International Journal of Computer Applications (0975 – 8887) Volume 103 – No.9, October 2014 11 Novel Adaptive Evolutionary Computation Approaches to the Dynamic Economic Load Dispatch Problems with Non-Smooth Fuel Cost Function Sunny Orike Department of Computer Science Heriot-Watt University Edinburgh, United Kingdom ABSTRACT The paper presents novel approaches to solving the dynamic economic load dispatch (DELD) problem with valve-point loading effects. In dynamic environments, optimization problems change over time. They are also called time dependent or dynamic time-linkage problems, where decisions made at a given time may affect output obtained in a later time. It is therefore expected of algorithms solving dynamic optimization problems to both locate optimal solutions of the given problem and keep track of such solutions as they change with time. An investigation was made of three optimization methods in conjunction with three smart mutation variants, on benchmark problem cases involving 5 and 10 generating units, the major test cases in the literature with comparative results for other algorithms. The results suggest that the third approach which exploits the dynamic nature of the problem was capable of superior to the other two approaches. Comparisons with all approaches so far in the literature that have addressed these problems show that these evolutionary computation approaches are superior to other algorithms. General Terms Algorithm, Artificial Intelligence, Optimization, Electrical Power System, Performance, Simulation Keywords Economic Load Dispatch, Evolutionary Computation, Fitness Evaluation, Ramp-Rates, Valve-Point Effects 1. INTRODUCTION A great majority of problems in real-world applications are very complex and adaptive, as well as ways and methods used in solving them. The solutions are obtained by balancing several (and sometimes) conflicting multiple criteria in dynamic environments. Artificial intelligent techniques present powerful machine learning and nature-inspired approaches to investigate ways of solving problems in such dynamic environments which pose great challenges. This is the case of electrical power system optimization problems with multiple objectives, challenging constraints and dynamic demands. Here, there are constant changes affecting the power variables, various problem scenarios, with lots of operational constraints, resulting in the optimal solutions changing over time, and across dispatch periods. The Economic Load Dispatch (ELD) problem is concerned with the determination of the optimal combination of electrical power output for generating units in power stations on a near-real time basis, and with respect to a predicted load demand. The aim is to minimize the cost of producing power among those units (basically fuel cost), while obeying all operational constraints. The Dynamic Economic Load Dispatch (DELD) problem extends the traditional, also called Static Economic Load Dispatch (SELD) problem. It exists in practical systems involving ramp-rate limits (which constrain the changes that can be made to the settings of an individual generator between periods), where operational decisions at a given hour will affect the decision at a later hour [1]. This is one of several optimization problems that need repeatedly to be solved in the electricity industry. Formulation of DELD addresses two major issues: (1) the changes caused by ramp-rate limits, which makes the power generation to be periodically adjusted to meet targeted demands; and (2) the dynamic costs involved in changing from one output level to another [2]. This makes it a more applicable formulation of the economic load dispatch problem, but also a more difficult and complex optimization problem. Until now, the DELD has been treated as a series of unconnected static problems. Evolutionary computation (EC) is one of the four main paradigms of computational intelligence, a branch of artificial intelligence. Others are: Artificial Neural Network (ANN), Fuzzy Logic (FL) and Swarm Intelligence (SI) [3]. EC consists of the following algorithms/techniques: Genetic Algorithms (GA), Genetic Programming (GP), Evolutionary Programming (EP), Simulated Annealing (SA), Evolution Strategies (ES), Differential Evolution (DE), and Estimation of Distribution Algorithms (EDA). All EC approaches are based on Charles Darwin’s theory of natural evolution [4]. GA is the basis of all evolutionary algorithms, and models genetic evolutions, based on the concept of natural selection (survival of the fittest). Others are resemblance, with variations in structure and implementation. GP is a specialization of GA, but each individual is a computer program (represented as trees) that performs a user-defined task. EP is similar to GP, but the structure of the programs is fixed, whereas the parameters are allowed to gradually evolve. SA originated from the annealing process found in the thermodynamics and metallurgies. It involves a controlled heating and cooling of materials in order to increase the sizes of their crystals and reduce unwanted defects. ES do not implement crossover, with its results primarily dependent on mutation and selection. DE uses few control parameters and like ES, relies mainly on genetic mutation to achieve its solutions. EDA is also known as Probabilistic Model Building Genetic Algorithms, and motivated by the idea of discovering and exploiting interactions between variables in the solution,