Model And Algorithm Of Approximate Dynamic Programming For Medium And Long Term Optimal Scheduling

With the large-scale exploitation of clean energy such as hydroelectric power and wind power, improving the acceptance of clean energy in power grid and optimizing various power generation resources has become important task of optimal scheduling.The exploitation of cascade hydropower stations make the implementation of optimal scheduling for hydro-thermal coordination, cross-basin complementation and interaction between cascade hydro plants come true, but also leads to complexity of electrical coupling and water coupling, etc. And what’s worse, uncertainty of water inflows greatly affects the exertion of hydro optimal operation. As the wind integration increases, it would become more difficult for grid to perform peak regulation and optimal operation because of the volatility, intermittency and unpredictability of wind power. Moreover, the introduction of the market competitive mechanism has broken the traditional vertical monopoly management model and generation dispatching is transferred to the market-oriented pattern. The market agents’ behaviors, such as maximizing profits or minimizing cost, add to the complexity of the system operation. And the according uncertain factors including prices fluctuation incur the new difficulty with generation dispatching.Medium & long term optimal scheduling focuses on a variety of power generation resources within a long period of time. It is essential for a stochastic dynamic optimization problem. Longer period, wider range and stronger randomness make it harder for modeling and solving. This is because the traditional models derived from deterministic mathematical programming theory have some limitations to uncertainties.In 2011, Professor Powell from Princeton University proposed approximate dynamic programming (ADP) theory, and offered a powerful framework for dealing with certainty, uncertainty, continuous and discrete programming problems. ADP has been widely considered to be an effective solution to large-scale stochastic dynamic programming problems, also known as the basis of medium & long-term optimal scheduling models in study. ADP’s idea is perfect and concise. Firstly, the concepts and symbols of dynamic programming (DP) are used to describe multi-stage optimization decision problems. By the way of stochastic process embedded in the states, uncertainties are included in the optimal decisions. At last, the rich and open approximation policies expand the application of ADP.Based on the above backgrounds, this dissertation uses ADP methods to model and solve medium & long term optimal scheduling under different conditions respectively. The main research work is as follows:(1) Models for operation of cascade hydropower and hydro-thermal coordination are proposed on the basis of technologies such as thermal power, hydropower and wind power. In this process, some non-dispatchable technologies are analyzed, as well as the new more uncertainties under the electricity market. The stochastic process based on scenario tree was used to show the uncertainties. For reducing the number of scenarios but still keeping the essential features of the scenario tree, stochastic simulation and scenario reduction are combined to generated predicted scenario set.(2) A decision-making model for hydroelectric agents’self-scheduling under uncertainties is proposed according to the ADP’s multi-stage optimization theory. Putting the producers first and derived from cascade hydropower operation model, forward contracting decisions and day-ahead market trading decisions are recognized as different decisions of stages in an ADP framework. New model considers the uncertainties in natural inflows and market prices simultaneously with stochastic process based on predicted scenario set, optimizes the forward contracting decisions to obtain stable earnings. And then, approximation value function policy was designed to solve the model. Finally, an example of hydropower system in a province shows the superiority of ADP for stochastic programming problems. At the same time, the relationship between the decisions and profit, and the effects of reservoir operation are discussed.(3) An ADP-RO model for medium & long term optimal scheduling is proposed based on the multi-stage optimization of ADP, which proceed from the perspective of dispatching department and treated the scheduling as the resource optimization (RO) about the thermal, hydro, wind and so on. Prediction scenarios are used to represent uncertainties in day-ahead coal prices, water inflows and wind speeds. Decisions including forward contracts coal, day-ahead coal and water usage are recognized as stage decisions in ADP framework. Staged solution is used to reduce the size and difficulty of solving the problem. The proposed value function approximation policy of coal inventory and reservoir storage is used to make decisions and maintain the overall optimization after staged decomposition. Through the iteration between the decisions and the approximate value functions, the optimal decisions sequence is solved to obtain the optimal scheduling, which contains thermal-hydro-wind generation plan, coal supply plan, reservoir optimal operation, etc.(4) A simulation of optimal operation model base on ADP is proposed for analyzing the influence of large number of wind power access to grid. In the model, spinning reserves and pre-arrangement for wind power are set as tunable parameters, day-ahead scheduling and real-time rebalancing are connected to operate alternately. The simulation runs over days using predicted and real-time data to optimize the parameters and get the output of the units. Influences between power system and spinning reserves capacity, wind power forecasting accuracy, storage capacity are discussed. The simulation model can provide a tool for planning on large-scale wind power integration from theory and demonstration.Finally, summarize the whole thesis and explain the conclusion that the ADP method can behave well for the medium & long term optimal scheduling and large-scale stochastic dynamic optimization problems. The research achievements can be applied to the dispatching department for medium and long-term generation plan, coal supply plan, and reservoir water management and so on. At the same time, the work expands the application range of ADP.