum1 = 0
sum2 = 0
for i in range(len(args[0]) - 1):
sum1 += ((0.25 * (args[0][i] + 1)) ** 2) * (
1 + 10 * (np.sin(np.pi * (1 + 0.25 * (args[0][i + 1] + 1))) ** 2))
for j in range(len(args[0])):
sum2 += mu_fun(args[0][j], 10, 100, 4)
return np.pi / len(args[0]) * (10 * (np.sin(np.pi * (1 + 0.25 * (args[0][0] + 1))) ** 2) + sum1 + (
0.25 * (args[0][-1] + 1)) ** 2) + sum2
result = f17
elif fname == "f18":
def f18():
sum1 = 0
sum2 = 0
for i in range(len(args[0]) - 1):
sum1 += ((args[0][i] - 1) ** 2) * (1 + np.sin(3 * np.pi * args[0][i + 1]) ** 2)
for j in range(len(args[0])):
sum2 += mu_fun(args[0][j], 5, 100, 4)
return 0.1 * (np.sin(3 * np.pi * args[0][0]) ** 2 + sum1 + ((args[0][-1] - 1) ** 2) * (
1 + np.sin(2 * np.pi * args[0][-1]) ** 2)) + sum2
result = f18
elif fname == "f19":
def f19():
sum2 = 0
a = [[-32, -16, 0, 16, 32, -32, -16, 0, 16, 32, -32, -16, 0, 16, 32, -32, -16, 0, 16, 32, -32, -16, 0, 16,
32],
[-32, -32, -32, -32, -32, -16, -16, -16, -16, -16, 0, 0, 0, 0, 0, 16, 16, 16, 16, 16, 32, 32, 32, 32,
32]]
for i in range(25):
sum1 = 0
for j in range(len(args[0])):
sum1 += (args[0][j] - a[j][i]) ** 6
sum2 += 1 / ((i + 1) + sum1)
return 1 / (1 / 500 + sum2)
result = f19
elif fname == "f20":
def f20():
sum1 = 0
a = [0.1957, 0.1947, 0.1735, 0.16, 0.0844, 0.0627, 0.0456, 0.0342, 0.0323, 0.0235, 0.0246]
b = 1 / np.asarray([0.25, 0.5, 1, 2, 4, 6, 8, 10, 12, 14, 16])
for i in range(11):
sum1 += abs(a[i] - (args[0][0] * (b[i] ** 2 + b[i] * args[0][1])) / (
b[i] ** 2 + b[i] * args[0][2] + args[0][3])) ** 2
return sum1
result = f20
elif fname == "f21":
def f21():
sum1 = 4 * (args[0][0] ** 2) - 2.1 * (args[0][0] ** 4) + (args[0][0] ** 6) / 3 + args[0][0] * args[0][
1] - 4 * (args[0][1] ** 2) + 4 * (args[0][1] ** 4)
return sum1
result = f21
elif fname == "f22":
def f22():
sum1 = 0
a = [[4, 4, 4, 4], [1, 1, 1, 1], [8, 8, 8, 8], [6, 6, 6, 6], [3, 7, 3, 7], [2, 9, 2, 9], [5, 5, 3, 3],
[8, 1, 8, 1], [6, 2, 6, 2], [7, 3.6, 7, 3.6]]
c = [0.1, 0.2, 0.2, 0.4, 0.4, 0.6, 0.3, 0.7, 0.5, 0.5]
for i in range(5):
sum1 += 1 / abs(sum((args[0] - np.asarray(a[i])) ** 2) + c[i])
return -sum1
result = f22
elif fname == "f23":
def f23():
sum1 = 0
a = [[4, 4, 4, 4], [1, 1, 1, 1], [8, 8, 8, 8], [6, 6, 6, 6], [3, 7, 3, 7], [2, 9, 2, 9], [5, 5, 3, 3],
[8, 1, 8, 1], [6, 2, 6, 2], [7, 3.6, 7, 3.6]]
c = [0.1, 0.2, 0.2, 0.4, 0.4, 0.6, 0.3, 0.7, 0.5, 0.5]
for i in range(7):
sum1 += 1 / abs(sum((args[0] - np.asarray(a[i])) ** 2) + c[i])
return -sum1
result = f23
elif fname == "f24":
def f24():
sum1 = 0
a = [[4, 4, 4, 4], [1, 1, 1, 1], [8, 8, 8, 8], [6, 6, 6, 6], [3, 7, 3, 7], [2, 9, 2, 9], [5, 5, 3, 3],
[8, 1, 8, 1], [6, 2, 6, 2], [7, 3.6, 7, 3.6]]
c = [0.1, 0.2, 0.2, 0.4, 0.4, 0.6, 0.3, 0.7, 0.5, 0.5]
for i in range(10):
sum1 += 1 / abs(sum((args[0] - np.asarray(a[i])) ** 2) + c[i])
return -sum1
result = f24
return result
# 获取目标函数相关参数
def get_functions_details(fname):
if fname == "f1":
lb = -100
ub = 100
nd = 30
fobj = lambda x: get_function_result("f1", x)
return lb, ub, nd, fobj
elif fname == "f2":
lb = -10
ub = 10
nd = 30
fobj = lambda x: get_function_result("f2", x)
return lb, ub, nd, fobj
elif fname == "f3":
lb = -1
ub = 1
nd = 30
fobj = lambda x: get_function_result("f3", x)
return lb, ub, nd, fobj
elif fname == "f4":
lb = -100
ub = 100
nd = 30
fobj = lambda x: get_function_result("f4", x)
