RIME（雾凇优化算法）介绍

RIME optimization algorithm

Rime-ice is resulted from accumulated water vapor in the air that has not yet condensed. It freezes and sticks to objects such as tree branches at low temperatures. Due to their unique climatic characteristics and topography, some regions form a unique landscape like rime-ice every year, as shown in Figure 1.

雾凇是由空气中尚未凝结的累积水汽形成的。在低温条件下，它会冻结并附着在树枝等物体上。由于其独特的气候特性和地形条件，某些地区每年都会形成如图1所示的独特雾凇景观。

The growth process of rime ice is determined by the temperature, wind speed, humidity, air, and other factors, and rime formation varies under different conditions. At the same time, due to the influence of environmental factors and the growth pattern, rime-ice cannot grow indefinitely, and it will stop growing when it reaches a relatively stable state. The growth pattern of rime is generally divided into two types: soft-rime and hard-rime, which is mainly determined by the wind speed during the formation process, as shown in Figure 2, where ΔABC represents the growth plane of rime, and D1, D2, D3, D4 represent the birth points of rime. Usually, soft rime is generated in a breeze environment, and hard rime is formed in a high-wind environment. The breeze is characterized by small wind speed and variable wind direction, and the wind exists in all directions simultaneously and in the same height plane, as shown in Figure 2(a). Therefore, the soft rime formed by the breeze grows slowly and randomly. On the other hand, a gale is characterized by high wind speed and roughly the same wind direction in the same height plane, as shown in Fig. 2(b). Therefore, the hard rime formed by the gale is fast and grows in approximately the same direction.

霜冰的生长过程由温度、风速、湿度、气流等多种因素决定，且在不同条件下霜冰的形成存在差异。同时，受环境因素和生长模式的影响，霜冰无法无限生长，当达到相对稳定状态时便会停止生长。霜的生长模式通常分为软霜和硬霜两种类型，主要由形成过程中风速决定，如图2所示，其中ΔABC表示霜的生长面，D1、D2、D3、D4代表霜的生成点。通常，软霜在微风环境中生成，而硬霜则在强风环境中形成。微风以风速小、风向多变为特点，风同时存在于各个方向并在同一高度平面上，如图2(a)所示。因此，由微风形成的软霜生长缓慢且随机。另一方面，大风以风速高、在同一高度平面上风向大致相同为特点，如图2(b)所示。因此，由大风形成的硬霜生长迅速且大致沿同一方向。

(a) soft-rime	(b) hard-rime
图2 在不同环境下软霜和硬霜的形成过程

In summary, this study is inspired by the growth mechanism of rime-ice and proposes a soft-rime search strategy for algorithm search by simulating the motion of soft-rime particles. Also, a hard-rime puncture mechanism is proposed to exploit the algorithm by simulating the crossover behavior between hard rime agents. Finally, the selection mechanism of the metaheuristic algorithm is improved, and the positive greedy selection mechanism is proposed. This paper proposes the RIME algorithm with better performance by combining the above three mechanisms.

总的来说，这项研究受到了霜的生长机制的启发，并提出了一种软霜搜索策略，通过模拟软霜颗粒的运动来进行算法搜索。同时，提出了一种硬霜穿刺机制，通过模拟硬霜粒子之间的交叉行为来开发算法。最后，改进了元启发式算法的选择机制，并提出了积极贪婪选择机制。本文通过结合以上三种机制提出了性能更好的RIME算法。

3 Mathematical model of the RIME

In this section, the growth process of each rime strip is simulated by analyzing the effects of wind speed, freezing coefficient, the cross-sectional area of the attached material, and growth time. On the other hand, inspired by the diffusion-limited aggregation [50] method of simulating metal particle aggregation, the motion process of each rime particle coalescing into a rime agent is simulated by modeling the motion behavior of each rime particle, and the final generated rime-agent is in the form of a strip crystal. The RIME consists of four stages: the initialization of rime clusters, the proposed soft-rime search strategy, the proposed hard-rime puncture mechanism, and the improvement of the greedy selection mechanism.

本节通过分析风速、冻结系数、附着材料截面积以及生长时间等因素对每条霜带生长过程的影响，对其进行模拟。另一方面，受到金属粒子聚集模拟中扩散限制聚团[50]方法的启发，通过对每个霜粒运动行为建模，模拟各霜粒合并成霜体的过程，最终生成的霜体呈条状晶体形态。RIME算法包含四个阶段：霜簇初始化、所提出的软霜搜索策略、所提出的硬霜穿透机制以及贪婪选择机制的改进。

3.1 Rime cluster initialization

Inspired by reality, this paper treats each agent rime as the searched agent of the algorithm and the rime-population formed by all agents as the population of the algorithm. Firstly, the whole rime-population  is initialized. The rime population consists of  rime agents  and each rime- agent consists of  rime-particles , as shown in Figure 3 and Eq. (1). Thus, the rime-population  can be directly represented by the rime-particles , as shown in Eq. (2).

