OPERATION OF A MULTIPURPOSE RESERVOIR USING STOCHASTIC DYNAMIC PROGRAMMING

Abstract

A surface water system is characterized by a set of variables and parameters such as inflows, demands, storage, losses and release limitations and operation planning aims at the most acceptable inter relationship between these governing variables and parameters. This is needed both for the effective control of the system as well as adequate utilization of the available resources. The essence of the problem is to take rational decisions such that water can be supplied at the right time and at the right place in adequate quantity, quality and dependability ensuring at the same time the safety of the storage structure and other related facilities as well as the safety of the downstream interests. That is why Reservoir Operation plays an important role in planning and management of water resources systems. After completion of reservoir when it is fully developed, detailed guidelines are to be given to the operator to enable him to take decisions about storing or releasing water. The specific problem in question is of reservoir operation, i.e., finding out an optimal release policy for the Sardar Sarovar Project (SSP) reservoir, a multipurpose and one of the most controversial reservoirs from ecological point of view on the Narmada river in Central India. The reservoir is the tail end reservoir in the river system located in the state of Gujarat (India), 95 km upstream from Gulf of Cambay and has a gross capacity of 9500 Mem with live capacity of 5800 Mem. It has irrigation and hydropower as main objectives besides M&I water supply and flood control as other objectives. The project development is proposed in three phases that spread over 45 years. During these three phases of development the irrigation release gradually increases, while hydropower production gradually decreases. It will irrigate almost 17 lakh (105) hectares of irrigable area in the state of Gujarat and Rajasthan, at the same time producing substantial amount of hydropower and supply drinking water to 135 urban centers (iv) and 8215 drought, prone villages. The various objectives and constraints of the reservoir are built in the model and optimal operating policy is obtained and presented to the reservoir operator. The approach is to arrive at a suitable optimal reservoir operating policy for efficient use of water resources. The main objective of the study is to develop models in stochastic and multiobjective environment. The dynamic programming (DP) technique is used here, as it is the most widely used and well established approach for solving such problems. All models developed in this study are discrete in nature, since the continuous variables of time, storage, and inflow are approximated by discrete units. As SSP being the tail end reservoir, the exact amount of inflow to the reservoir including contributions from just upstream reservoirs (five in numbers) is computed by simulation for the entire Narmada basin consisting of all 30 major reservoirs (which includes SSP and 29 upstream reservoirs). For this purpose 22 year (1948-1969) flow series is used. In calculating the total inflows to the SSP reservoir due weightage is given to the probability of occurrences of the regulated return flows from the upstream reservoirs. A backward moving deterministic dynamic programming (DEDP) algorithm has been applied to determine the monthly optimal operating policy for the SSP reservoir to provide an initial guideline for developing stochastic optimal reservoir operation model. This DEDP optimal operating policy is regressed to find out a general operating rule (DEDP Policy-I). Next, backward moving stochastic dynamic programming (SDP) algorithm has been used to develop monthly optimal reservoir operation policy for the reservoir (SDP Policy-I). Linear programming (LP) model has been introduced for determination of reliability of reservoir releases. Performance of these optimal reservoir operation policies is evaluated by simulation for the historical inflow series. Sensitivity analysis for the effect of number of discrete storage states and inflows on the operating policies (a second (v) set of policies, i.e., DEDP Policy-II and SDP Policy-II) is carried out and again evaluated by simulation. Finally, multiobjective stochastic dynamic programming optimal reservoir operation model (MOSDP Model-I) has been proposed to resolve the conflict between annual irrigation target, and annual hydropower generation with different reliability levels of achieving it. Another model (MOSDP Model-II) is used to resolve the conflict between flood control and conservation benefits from the reservoir. The flood control objective has been considered implicitly in this model as a maximum permissible storage volume in the reservoir. The best compromise solutions have been found out using SWT method. The SDP optimal operating policy performed better than the regressed DEDP operating policy considering reduction in power generation and canal release deficits as the main criteria in simulation. The SDP policy improves with increased numbers of discrete storage volumes and inflows as shown by SDP Policy-II. The problems of identification and elimination of redundant equations associated with determination of reservoir release reliability have been effectively solved by a newly introduced linear programming model. A computer program has been developed to reduce the Herculean task involved in a large size problem in preparation of data matrices for LP package. Procedure for using SDP monthly optimal operating policy for daily operation of a reservoir is also suggested. For the given project demand of SSP reservoir (11681.49 Mem), the MOSDP Model-I produced 37.5% more canal release (75% reliable) and 15% more power (90% reliable) than SDP Policy-I. Here MOSDP Model-I performs better than SDP Policy-I for SSP reservoir. It would also be possible to generate 49 MW of firm power as per MOSDP Model-I, whereas the firm power generation at 3r phase (ultimate) of development of SSP was proposed to be zero in the project report. (vi) For effective control of a particular return period flood (say 1000-year), the MOSDP Model-II found out that the maximum permissible storage level upto which the reservoir should be brought down is 135.09m in order to minimize downstream inundation This would inundate only 120 sq.km. of area and thus save as much as 266 sq.km. (69%) of downstream area from inundation. MOSDP Model-I and MOSDP Model-II can be used simultaneously for a reservoir operation problem. While the MOSDP Model-I can be used for the whole year, the MOSDP Model-II can be used only during a flood. Finally, some guidelines are recommended for selecting a proper reservoir operation policy considering advantages and disadvantages of various developed models, nature and selection of objective functions, serial correlation of inflows, length of data etc. The use of the proposed strategy of exhaustive analysis through these models suggests a promising scheme for reservoir operation of a single multipurpose reservoir, which is likely to guarantee an optimal operation policy. All the approaches are feasible and efficient with respect to computer memory and efforts.