Optimization Of Initial Pressure Of Steam Turbine Based On Krill Herd Algorithm And Feedback Fast Learning Network

With the rapid development of Chinese economy, the amount of energy consumption is increasing over the past decade. Coal-based energy structure determines that the development of electric industry will mainly rely on thermal power generation in China. At present, thermal power generation accounts for more than 70% of the total power generation in electric industry in China. Due to the increasing amount of electricity consumption in society, the gap between peak and low load of power grid is gradually broadening. Hence, large thermal power units have to participate in peaking operation of power grid. Steam turbine operates in the state of long-term variable load off its design conditions, so its thermal economy decreases severely.The study focuses on the research of optimal operation of the steam turbine in the thermal power plant. In order to solve the main problems in the optimization of the initial pressure of steam turbine operation, the latest technology and method are adopted, and the following researches are made:First, the thermal economic analysis method of steam turbine in thermal power plant is studied in depth. It is concluded that the enthalpy of steam and the noise signal in DCS data have an important influence on the thermal economic analysis. Therefore, an improved median filtering MMEM algorithm is introduced in the data pre-processing stage, which eliminates the Gauss noise in the original data. Then, the international calculation formula of water and steam properties, IAPWS-IF97, is adopted. The accuracy of water enthalpy in the analysis of thermal economy is guaranteed.Second, the fast learning network(FLN) is a feedforward neural network with compact network structure, fast learning and training speed. However, the fast learning network ignores the inertia and time continuity of the system. The feedback fast learning network(B-FLN) is proposed by adding a time delay feedback process in the fast learning network. The experiment results show: the precision of modeling and the performance of noise suppression in the inertial and delay system are improved.Third, the analysis model of thermal system is established based on mass balance and energy balance of extraction heaters. The model includes thermal balance matrix equation, heat equation and power equation. The neural network of B-FLN is proposed to calculate the heat rate of steam turbine, which improved the accuracy of the thermal equilibrium analysis method.Last, the hybrid model of steam turbine thermal system is created, which mixed thermal balance analysis and B-FLN method. In addition, the global intelligent optimization algorithm, krill herd optimization algorithm(KH), is adopted as the optimization tool. By optimizating the main steam pressure of steam turbine under different loads, the optimal operating pressure of the 600 MW steam turbine was obtained under varying conditions. It reduces the heat rate of steam turbine, and guide the energy saving for thermal power plant.