Akimoto Kamiya

Fusion of soft computing and hard computing in industrial applications: an overview

Soft computing (SC) is an emerging collection of methodologies which aims to exploit tolerance for imprecision, uncertainty, and partial truth to achieve robustness, tractability, and low total cost. It differs from conventional hard computing (HC) in the sense that, unlike hard computing, it is strongly based on intuition or subjectivity. Therefore, soft computing provides an attractive opportunity to represent the ambiguity in human thinking with real life uncertainty. Fuzzy logic (FL), neural networks (NN), and genetic algorithms (GA) are the core methodologies of soft computing. However, FL, NN, and GA should not be viewed as competing with each other, but synergistic and complementary instead. Considering the number of available journal and conference papers on various combinations of these three methods, it is easy to conclude that the fusion of individual soft computing methodologies has already been advantageous in numerous applications. On the other hand, hard computing solutions are usually more straightforward to analyze; their behavior and stability are more predictable; and, the computational burden of algorithms is typically either low or moderate. These characteristics. are particularly important in real-time applications. Thus, it is natural to see SC and HC as potentially complementary methodologies. Novel combinations of different methods are needed when developing high-performance, cost-effective, and safe products for the demanding global market. We present an overview of applications in which the fusion of soft computing and hard computing has provided innovative solutions for challenging real-world problems. A carefully selected list of references is considered with evaluative discussions and conclusions.

Fusion of soft computing and hard computing for large-scale plants: a general model

The design of control systems for large-scale and complex industrial plants involves numerous trade-off problems, such as costs, quality, environmental impact, safety, reliability, accuracy, and robustness. Some of these parameters are even conflicting. Thus, the use of a multidiscipline approach is suggested to satisfy these requirements in an acceptable and well-balanced manner, and a fusion of soft computing and hard computing appears to be a natural and practical choice. Although the state-of-the-art soft computing technology has distinguished features, the use of soft computing technology would be ineffective, and may be doomed to fail, if it is improperly fused with conventional hard computing technology and control processes. Proper fusion is key to success, and a general model of fusion is worth examining. In this paper, through a survey of published literature, a general model of fusion is shown at the system level as well as at the algorithm level. In the system level, soft computing is applied to the upper level in a hierarchical control system, performing human-like tasks, such as forecasting and scheduling, or applied to ill-defined process models for carrying out intelligent control. Hard computing is used at the middle or lower control level for well-defined process models, carrying out coordinate control tasks while maintaining a high level of accurate and safety control. In the case of fusion at the algorithm level, this paper will discuss several types of tasks, such as scheduling and control.

Fusion of soft computing and hard computing: computational structures and characteristic features

Soft computing (SC) and hard computing (HC) methodologies are fused together successfully in numerous industrial applications. The principal aim is to develop computationally intelligent hybrid systems that are straightforward to analyze, with highly predictable behavior and stability, and with computational burden that is no more than moderate. All these goals are particularly important in embedded real-time applications. This paper is intended to clarify the present vagueness related to the fusion of SC and HC methodologies. We classify the different fusion schemes to 12 core categories and six supplementary categories, and discuss the characteristic features of SC and HC constituents in practical fusion implementations. The emerging fusion approach offers a natural evolution path from pure hard computing toward dominating soft computing

Fusion of soft computing and hard computing techniques: a review of applications

Soft computing (SC) is an emerging collection of methodologies, which aim to exploit tolerance for imprecision, uncertainty, and partial truth to achieve robustness, tractability, and low total cost. It differs from conventional hard computing (HC) in the sense that, unlike hard computing, it uses intuition or subjectivity. Therefore, soft computing provides an attractive opportunity to represent ambiguity in human thinking with the real life uncertainty. Fuzzy logic (FL), neural networks (NN), and genetic algorithms (GA) are the core methodologies of soft computing. However, FL, NN, and GA should not be viewed as competing with each other but synergistic and complementary instead, as emphasized by Dr. Zadeh . Considering the available literature, it is easy to conclude that the fusion of individual soft computing methodologies has been advantageous in numerous applications. In this paper, we give a review of applications where the fusion of soft computing and hard computing has provided innovative solutions for demanding real-world problems. A representative list of references is provided with evaluative discussions and conclusions.

Adaptive-edge search for power plant start-up scheduling

Power plant start-up scheduling is aimed at minimizing the start-up time while limiting maximum turbine-rotor stresses. A shorter start-up time not only reduces fuel and electricity consumption during the start-up process, but also increases its capability of adapting to changes in electricity demand. The start-up scheduling problem can be formulated as a function optimization problem with constraints. We have constructed an efficient and robust search model-a genetic algorithm (GA) with an enforcement operation-which forces the search along the edge of the feasible space, where the optimal schedule is supposed to exist. However, this model has to perform a prior Monte Carlo test to obtain the enforcement gains used for the implementation of the enforcement operation. In this paper, we attempt to eliminate the Monte Carlo test by proposing a self-reliant search model by introducing a GA with an adaptive enforcement operation that can generate and adapt enforcement gains during the search process. The test results of this proposed model show that the overall number of time-consuming dynamic simulations for the constraints calculation can be reduced further, thus increasing the overall efficiency of finding the optimal or near-optimal schedules.