本论文借鉴现实情况，将每个个体霜体视为算法的搜索粒子，并将由所有粒子构成的霜体群体视作算法的种群。首先，对整个霜体种群进行初始化。霜体种群由个霜体粒子组成，而每个霜体粒子又包含个霜粒，如图3和式(1)所示。因此，霜体种群R可直接由霜粒表示，如式(2)所示。

​​​​​​​​​​​​​ ​​​​​​​

式(1)

式(2)

where  is the ordinal number of the rime agent and ​​​​​​​​​​​​​​  is the ordinal number of the rime particle. In addition,  is used to denote the growth state of each rime-agent, i.e., the fitness value of the agent in the meta-heuristic algorithm.

其中， 表示霜体代理的序号， 表示霜粒的序号。另外，用 表示每个霜体代理的增长状态，即该代理在元启发式算法中的适应度值。

3.2 Soft-rime search strategy

In a breezy environment, soft-rime growth is strongly random, and the rime particles can freely cover most of the surface of the attached object but grow slowly in the same direction. Inspired by the growth of soft-rime, this study proposes a soft-rime search strategy using the strong randomness and coverage of rime particles, which enables the algorithm to cover the entire search space in the early iteration quickly and does not easily fall into the local optimum.

在微风环境中，软霜生长具有很强的随机性，霜粒能够自由覆盖附着物表面的大部分区域，但在相同方向上生长较慢。受软霜生长的启发，本研究提出了一种利用霜粒强随机性和覆盖性的软霜搜索策略，使算法在早期迭代中能快速覆盖整个搜索空间，不易陷入局部最优解。

In a breezy environment, soft-rime growth is strongly random, and the rime particles can freely cover most of the surface of the attached object but grow slowly in the same direction. Inspired by the growth of soft-rime, this study proposes a soft-rime search strategy using the strong randomness and coverage of rime particles, which enables the algorithm to cover the entire search space in the early iteration quickly and does not easily fall into the local optimum.

在微风环境中，软霜生长具有很强的随机性，霜粒能够自由覆盖附着物表面的大部分区域，但在相同方向上生长较慢。受软霜生长的启发，本研究提出了一种利用霜粒强随机性和覆盖性的软霜搜索策略，使算法在早期迭代中能快速覆盖整个搜索空间，不易陷入局部最优解。

When the rime particles condense into soft-rime agents, there are the following characteristics:

当霜粒凝结成软霜代理时，具有以下特征：

 1) Before the particles condense to form a soft rime agent, each particle xij will wander according to a certain law, and the efficiency of the wandering is affected by environmental factors.

2) If the free-state rime particles move to the vicinity of a soft-rime agent, they will condense with the particles in the agent so that the stability of the soft-rime agent will change.

3) The distance between the centers of the two particles adhering to each other is not fixed, as the degree of condensation varies between each particle.

4) If the particles move directly outside the escape radius, no interparticle condensation occurs.

5) During the formation of a soft rime, the random condensation of each particle increases the area to which the agent is attached, resulting in a greater probability of free particle condensation. However, the agent will not grow indefinitely and will eventually reach a stable state due to environmental factors.

1）在霜粒凝结形成软霜代理之前，每个粒子xij会按照一定规律游动，游动效率受环境因素影响。

2）如果处于自由状态的霜粒移动到软霜代理附近，它们会与代理内的粒子发生凝结，从而改变软霜代理的稳定性。

3）相互粘附的两个粒子中心之间的距离不固定，因为每粒子之间的凝结程度不同。

4）如果粒子直接移出逃逸半径之外，则不会发生粒子间凝结。

5）软霜形成过程中，每个粒子的随机凝结会增加代理附着的面积，导致自由粒子发生凝结的概率增大。然而，由于环境因素，代理不会无限制地增长，最终会达到一个稳定状态。

In this paper, corresponding to the five motion characteristics of the rime particles, the process of condensation of each particle is concisely simulated, as shown in Figure 4, and the position of the rime- particles is calculated as shown in Eq. (3).

