Adaptive Neural Fuzzy Reasoning Model for Smart Advanced Science and Technology Letters Vol.42 (Mobile and Wireless 2013), pp.141-144 http://dx.doi.org/10.14257/astl.2013.42.33 Adaptive Neural Fuzzy Reasoning Model for Smart Grid Sang-Hyun Lee 1, Sang-Joon Lee 2, Kyung-Il Moon 1, 1 Dept. of Computer Engineering, Honam University, Korea, 2 Dept. of School of Business Administration, Chonnam National University, Korea { leesang64, kimoon}@honam.ac.kr, s-lee@chonnam.ac.kr Abstract. The current work formulates an automated decision-making model for electric-grid resource allocation. Resource allocation is primarily in the form of assigning the best power source to a sink. The model is built in Adaptive Neural Fuzzy Inference System (ANFIS). The input parameters for the model are the power balance and the consumption size. The fuzzy model did not have any clear boundary for the input values, so the model output was largely dependent on the contributions from individual portions. A separate model based on ANFIS has also been constructed to compare the results with Fuzzy model. The results obtained from this model show agreement in decision output and reveal the potential application areas of the smart grid decision-making. Keywords: Smart grid, fuzzy system, ANFIS. 1 Introduction There have been numerous attempts to create simulation systems for a smart grid environment [1-3]. In [3], the authors created an accurate hardware simulation of a simple micro-grid using MATLAB and Simulink to implement low-level electric circuits’ functionality. The agent interaction and collaboration were not thoroughly tested or evaluated in their simple simulation. In [2], an adaptive, self-healing framework for power grids based on intelligent-agent technologies is proposed, but little information is presented about any actual working simulation. In [1], an agent based simulation of a dynamic smart city is implemented; the simulation is promising because it has a degree of flexibility. The current work automates the operator’s decision making. Routing decisions are made by this study. The solution incorporates the self-healing characteristic of the smart grid. The decision comes in the form of picking the best power source for a particular demand area, so the model is capable of making the routing decisions about which source to connect with which sink or demand area. These decisions are made both by fuzzy logic and neural network model. For developing both the models, the initial assumption is to have power sources with an adequate supply. Hence, the supply side always needs a higher amount of power than the demand side. Based on ANFIS (Adaptive Neuro-Fuzzy Inference System) [4, 6], this paper explore fuzzy ISSN: 2287-1233 ASTL Copyright © 2013 SERSC

3 Fuzzy knowledge representation of smart grid Advanced Science and Technology Letters Vol.42 (Mobile and Wireless 2013) knowledge and its negative representation and reasoning by a specific example of power transmission and distribution. 2 ANFIS Architecture The smart grid Fuzzy controller is capable of making automated decisions for resource allocation. It takes some predefined inputs, and then based on the given parameters it picks the best source to connect with a sink. First, the Fuzzy variables are identified. After that, the membership functions are constructed. The membership functions have the trapezoidal and triangular functions. Fuzzy Logic does not provide any specific boundary. For example, if we see the power membership function, we notice that the segment boundaries of low, normal and big do not necessarily have any concrete value for their ranges. Boundary-value analysis is an approach that can be used to test any model. The idea of boundary-value analysis is to check if there has been any unexpected change in the output if the input values were changed close to the boundary. To test any model with boundary-value analysis, a clearly defined boundary is needed. An ANFIS largely removes the requirement for manual optimization of the fuzzy system parameters. A neural network is used to automatically tune the system parameters, for example the membership functions, leading to improved performance without operator intervention. In addition to a pure fuzzy logic approach, an ANFIS was also developed for the estimation of spray penetration length because the combination of neural network and fuzzy logic enables the system to learn and improve its performance based on warranty data. The neuro-fuzzy system with the learning capability of a neural network and with the advantages of the rule-based fuzzy system can improve the performance significantly and can provide a mechanism to incorporate past observations into the classification process. In a neural network the training essentially builds the system. However using a neuro-fuzzy scheme, the system is built by fuzzy logic definitions and then it is refined using neural network training algorithms. ANFIS architecture is consisted of five layers. The nodes in the input layer are adaptive. Any appropriate membership functions can be used. In this experiment Gauss membership functions were chosen to describe the input parameters because of their smoothness and concise notation. There are five main processing stages in ANFIS operation, including input fuzzification, application of fuzzy operators, application method, output aggregation, and defuzzification. 3 Fuzzy knowledge representation of smart grid Set the amount of electricity of the region is m, actual consumption is n, and power balance is t = n/m, then the monitoring level division depends on t, n according to the following rules: Rule1: If t is small, whatever the actual output m and actual consumption are, we recognize it as a good level. 142 Copyright © 2013 SERSC

