Structure and Component of Biological-Environment Adaptive Control System

Proceeding of 2013 International Seminar on Climate Change and Food Security ISCCFS 2013 Palembang, South Sumatra -Indonesia, 24-25 October,2013 118 the similarity of variable process to reference. These parameters should be optimized to avoid instability of the control system by reference change or disturbance. Biological-environment adaptive control system based on the identification of the optimal set point and the estimation of optimal control parameter consists of sub-system of optimal contro l parameter‘s determination and sub-system of real-time control. The results of those sub-systems can be applied on the real time controlling of biological environment in the plants house. There are two terms of biological-environment. Firstly, it is the relationship between environmental factors as inputs and plant produce as output. Secondly, the sensors for detecting the inputs and outputs are placed near the plant canopy in order to detect the surrounding conditions of plants related to photosynthesis, respiration and transpiration. The optimal environmental condition for plant growth has to be established prior to controlling by using a mathematical equation model.

3. Structure and Component of Biological-Environment Adaptive Control System

Biological-environment adaptive control system includes some operators. Those are identification, optimization, control fuzzy and PID, image processing, as well as mass-transfer model [16]. Artificial neural network ANN structure consists of layers i.e. input, hidden and output layers. The nodes within layers are connected by weights, and they receive inputs and process them to obtain an output. The connections of nodes between layers determine the information flow between nodes, either one direction or bidirectional. Through a set of learning process, the weights are modified in such a way by using a particular algorithm, such as back propagation. The back propagation algorithm is used in layered feed-forward ANNs. It means that the nodes are organized in layers, and send their signals forward, and the errors are propagated backwards. The inputs are received by nodes in the input layers, and the output is node in the output layers. There may be one or more hidden layers between input and output layers. Back propagation method is successfully proved by the process of multi-layer neural network training. Information about the error is also controlled through the system, and it is used to justify the relationship between the layers, so that the network performance will increase. Back propagation algorithm is a commonly used to train the ANN. The network weight is modified by minimizing the sum of squared errors which is calculated against all the output nodes. Back propagation algorithm is a form of differential gradient which is aimed to reduce the error. The training begins with random weights, and the goal is to adjust them so that the error will be minimal. Genetic algorithm GA is one of frequent optimization techniques used, for example in controlling micro climate in greenhouse [17]. Genetic algorithm uses a natural analog phenomenon such as biological evolution, in which the best individuals in a population will experience mutations. The population consists of individual who each is possible to solve a problem. Each individual, which in this case is similar chromosomes, has fitness value corresponding to the feasibility of the solution of the problem. Some individuals in the population are with better fitness value opportunity for reproduction. Cross over or mutation might occur during the process. At the end of process, the best individual would appear and therefore it would be selected. Genetic algorithms use direct analogy of natural properties. This algorithm uses a population of individual which in each represents a possible solution to a given problem. Each individual has a value of fitness function in accordance with the completion of the feasibility problem. Highly fit individual has the opportunity to recombine with the individuals in the population. This process produces new individuals as offspring that has parent nature. Individuals that have low fitness value could not be selected for reproduction so that the species will become extinct. In addition to cross-over between individuals, the natural evolution of the well known mutations, i.e. changes in individuals who are not influenced by other individuals. The best fitness value would produce the next generation of individuals. On the other hand, low fitness value of individual would be discarded in order to produce the similar next generation. This process is repeated until the desired generation or a high value of the fitness function for solving problems is obtained. Proceeding of 2013 International Seminar on Climate Change and Food Security ISCCFS 2013 Palembang, South Sumatra -Indonesia, 24-25 October,2013 119 The characteristic of GA is different from other optimization techniques. The differences are found in the operating system which GA runs based on the coding set in the parameter, the selecting system is in a population, information of objective function as technique for evaluating the individual who has the best solution, not the derivative of a function. In addition, AG uses probabilistic transition rules, not deterministic rules. The variables used in the genetic algorithm are as follows: 1 the fitness function, which is owned by each individual for analyzing the individuals level of compliance with the criteria to be achieved. Fitness is maximized with the application of genetic algorithms, 2 population is the number of individuals, who are involved in every generation, 3 Opportunity probability in recombination occurs in a generation, 4 opportunity mutations occur in every transfer bits, and 5 number of generations to be established would determine the number of the application of genetic algorithms. Logical fuzzy and PID techniques have been widely applied in the control system. The steps of the application are as follows: 1 calculation of the error and error difference, 2 fuzzification, 3 determination of the control rules decision matrix and calculation of the maximum value, and 4 de-fuzzification. Proportional integral differential control and its parameters follow Ziegler and Nichols method.

4. Determination of Optimal Set Point and Controlling