智能控制（双语）期末答案2023秋

1.（ ） is not the achievement of reinforcement learning?
A:Face recognition B:AlphaZero C:AlphaGo D:OpenAI
答案:AI参考:正确答案是C：AlphaGo。\nAlphaGo并不是强化学习的一种具体应用或模型，而是围棋人工智能的一种应用。而face recognition是面识技术，OpenAI则是一家专门研发AI技术的公司，这些都可以属于强化学习技术范畴的应用领域。而AlphaZero则是一种深度强化学习算法，它的研究成果已经被广泛应用于各个领域。因此，根据题目描述，只能选C，AlphaGo。因此答案是C。'
2.（ ）is regarded as one of the best effective management techniques and is widely applied in practice.
A:Ramp metering B:Queue management C:Off-ramp metering D:Density metering
答案:AI参考:正确答案是B:Queue management。\n排队管理是公认的最有效的管理技术之一，在实践中被广泛应用。因此，排队管理是最佳选项。其他选项如ramp metering（坡道计量）、off-ramp metering（出匝道计量）和密度计量（Density Metering）都不是有效的管理技术，也不是广泛应用的实践技术。因此，正确答案是B。'
3.Ordinary sets can be represented by characteristic functions, and fuzzy sets can be represented by ( ）function.
A:Discrete B:Membership C:Triangle D:Distribution
答案:AI参考:正确答案是B：Membership。\n\n在模糊集合理论中，通常使用隶属度函数来表示模糊集合，而普通集合则使用特征函数来表示。因此，选项B“Membership”函数是表示模糊集合的正确选项。选项A“Discrete”表示离散集合，选项C“Triangle”表示三角形分布，选项D“Distribution”表示分布，它们都不是表示模糊集合的合适选项。'
4.Conventional PID algorithm is insufficient for controlling the pulp consistency because（ ）。
A:Pulp consistency’s fluctuation B:The change of pulp-motor’s speed C:Lagging and interference in the process D:The shake of pulp’s consistency
答案:AI参考:正确选项是：B:The change of pulp-motor’s speed\n\n根据题目中的描述，传统的PID算法不足以控制纸浆浓度，这是因为纸浆浓度的变化、纸浆电机速度的变化、工艺中的滞后和干扰以及纸浆浓度的抖动。其中，纸浆电机速度的变化可能会影响纸浆的浓度，因此需要考虑到这一点。因此，选项B是正确的。其他选项在题目中未提及或与题目描述不符。'
5.The basic genetic operators of genetic algorithms include（ ）.
A:Selection and copy B:Mutation C:Crossover D:Learning
答案:AI参考:正确选项是：C:Crossover。\n\n遗传算法的基本遗传操作包括选择、交叉、变异和学习等。其中，交叉（Crossover）是遗传算法中最为核心的遗传操作，它通过对个体进行重组来产生新的个体，从而在搜索过程中产生新的解，加快搜索速度并保持搜索方向的正确性。因此，选项C是正确的。其他选项中，选择和复制是用来确定哪些个体将被用于后续的遗传操作；变异是一种随机扰动操作，可以保持种群的多样性，防止算法陷入局部最优解；学习通常是指根据一定的学习规则对遗传操作进行修正或调整。'
6.The three elements of neural networks are（ ）.
A:Defuzzification interface B:Topology C:Learning rules D:The characteristics of neurons
答案:AI参考:答案是：BCD。根据题意，神经网络由三个元素组成，而这三个元素是拓扑结构、学习规则和神经元的特性。因此，正确选项为BCD。Defuzzification interface（解模糊化接口）不是神经网络的组成部分，它通常在模糊逻辑系统或模糊控制器中使用。因此，A选项不正确。'
7.The three inference methods that the reasoning machine of the expert system can use are（ ）.
A:Chain reasoning B:Bidirectional reasoning C:Forward reasoning D:Reverse reasoning
答案:AI参考:正确选项是：C:Forward reasoning。\n\n推理机在专家系统中使用的三种推理方法分别是正向推理（Forward reasoning）、反向推理（Reverse reasoning）和双向推理（Bidirectional reasoning）。正向推理是从已知事实出发，尝试找出可以解释其他已知事实的结论。反向推理则是在一些可能的结论中，根据现有的事实选择一个最可能的结论。双向推理是正向推理和反向推理的结合，即在探索可能的结论时，既考虑结论的可信度，也考虑前提的正确性。因此，选项C“正向推理”是正确答案。'
8.The disadvantage of early parameter learning control include（ ）
A:Slow speed of convergence B:Large amounts of memory C:Complex selection of classifier D:Difficult selection and extraction of features
答案:AI参考:正确答案是：A:Slow speed of convergence 和 B:Large amounts of memory。\n\n早期参数学习控制的主要缺点包括收敛速度慢和需要大量的内存。因此，答案为A和B。C和D选项虽然也与早期参数学习控制有关，但不是其主要缺点。'
9.The intelligent control method is more suitable for the object with the characteristics of（ ）than the traditional control method.