Fig. 1. Decision making surface Advanced Science and Technology Letters Vol.42 (Mobile and Wireless 2013) Rule2: If t is close to zero and n is low, we recognize it as normal. Rule3: If t is close to zero and n is high, it is the critical degree, and we regard it as a control level, which means manual supervision is required. Rule4: If t is close to one, we regard it as warning level. This means artificial debugging and maintenance treatment is strongly recommended. Suppose that ANFIS has two inputs, balance and consumption, and one output level. Linguistic labels are small, normal and big. If-then rules: Rule1: if balance is very small then good=a1⋅ balance+b1⋅ consumption+c1 Rule2: if balance is very small and consumption is more or less small or very small then normal=a2⋅balance+b2⋅consumption+c2 Rule3: if balance is more or less small and consumption is more or less big then critical=a3⋅balance+b3⋅consumption+c3 Rule4: if balance is more or less big then warning=a4⋅balance+b4⋅consumption+c4 The outputs are fuzzy membership grade of inputs. If the gauss membership function is taken, the fuzzy set (very small) of input variable balance is given by  2  − µ ( balance ) =  exp   ( α balance )  2 (1)   σ 2 2   Here α and σ are the center and width of the fuzzy set Small. Also, the fuzzy set (more or less small) is given by  2 1  − µ ( balance ) =  exp  −  ( α balance )    2     σ    2  / 2 (2) 4 Application We use 5 years data in 4 regions by Wang, etc. [5]. To comprehensive analysis survey data in the same area over the years, we can average them respectively. Two fuzzy variables (power balance and consumption size) are used, and for each variable, there are seven different outcomes. The used rules are 19 of the total rules. The outcome of the rules is considered in terms of decision level. This linguistic outcome clearly captured the notion of the input. The output with a good level would generally be preferred over more or less low or very low. Figure 1 represents a decision making surface. Table 1 denotes city power distribution results. Fig. 1. Decision making surface Copyright © 2013 SERSC 143

Table 1. City power distribution results Advanced Science and Technology Letters Vol.42 (Mobile and Wireless 2013) Table 1. City power distribution results region Power balance Consumption size level a 0.7067 0.8411 Warning b 0.6475 0.9696 c 0.6807 1.0439 d 0.7433 0.8436 5 Conclusion The ANFIS model is fast in computation, so it can be used to filter out the best source-sink combination when a decision is crucial in terms of time. This kind of situations may arise when a sudden power failure has occurred due to any other unanticipated occurrences. The model would pick the best solution quickly in those situations. In particular, ANFIS provides some specific boundaries. Boundary-value analysis is an approach that can be used to test any model. The idea of boundary-value analysis is to check if there has been any unexpected change in the output if the input values were changed close to the boundary. In this way, the models can incorporate self-healing characteristic in the grid function. The time complexity is also in polynomial order compared to other mathematical models that generally have exponential-order complexity. Also, the models were straight forward and easy to implement. References Karnouskos, S.T., Holanda, N.: Simulation of a smart grid city with software agents. European Modeling Symposium 2009, Athens, Athens (2009) Noorian, Z.: An autonomous agent-based framework for self-healing power grid. Proc. IEEE Conference on Systems, Man and Cybernetics, 18--22, San Antonio, TX, (2009) Pipattanasomporn, M.: Multi-agent systems in a distributed smart grid: Design and implementation. IEEE PES 2009 Power Systems Conference and Exposition, Seattle,WA, (2009) Roger, J.J.: ANFIS: Adaptive-Network-Based Fuzzy Inference System, IEEE Transaction on Systems, Man, and Cybernetics, 23, No. 3, 665--685, (1993) Wang, J., Wang, Q., WeiQing, M.: Fuzzy Knowledge Representation and Reasoning of the Smart Grid Based on Medium Logic and Its Application, Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013), Published by Atlantis Press, Paris, France, 2786--2789, (2013) Yang, L., Yang, H., Huairong, S.: The application of ANFIS and WT in filtering, 2010 2nd Int. Conf. on Information Engineering and Computer Science, 1--3, (2010) 144 Copyright © 2013 SERSC