OPTIMAL HYDRO GENERATION SCHEDULING

Abstract

Due to fast growing energy requirements in a competitive and deregulated business environment, utilities are to maximize the generation from their reservoir based and run off river hydropower plants by carrying out hydropower scheduling in a fully integrated manner rather than considering individual plant and facility. Reservoir based hydropower plants are operated as per power or downstream water demand subject to storage available whereas the run off river hydropower plants since do not have any storage, generate power as per available water in the stream/canal. Available water resources are limited and thus their optimum utilization has become increasingly important. Many reservoir based hydropower plants are still operated on the basis of experience, rules of thumb or static rules set at the beginning of operation. Even small improvements in the operating policies of these plants can lead to large benefits to plant owners. Different hydropower plants are of different unit capacity, having different discharge and operating conditions with different performance and hence for meeting the power demand and water requirements, the optimize scheduling is necessary. The purpose of hydropower scheduling problemis to find the water releases from each reservoir and through each hydropower plant so as to maximize the total benefits of generated hydro energy and downstream water releases for its designated use like irrigation, drinking or navigation, while the environmental, physical, legal and contractual constraints are satisfied. Scheduling of hydropower plants is a challenging task that needs to take into account all relevant information ranging from long term to short term, reservoir size, installed capacity, demand for irrigation orother designated use. River inflows, power demand and downstream water requirements vary in time and space. In a pure hydro system, the function of the reservoir is to provide flow regulation, which is determined by the power and water requirements. When the reservoir serves more than one purpose, flow regulation becomes more complex. The random nature of stream flow and other hydrologic variables are fundamental complicating factors in determining optimum hydropower scheduling specially in case of longer study period. Scheduling of hydropower generation may be long term (several year basis), mid-term (monthly basis) or short term (hourly basis) each being dealt by dedicated Abstract models and solution techniques. The long term scheduling aims at evaluating the seasonal and multi annual optimal strategy in view of peak power demand, firm power supply, river inflows statistics, discharge and power generation limits of plant and water requirement. The main inputs for long term scheduling process are river inflow statistics, power demand and water demand etc. In the mid-term scheduling one assumes a deterministic period of one month and aims to match supply and demand optimally within the mid-term period. The mid-term scheduling provides the boundary condition for ensuring short term scheduling of the hydro system. It provides a link or a means of transforming results from long term scheduling process to a form suitable for the short term scheduling process. Short term scheduling is often solved as a deterministic problem within the constraints of hydraulic, electrical, firm power demand, environment and ramping so that the optimal benefits are achieved. The short term scheduling problem is normally solved plant wise. In the present study mid-term and short term scheduling problems have been solved and applied on a real system Satluj, Beas and Ravi rivers network (SBR) in Himalayan region of North West India for which field data has been collected from two utilities owning the reservoirs and hydro plants. This network includes inter river water transfer also. In the context of a much needed efficient management of hydropower network, mathematical optimization models become valuable tools, for improved planning and designing of complex operational schedules for any hydropower plant network system. Many techniques, particularly in the area of mathematical programming, such as maximum principal, variational calculus, dynamic programming, nonlinear programming, decomposition methods, network flow programming including multistage decision processes, were the subject of development for multi-reservoir system operation scheduling. Recently certain new techniques such as artificial neural network (ANN) and heuristics, fuzzy systems, genetic algorithms and evolutionary programming are being used for hydropower optimization. Multilayer Perceptron (MLP), Kohonen's self-organizing ANN and Hopfield neural network based approaches for hydropower generation scheduling have been reported by several researchers. The short term scheduling problem of hydro generation has been one of the most important and challenging optimization problems for economic operation and control of interconnected hydropower system. A large number of researchers have intensively investigated this scheduling problem. Main computational techniques that •> Abstract have been used are dynamic programming and its variants, network flow algorithms, mixed integer programming, lagrangian relaxation, decomposition method, branch and bound and projected gradient methods, genetic algorithms and continuous non linear optimization techniques. These methods have one or the other drawback as some of them do not take into account all important aspects of short term scheduling together such as unit efficiency, forbidden zone and ramping constraints and are unable to handle nonlinear, discontinuous, non-separable objective function and constraints. For the large size hydropower plant system having many operational strategies, the main issue is to assess its operation under different conditions with sufficient accuracy and less computing time. From the literature it has been observed that the inherent complexity of big systems limits the use of a very few techniques that can solve large scale hydropower scheduling systems. Beside this they require setting of parameters which are problem dependent and require large memory and computational efforts. The scheduling based on operating rules uses approximate models to produce fast results which may not be optimal solution. The objective of the present study is to develop fast and efficient methods for finding optimal generation scheduling of large size hydropower plant network. In recent years ANNs and evolutionary techniques have emerged as powerful tool due to their ability to map