Learning of communication codes in multi-agent reinforcement learning problem

Realization of cooperative behavior in multi-agent system is important for improving problem solving ability. Reinforcement learning is one of the learning methods for such cooperative behavior of agents. In this paper, we consider pursuit problem for multi-agent reinforcement learning with communication between the agents. In our study, the agents obtain communication codes through learning. Here, the codes are rules for communicating appropriate information under various situations. We call the learning of communication codes signal learning. The signal is expressed by bit sequence, and its length is set to be variable. We carried out experiment for performance comparison with varying the signal length from 0 to 4 bits. As a result, it has been shown that, in learning precision, the case of 1 bit or more bits communication outperformed the case of no communication. It also has been shown that 4 bits communication produced the best result among the five cases, while learning with longer signals required much more iterations.

Theoretical proof of edge search strategy applied to power plant start-up scheduling

Power plant start-up scheduling is aimed at minimizing the start-up time while limiting maximum turbine rotor stresses. This scheduling problem is highly nonlinear and has a number of local optima. In our previous research, we proposed an efficient search model: genetic algorithms (GAs) with enforcement operation to focus the search along the edge of the feasible space where the optimal schedule is supposed to stay. Based on a nonlinear dynamic simulation and a linear inverse calculation with the iteration method, the enforcement operation is applied to move schedules generated by GA toward the edge. We prove that the optimal schedule lies on the edge, ensuring that searching along the edge instead of the entire space can improve the search efficiency significantly without missing the optimum. Furthermore, we provide a theoretical setting equation for the inverse enforcement gains of the linear inverse calculation, intended to move schedules closer to the edge at each iteration of the enforcement operation. The theoretical setting equation is verified and discussed with the test results. We propose the theoretical setting equation with the test results as a guideline for the use of our proposed search model: GA with enforcement operation.

Advanced power plant start-up automation based on the integration of soft computing and hard computing techniques

The conventional HC (Hard Computing) techniques, i.e., ES (Expert System) and DDC (Digital Direct Control), have so far played a key role in large-scale thermal power plant automation systems which are based on a hierarchy structure. In this paper, we propose to integrate these HC techniques with the emerging SC (Soft Computing) in order to achieve the next-generation optimal automation system. SC, implemented through the integration of reinforcement-learning-based NN (Neural Networks) and GA (Genetic Algorithms), is capable of stochastic searching, learning and generalization, in solving those online optimization problems that are highly non-linear and accompanied with local optima. During the start-up process, SC is applied to generate and search the optimal or near-optimal schedule for the ES, which in turn controls the DDC-based controllers and monitors the whole power plant process with the given schedule. In accordance with our previous research, it has been verified that the optimal or near-optimal schedule can be obtained within tens of seconds, a time range which should be acceptable in power plant operation. The optimal schedule reduces the start-up time by approximately 10% for warm start-up mode.

Soft Computing-Based Optimal Operation in Power Energy System

Fossil fuels, chiefly coal, oil and natural gas, currently account for more than 60% of the primary energy used for electricity generation worldwide. This share will continue to increase steadily along with the growing global electricity demand [48]. There is therefore great demand for optimal operation of power energy systems aimed at reducing fossil fuel consumption. Such optimal operation could not only save fuel cost but also reduce CO2 emission, which is considered the main contributor to global warming.

Reward strategies for adaptive start-up scheduling of power plant

Power plant start-up scheduling is aimed at minimizing the start-up time while limiting maximum turbine-rotor stresses. A shorter start-up time reduces fuel and electricity consumption during the start-up process and increases its adaptability to changes in electricity demand. Online start-up scheduling increases the flexibility of power plant operation. The start-up scheduling problem can be formulated as a combinatorial optimization problem with constraints. This problem has a number of local optima with a wide and high-dimension search space. The optimal schedule lies somewhere near the boundary of the feasible space. To achieve an efficient and robust search model, we propose the use of an enforcement operator to focus the search along the boundary and other local search strategies such as the reuse function and tabu search used in combination with genetic algorithms (GAs). We also propose integrating GAs with reinforcement learning. During the search process, GAs would guide the learning toward the promising areas. Reinforcement learning can generate a good schedule in the earlier stage of the search process. After learning representative optimal schedules, the search performance virtually satisfies the goal of this research: to search for optimal or near-optimal schedules in 30 seconds. For industrial use, the design of a reward strategy is crucial. We show that (a) positive rewards succeed with both low and high-dimension reinforcement-learning output, and (b) negative rewards succeed only with low-dimension output. We present our proposed model with analysis and test results.