return lb, ub, nd, fobj
elif fname == "f5":
lb = -100
ub = 100
nd = 30
fobj = lambda x: get_function_result("f5", x)
return lb, ub, nd, fobj
elif fname == "f6":
lb = -30
ub = 30
nd = 30
fobj = lambda x: get_function_result("f6", x)
return lb, ub, nd, fobj
elif fname == "f7":
lb = -100
ub = 100
nd = 30
fobj = lambda x: get_function_result("f7", x)
return lb, ub, nd, fobj
elif fname == "f8":
lb = -1.28
ub = 1.28
nd = 30
fobj = lambda x: get_function_result("f8", x)
return lb, ub, nd, fobj
elif fname == "f9":
lb = -5
ub = 10
nd = 30
fobj = lambda x: get_function_result("f8", x)
return lb, ub, nd, fobj
elif fname == "f10":
lb = -500
ub = 500
nd = 30
fobj = lambda x: get_function_result("f10", x)
return lb, ub, nd, fobj
elif fname == "f11":
lb = -10
ub = 10
nd = 30
fobj = lambda x: get_function_result("f11", x)
return lb, ub, nd, fobj
elif fname == "f12":
lb = -5
ub = 5
nd = 30
fobj = lambda x: get_function_result("f12", x)
return lb, ub, nd, fobj
elif fname == "f13":
lb = -5.12
ub = 5.12
nd = 30
fobj = lambda x: get_function_result("f13", x)
return lb, ub, nd, fobj
elif fname == "f14":
lb = -32
ub = 32
nd = 30
fobj = lambda x: get_function_result("f14", x)
return lb, ub, nd, fobj
elif fname == "f15":
lb = -600
ub = 600
nd = 30
fobj = lambda x: get_function_result("f15", x)
return lb, ub, nd, fobj
elif fname == "f16":
lb = -10
ub = 10
nd = 30
fobj = lambda x: get_function_result("f16", x)
return lb, ub, nd, fobj
elif fname == "f17":
lb = -50
ub = 50
nd = 30
fobj = lambda x: get_function_result("f17", x)
return lb, ub, nd, fobj
elif fname == "f18":
lb = -50
ub = 50
nd = 30
fobj = lambda x: get_function_result("f18", x)
return lb, ub, nd, fobj
elif fname == "f19":
lb = -65.536
ub = 65.536
nd = 2
fobj = lambda x: get_function_result("f19", x)
return lb, ub, nd, fobj
elif fname == "f20":
lb = -5
ub = 5
nd = 4
fobj = lambda x: get_function_result("f20", x)
return lb, ub, nd, fobj
elif fname == "f21":
lb = -5
ub = 5
nd = 2
fobj = lambda x: get_function_result("f21", x)
return lb, ub, nd, fobj
elif fname == "f22":
lb = 0
ub = 10
nd = 4
fobj = lambda x: get_function_result("f22", x)
return lb, ub, nd, fobj
elif fname == "f23":
lb = 0
ub = 10
nd = 4
fobj = lambda x: get_function_result("f23", x)
return lb, ub, nd, fobj
elif fname == "f24":
lb = 0
ub = 10
nd = 4
fobj = lambda x: get_function_result("f24", x)
return lb, ub, nd, fobj
if __name__ == '__main__':
# 白鲸种群数
out_n = 50
# 最大迭代次数
out_tmax = 1000
# 目标函数
out_fobj = "f22"
# 获取函数参数
out_lb, out_ub, out_nd, out_fobj = get_functions_details(out_fobj)
out_xposbest, out_fvalbest, out_best_fval = bwo(out_n, out_tmax, out_lb, out_ub, out_nd, out_fobj)
# 画图
plt.plot(range(len(out_best_fval)), out_best_fval)
plt.xlabel("t")
plt.ylabel("best value")
plt.show()
print("xposbest = ", out_xposbest)
print("fvalbest = ", f"{out_fvalbest: .5f}")
白鲸优化算法有哪些应用场景：
白鲸优化算法可以应用于各种优化问题，包括但不限于函数优化、约束优化、多目标优化等。具体的应用场景包括但不限于：
工程优化：白鲸优化算法可以用于机械设计、电力系统、交通运输等领域的优化问题，如结构优化设计、输电线路布局优化、公交车线路优化等。
无线通信优化：白鲸优化算法可以用于无线传感器网络的布局和能量管理，以及无线信道分配和频谱分配等优化问题。
图像处理：白鲸优化算法可以用于图像处理中的参数选择和优化问题，如图像去噪、图像增强等。
总之，白鲸优化算法可以应用于各种需要寻找最优解的问题，通过全局搜索和局部搜索的平衡，找到问题的最优解。
参考资料
https://mp.weixin.qq.com/s/vGcZ1_Bh4M5_CcuCclWt5Q
https://blog.csdn.net/jiaheming1983/article/details/129632261
https://blog.csdn.net/qq_41851955/article/details/127360805