Study On Reservoir Operation Models And Methods With Probabilistic Constraints Based On Stochastic Linear Programming

Reservoirs play a decisive role in the redistribution of water resources,especially surface water resources.In the medium-and long-term temporal scale,an inherent characteristic of reservoir operation is the randomness of runoff,which makes many constraints not be able to be completely satisfied,but only in a probabilistic or statistical sense.Handling these probabilistic constraints is a technical bottleneck in the stochastic optimal operation of reservoirs,especially to explicitly model and solve them,which is rarely discussed in previous research works.This paper focuses on the modeling method of reservoir operation based on the stochastic linear programming,and theoretically achieves explicit expression of probabilistic constraints of performance indicators,including the reliability,vulnerability and resilience in reservoir operation.Moreover,it also makes in-depth research on how to improve the operation benefit by using runoff forecast information.The main research works and innovative achievements of this paper include:(1)A stochastic linear programming model considering the probability of state decision in reservoir operation is constructed,which takes into account the probability of runoff state transition and makes the linear expression of nonlinear output function possible by introducing the probability of state decision variables.Also,the equivalence between the proposed stochastic linear programming model and the traditional stochastic dynamic model is not only proved theoretically,but also verified through the operation simulation with both the historical and simulated runoffs.(2)A method is proposed to explicitly express probabilistic constraints including the reliability,vulnerability and resilience in the stochastic linear programming model of reservoir operation.Although in case studies,the traditional stochastic dynamic programming model of reservoir operation can also derive the operation strategy in an implicit way by updating the punishment factors or Lagrange multipliers when using the penalty function or Lagrange relaxation method,but the stochastic linear programming model can express the probabilistic constraints in an explicit way and obtain the optimal operation strategy once for all,showing an advantage in theory.(3)A stochastic linear programming model is proposed to maximize the firm power of a hydropower station with a given reliability in reservoir operation.The explicit expression of the probabilistic constraint with the firm power is achieved by dividing the set of solutions that meet constraints on the release into 2 subsets depending on whether or not the firm power constraint is met.However,the stochastic dynamic programming is unable or very difficult to deal with the operation problem with the firm power to be a variable because of its requirement for the optimization objective and constraints be separable in stages.By comparing the simulation results of two optimization objectives,to maximize the firm power and to maximize the expected power generation,the feasibility and rationality of the proposed model in obtaining the operation strategy are verified.(4)A stochastic linear programming model with long-term runoff forecast information incorporated is presented.Still based on the probability of state decision,the model obtains satisfactory reservoir operation strategies by utilizing the inflow forecasting information with the normal feedforward artificial neural networks,as well as using two different methods to recursively obtain the transition probability of two runoff states including the runoff prediction information,which is integrated into the stochastic linear programming model.Two stochastic linear programming models with and without forecast information are compared in the case studies,which surprisingly show not much improvement on the expected energy production in the model with the forecast information.It is likely that the amount of information added by the second state of forecasted runoff is relatively small on the basis of the original probability transfer equation of runoff state.