基于智能体和元胞自动机的鱼道内鱼类行为模拟

为建立基于智能体和元胞自动机(Agent-CA)的鱼道内鱼类行为模型,该研究以计算流体力学软件得到的水流流场数据为基础,根据过鱼试验研究认识到的鱼上溯行为特征和规律,定义了不同场景下鱼的避开障碍、寻找主流、逆流向前、逆流后退、冲刺、随机6种行为和行为规则,将鱼道划分成多个二维元胞空间,将元胞空间中元胞状态为"鱼"的网格看作是一个智能体(Agent)。在3种体型的竖缝式鱼道中应用该模型对草鱼幼鱼的上溯行为进行模拟,并将模拟结果与实际过鱼轨迹进行对比。结果表明:该研究构建的模型能够较好地描述实际过鱼的主要上溯轨迹,针对3种体型竖缝式鱼道的模拟特征轨迹范围相对误差分别为23.5%、7.7%和2.3%。本研究提出的基于智能体和元胞自动机的鱼道内鱼类行为模型具有较好的适用性,可为鱼类行为模拟研究提供重要基础数据。

Fish behavior simulation in fishway based on agent and cellular automata

Fish passing is closely related to the water flow and fish behavior in a fish-way experiment. The complex mechanism is often simulated on the basis of the flow field in most model tests, although the numerical simulation is still time-consuming and laborious. Moreover, a typical fish is generally selected to conduct the fish passing experiments, but the fish habits vary mainly on the type of fish, particularly on the contradiction of "fishway cannot meet the needs of all fish". Therefore, it is very necessary to consider the fish behavior in the numerical simulation of the flow field in the fish-way. An Agent-based model was widely used to simulate the fish behavior in recent years. However, the complicated calculation and difficult implementation usually occur especially in the case of complex behavior in the fish movement. Alternatively, a Cellular Automata (CA) model was highly suitable for the various complex systems. In this study, a feasible fish behavior model was proposed for the fish movement system in the fish-way using the integrated Agent-CA. The fish was treated as agents with independent behavior, where each agent was randomly chosen in the direction of movement through the probability of grid attraction, thereby evolving the whole process of fish behavior in the combined Agent-CA model. The model assumed that the different movement modes were taken for the fish, according to the velocity fields. Three velocity zones were divided according to fish behavior. The variety of movement patterns were set in each velocity field, while the probability of grid attraction was calculated using the patterns of fish movement behavior in the model. The patterns of fish movement behavior included obstacle-avoiding, mainstream seeking, random, upstream forward, upstream backward, and sprint behavior. A field experiment of fish passing was conducted to verify the efficiency of the model in the simulation. First of all, three kinds of vertical slot fish-ways were built with G/B=0.1, 0.25, and 0.5, by changing the relative length G/B of the guide wall. An experimental fish was selected as an eighty-five grass carp with a body length range of (10±2) cm. Secondly, the fish experiments were conducted in the three types of vertical slot fish-ways to record the track, particularly the characteristic track. And then, the velocity fields of three fish-ways were calculated using CFD numerical calculation, thereby serving as the Cellular space data in the fish behavior model. Finally, a randomly generated Agent (fish) was selected to simulate the fish behavior, with the body length in the range of (10±2) cm in the last fish-way chamber. The distribution of simulated and characteristic track of the Agent (fish) was obtained after repeating 85 times. Specifically, the scope of simulated track (W/B) for G/B=0.1, 0.25 and 0.5 were 42.5%, 1.1%, and 34.4%, while, the scope of simulated characteristic track (W/B) for G/B=0.1, 0.25 and 0.5 were 23.5%, 7.7%, and 2.3%, respectively. The data demonstrated that the constructed model can represent the main track of actual fish, but it is difficult to simulate the movement of a few fish deviating from the mainstream. The Agent-CA fish behavior model in fish-way can greatly contribute to clarifying the behavior laws of fish movement in modern aquaculture production.