该论文针对霜粒的五种运动特性，对单个粒子的凝结过程进行了简洁模拟，如图4所示，并通过式(3)计算了霜粒的位置。 

  ​​​​​​​式(3)

where,  is the new position of the updated particle, and  and  denote the -th particle of the -th rime-agent.  is the -th particle of the best rime-agent in the rime-population . The parameter  is a random number in the range (-1,1) and  controls the direction of particle movement together with  will change following the number of iterations, as shown in Eq. (4).  is the environmental factor, which follows the number of iterations to simulate the influence of the external environment and is used to ensure the convergence of the algorithm, as shown in Eq. (5).  is the degree of adhesion, which is a random number in the range (0,1), and is used to control the distance between the centers of two rime-particles.

其中，为更新后的粒子的新位置，和分别表示第个霜体代理中的第个粒子。是霜体种群R中最佳霜体代理的第个粒子。参数是(-1,1)范围内的一个随机数，与一起控制粒子的运动方向，并随迭代次数变化，如式(4)所示。为环境因子，随迭代次数变化以模拟外部环境的影响，用于保证算法的收敛，如式(5)所示。为粘附度，是(0,1)范围内的一个随机数，用于控制两霜粒中心之间的距离。

式(4) ​​​​​​​

where  is the current number of iterations and  is the maximum number of iterations of the algorithm.

其中t为当前迭代次数，T为算法的最大迭代次数。

式(5)

where the mathematical model of  is the step function,  denotes rounding; the default value of  is 5, which is used to control the number of segments of the step function. Returning to Eq. (3),  and  are the upper and lower bounds of the escape space, respectively, which limit the effective region of particle motion.  is the coefficient of being attached, which affects the condensation probability of an agent and increases with the number of iterations, as shown in Eq. (6).

在这个数学模型中，表示步进函数，表示四舍五入操作；的默认值为5，用于控制步进函数的段数。回到公式(3)， 和分别表示逃逸空间的上限和下限，它们分别限制了粒子运动的有效区域。是附着系数，它影响个体的凝聚概率，并且随着迭代次数的增加而增大，这一点体现在公式(6)中。​​​​​​​

式(6)

​​​​​​​

 is a random number in the range ([51]) which, together with , controls whether the particles condense, i.e., whether the particle positions are updated. The pseudo-code for the soft-rime search strategy is shown in Algorithm 1.

r2是一个在([51])范围内的随机数，它与共同控制粒子是否发生凝结，即粒子位置是否进行更新。软霜搜索策略的伪代码如算法1所示。

Figure 4. Soft-rime particles motion

3.3 Hard-rime puncture mechanism

In strong gale conditions, hard-rime growth is simpler and more regular than soft-rime growth. When the rime particle condenses into a hard rime, there are the following characteristics: 1) The gale is so strong that other influences are negligible, resulting in different hard-rime agents snowballing in the same direction. 2) Due to the growth direction being the same, each rime agent can easily cross over, a phenomenon called rime puncture. 3) Like soft-rime agents, hard-rime agents increase in size as they grow, resulting in a greater probability of puncturing between agents in better growing conditions.

在强风暴条件下，硬霜的生长相比软霜更为简单、规律。当霜粒凝结成硬霜时，具备以下特征：1）强风暴的力量使得其他影响可以忽略不计，导致不同的硬霜代理朝着同一方向滚雪球般增大。2）由于生长方向一致，每个霜体代理可以轻易发生交叉现象，这种现象被称为硬霜穿透。3）类似软霜代理，硬霜代理在生长过程中体积也会增大，这使得在更好生长条件下的代理之间产生更大穿透概率。

Therefore, this paper is inspired by the puncturing phenomenon and proposes a hard-rime puncture mechanism, which can be used to update the algorithm between agents, so that the particles of the algorithm can be exchanged and the convergence of the algorithm and the ability to jump out of the local optimum can be improved. The puncture phenomenon is shown in Figure 5, and the formula for replacement between particles is shown in Eq. (7).

因此，本研究受到穿透现象的启发，提出了硬霜穿透机制，可用于代理之间的算法更新，使得算法中的粒子得以交换，从而提高算法的收敛性以及跳出局部最优的能力。穿透现象如图5所示，粒子间替换的公式见式(7)。

  , ​​​​​​​式(7)

where  is the new position of the updated particle and Rbest,  is the -th particle of the best rime-agent in the rime-population .  denotes the normalized value of the current agent fitness value, indicating the chance of the -th rime-agent being selected.  is a random number in the range (−1,1).

其中 表示更新后粒子的新位置，而 是霜体种群中最佳霜体代理的第 个粒子的位置。 表示当前代理适应度值的规范化值，它表明第 个冰代理被选中的概率。 是一个在区间(−1,1)内的随机数。

The pseudo-code for the hard-rime puncture mechanism is shown in Algorithm 2.