A:Nonlinearity B:Robustness C:Time-varying D:Uncertainty
答案:AI参考:正确选项是：B：Robustness。\n\n智能控制方法更适用于具有鲁棒性的对象，因为鲁棒性是指控制系统在面对不确定性、干扰、扰动等因素时，仍能保持稳定性和控制效果的能力。相比之下，传统控制方法在面对这些因素时可能无法有效控制对象。因此，智能控制方法更适合具有鲁棒性特性的对象。选项A、C、D虽然也具有一定的特点，但不如鲁棒性在智能控制中应用广泛。'
10.A pattern recognition-based learning system consists of（ ）
A:The operation control level B:The task-scheduling level C:The organization level D:The self-turning level
答案:AI参考:多选题A pattern recognition-based learning system consists of（ ）的正确答案是：B:The task-scheduling level 和 C:The organization level。\n\n这个系统包括任务调度层和组织层。任务调度层负责根据学习算法的规则分配任务到合适的处理单元，组织层则负责组织学习资源并优化整体系统性能。这两个层次共同构成了基于模式识别的学习系统的核心组成部分。因此，选项B和C是正确答案。选项A的“操作控制水平”虽然也在一定程度上影响学习系统的性能，但并不是这个系统的核心组成部分，因此排除。选项D的“自我转化水平”也与基于模式识别的学习系统的核心功能不相关，因此排除。'
11.Membership functions that are frequently used in fuzzy control include Gaussian membership functions and（ ）。.
A:Trapezoidal membership function B:Triangular membership function C:Generalized bell-shaped membership function D:S-shaped membership function

12.According to the nature of high-level decision-making functions, indirect expert controllers include（ ）.
A:Organizational expert controller B:Adaptive expert controller C:Optimized expert controller D:Coordinated expert controller 13.（ ）are examples of “Vanilla” Neural Networks.
A:Video-captioning B:Image captioning C:Image classification D:Action prediction 14.In a neural network, the processing unit is often divided into （ ）, according to different processing functions.
A:Output unit B:Input unit C:Hidden layer unit D:Conversion unit 15.Mathematical tools for intelligent control research include（ ）.
A:Combination of discrete events and continuous time systems B:Optimization theory C:Neural Network Theory D:Fuzzy Set Theory E:Combination of symbolic reasoning and numerical calculation 16.For the feedback network, the more stable points, the stronger the association and recognition ability of the network. Therefore, the more data items of stable points, the better the association function.（ ）
A:错误 B:正确 17.Two ways to control the top temperature of a petrochemical plant are by a reflux rate and by reboiler temperature. （ ）
A:错误 B:正确 18.Support Vector Machine (SVM) is a kind of generalized linear classifier that classifies data according to supervised learning. （ ）
A:错误 B:正确 19.Two-dimensional fuzzy controller is equivalent to traditional PI controller. （ ）
A:正确 B:错误 20.Model Predictive Control (MPC) is an important branch of automatic control theory and has been applied in many fields. （ ）
A:错误 B:正确 21.For CNN, the purpose of stride is that the border pixels are not undervalued (or lost) from the output. （ ）
A:错误 B:正确 22.The main disadvantage of BP network is that there are many optimal in the objective function , and it is easy to fall into local minimum when learning by gradient descent method. （ ）
A:正确 B:错误 23.In the two link robot manipulator control system, the higher order multilayer neural controller gives very accurate control result even with only PD controller. （ ）
A:对 B:错 24.The availability of neural network direct inverse control depends to a great extent on the accuracy of the inverse model. （ ）
A:正确 B:错误 25.In fuzzy control, the result of fuzzy reasoning must be fuzzy quantity.（ ）
A:错误 B:正确 26.Fuzzy control does not require a mathematical model of the controlled object. （ ）
A:正确 B:错误 27.Multiple randomly spread initialization is an important reason for escaping from the local optimal. （ ）
A:错误 B:正确 28.Direct expert controller is used to replace conventional controller and directly control the production process. （ ）
A:正确 B:错误 29.The overall control structure using multilayer high-order neural network is asymptotically stable and the feedback control structure is uniformly bounded stable. （ ）
A:错误 B:正确 30.Policy-based RL algorithm is a typical model-based approach. （ ）
A:错误 B:正确 31.Darwin proposed Genetic Algorithm in 1975.（ ）
A:错误 B:正确 32.The predictive model of MPC must be mathematical model to describe input and output relationship of the controlled object. （ ）
A:错误 B:正确 33.For the expert control system of pulp consistency, the better the knowledge is, the better the system performs. （ ）
A:正确 B:错误 34.For the ant colony optimization, the concentration of pheromones on a trail works as a positive feedback loop. （ ）
A:错误 B:正确 35."Warm" can be used as a language variable value.（ ）
A:错误 B:正确 36.Iterative learning control cannot be used as a tool in nonlinear system theory. （ ）
A:错误 B:正确 37.Neural network control is an intelligent control method developed by combining neural network and control theory. （ ）
A:正确 B:错误 38.Learning control systems can be divided into off-line trainable system and on-line trainable system. （ ）
A:正确 B:错误 39.Uniform crossover is especially suited for problems with constraints, like TSP.（ ）
A:错误 B:正确 40.In PSO algorithm, the inertia weight influences tradeoff between global and local exploration. （ ）
A:正确 B:错误 41.The error surface of the network based on BP algorithm has and only one global optimal solution.（ ）
A:正确 B:错误