complex nonlinear functional relationships with good accuracy and speed. ANNs have proved to be successful in many areas of operation planning studies. Ease of hardware implementation is one of the features that attract the use ofANN techniques. Because of modular architecture, the nature of implementation of ANN is a parallel one, both in software and hardware and hence fast speed of execution can be achieved. The aim of this study is to develop methods for finding optimum hydropower generation scheduling of large size system both on mid-term and short term basis. The author's contribution is summarised as follows: • Mid-term scheduling of hydropower plants using SLP method A method based on sequential linear programming (SLP) has been developed to find optimal mid-term hydropower generation scheduling of a large size hydropower system. The objective function of the hydropower generation scheduling problem is to maximize total hydropower generation over the entire study period in Abstract subject to constraints on water balance in the reservoir, headworks/barrage, limits on reservoir storage, spillage, water release though turbines, end-of-horizon storage and downstream water requirement/ water rights determined from ecological, irrigation, drinking, recreation and navigation requirements etc. The formulated problem may be infeasible due to unrealistic downstream irrigation targets and hence constraints on downstream water requirements are treated as soft constraints which are augmented in the objective function with the help of weightages. In this method the problem is solved via a sequence of linear programmes using primal - dual relationship which results into the final solution close to that of original non linear programming problem solution. The advantages of the method are reduced number of variables, no requirement of exact solution of sub-problems resulting into small computation time per iteration, insensitive to the scaling of variables and not requiring evaluation of hessian matrix. The method has been applied on SBR rivers network. The benefits (revenue) from the hydropower generation scheduling in terms of additional power generation compared to generation achieved by utilities during the study year, deficit of water for meeting irrigation requirements and thereafter additional power generation combined with water deficit have been worked out. • Hybrid neural network approach for Mid-term scheduling of hydropower plants A hybrid neural network approach is presented for solving mid-term scheduling of hydropower plants. The proposed technique combines Kohonen Self Organizing Map (SOM) known as unsupervised network and Multilayer Perceptron (MLP) known as supervised network. Therefore advantages of both neural network architectures have been utilized in this approach. A 2D self-organizing feature map is used for clustering of similar input patterns. For each cluster two MLPs are trained one with hydropower generation and other one with discharge through turbines in each time period and from each plant as output features. The proposed approach has been applied on SBR rivers network. The proposed technique gives better performance and more accurate results then obtained by MLP or SOM alone. It is observed that total network consisting of small MLPs converges much faster as compared to a single MLP of the same total size for similar error performance. The method has been applied on SBRRivers network system. • PSHNN approach for mid-term scheduling of hydropower plants 4 Abstract A method for finding optimum mid-term hydropower generation scheduling using parallel self-organizing hierarchical neural network (PSHNN) is presented. The PSHNN design has four stage neural networks (SNN), which are connected to each other in parallel. Each SNN has been trained by scaled conjugate gradient algorithm. Care has been taken to generalize the neural network by using early stopping method and by generating a large set of training data points. The advantages of this method are (i) backpropagation of errors from stage to stage to learn weights are avoided (ii) It does not require differential and invertible nonlinearities. (iii) It has faster learning time due to fewer training vectors in later stages and parallel operation of SNN during testing, (iv) It has provision for real-time adjustment of error detection bound, making PSHNN high fault tolerant and robust and (v) It is self organising as the number of SNN during training are adjusted by itself. The performance of the trained PSHNN has been demonstrated on SBR rivers network system and found that the results are very close with the results obtained by using conventional methods. • Short term scheduling of hydropower plants using Hybrid PSO method Short term scheduling of hydropower system operation is an important task for finding out optimal commitment as well as loading of generation units. A short term hydropower generation scheduling problem is considered taking into account generation losses due to water conductor head losses and efficiency losses, start up and shutdown cost of units, discharge ramping constraint, forbidden zone for turbine operation, power balance and limits on the problem variables in an integrated manner. The formulated scheduling problem is a mixed integer discontinuous nonlinear programming problem and hence conventional methods may not provide optimal solution. In this study the short term scheduling problem has been solved by hybrid particle swarm optimization (HPSO) method. This method provides better solution than other methods due to its intrinsic nature of updates of positions and velocities in which benefits from previous history have been taken. HPSO method which makes use of binary PSO and real coded PSO provides global optimal solution and is robust. Disjoint constraints are handled easily. The technique is also derivative free and very few parameters are to be adjusted. Performance of HPSO is better than other evolutionary method as it is insensitive to the population size which should not be small and best solution is obtained quickly. Abstract The proposed method has been demonstrated on reservoir based Pong hydropower plant with and without ramp constraints. Ramping constraints were imposed on tailrace canal from the view of structural safety of its lining. The use of this method for short term scheduling improves operating efficiency of the plant, provide input to maintenance scheduling, satisfy release, spinning reserve requirement and environmental requirement at downstream. Savings in water use for generating the targeting power has been calculated.