伪码形式的硬冰穿透机制展示在算法2中。

Figure 5. Hard-rime puncturing

3.4 Positive greedy selection mechanism

Typically, metaheuristic optimization algorithms have a greedy selection mechanism that replaces and records the best fitness value and the best agent after each update. The typical idea is to compare the updated fitness value of an agent with the global optimum, and if the updated value is better than the current global optimum, then the optimum fitness value is replaced, and the agent is recorded as the optimum. The advantage of such an operation is that it is simple and fast, but it does not help in the exploration and exploitation of the population and only serves as a record.

通常情况下，元启发式优化算法具有贪婪选择机制，在每次更新后替换并记录最佳适应度值和最佳代理。典型思路是在更新某个代理的适应度值之后，将其与全局最优适应度值进行比较，若更新后的值优于当前全局最优值，则替换最优适应度值，并将该代理记录为最优代理。此类操作的优势在于简便快捷，但无助于种群的探索和开发，仅起到记录作用。

在此基础上，本研究提出了积极贪婪选择机制。这一机制旨在在保留原有优势的同时，增强对全局解空间的探索和挖掘能力。具体来说，不仅关注单次更新后的最优结果，还可能根据一定的策略选择保留部分非最优但是有助于进一步发现潜在优良解的个体，以此实现更好的全局寻优效果。不过，请注意以上内容是对积极贪婪选择机制可能的一种解读和扩展解释，原始文本未提供具体的积极贪婪选择机制如何运作的详细信息。

Therefore, the paper proposes an aggressive greedy selection mechanism for participating in population updates to improve global exploration efficiency. The specific idea is to compare the updated fitness value of an agent with the fitness value of an agent before the update, and if the updated fitness value is better than the value before the update, a replacement occurs, and also, the solution of both agents is replaced. On the one hand, this mechanism allows the population to continuously have good agents through active agent replacement, which improves the quality of the global solution. On the other hand, as the position of the agents of the population changes significantly with each iteration, there will inevitably be agents that are worse than the population before the update and are detrimental to the next iteration. Therefore, this operation can be used to ensure that the population evolves in a more optimal direction at each iteration.

因此，本文提出了一种积极贪婪选择机制来参与种群更新，以提高全局探索效率。具体思想是将代理更新后的适应度值与其更新前的适应度值进行比较，如果更新后的适应度值优于更新前的值，则执行替换操作，同时替换两个代理的解决方案。一方面，该机制允许种群通过积极的代理替换不断拥有较好的代理，从而提高全局解的质量。另一方面，由于种群中各个代理的位置在每次迭代时会发生显著变化，难免会出现比更新前种群状况更差、不利于下一次迭代的代理。因此，这种操作可以确保种群在每次迭代时都能朝更优的方向进化。

In this paper, the pseudo-code of the positive greedy selection mechanism for solving the minimum value problem, as an example, is shown in Algorithm 3.

在这篇论文中，作为求解最小值问题的一个示例，我们展示了积极贪婪选择机制的伪代码，如算法3所示。

3.5 Proposed RIME algorithm

In summary, firstly, inspired by the motion of soft-rime particles in this section, a unique stepwise search and exploitation approach is designed to propose a soft-rime search strategy as the core optimization-seeking method of the algorithm. Immediately afterward, inspired by the crossover of hard-rime agents, a hard-rime puncture mechanism is proposed to achieve dimensional crossover interchange between ordinary and optimal agents, which is conducive to improving the solution accuracy of the algorithm. Finally, based on the greedy selection mechanism, an improved positive greedy selection mechanism is proposed to increase the diversity of the population and prevent the algorithm from falling into the local optimum as far as possible by changing the selection of optimal solutions. The overall structure of the algorithm in terms of pseudo-code and flow chart is shown in Algorithm 4 and Figure 6.

总结起来，首先，本节从软霜粒子的运动中获得灵感，设计了一种独特的分步搜索和利用方法，进而提出了软霜搜索策略作为算法的核心优化探查手段。紧接着，借鉴硬霜代理之间的交叉现象，提出了硬霜穿透机制，以实现普通代理和最优代理之间的维度交叉交换，有利于提高算法的解精度。最后，在贪婪选择机制的基础上，提出了改进的积极贪婪选择机制，通过改变最优解的选择方式，增强种群多样性，尽可能防止算法陷入局部最优解。算法的伪代码及流程图整体结构如算法4和图6所示。

Figure 6. Flowchart of RIME