Parameter Optimization Of Boiler Steam-water System Model Based On Genetic Algorithm And Artificial Neural Network

Power plant simulator plays an important role in training operators, reduceing misoperation and unscheduled shutdown to enhance safety and economy. The model of power plant is usually built in mechanism method. Though its qualitative character is reasonable, its accuracy is unsatisfied. Model engineer has to spend a great many of hours to regulate model parameters manually. Focused on this problem, genetic algorithm and atrifical neural network are studied to optimize mechanism model parameters. The following research work has been done.The mechanism models of steam and water system are analysed, which are a part of drum boiler in a 300MW power unit. The model parameters are divided into two groups: stable characteristics parameter and dynamic characteristics parameter. The relation between stable characteristic parameters and model output errors is studied. Meanwhile, the influence of dynamic characteristic parameters variety is also studied under the condition of the disturbance,. The output enthalpy formula of single phase heater is analyzed to find out the cause of see-saw effect. Then the criterion to avoid it and engineering solving method are put forward.Genetic algorithm is applied to optimize high temperature superheater model. The two target functions for minimizing stable errors and dynamic characteristics are constructed seperately. The optimization parameters corresponding to the target functions are seleted. The errors of high temperature superheater model optimized by GA are less than stated value. According to thermal data summary given by boiler manufacturer, GA is applied to optimize other heater models of steam and water system to minimize their stable errors.Based on the nonlinear mapping and fast computing ability of aritfical neural network(ANN), BP neural network is adopted to save optimization result by GA. The model imputs, outputs and optimized parameters make up learning sample set. After training, BP neural network calculates model optimization parameters directly. The BP neural network for each heater model of steam and water system are built and trained. All of them constitute ANN optimization library.ANN optimization library and GA are combined to optimize steam and water system model in the global space. At first, ANN optimization library is used to optimize single module. Then aiming at minimizing key parameter errors of the model, GA is applied to optimize model parameters in the global space to make its errors lesst than stated value. The integrated and intelligent model structure (IIMS) is put forward on the basis of mathematics modeling process study in the viewpoint of cybernetics. IIMS binds equipment principle and structure, strcutre parameters, running data, control algorithm and final model. It has the characteristic of explicit expression and additivity of simulation object property, and openness of optimization algorithm. The parameter optimization mode is changed from manully adjustment to algorithm optimization to cut down debugging workload due to IIMS. The debugging experience of senior engineer can be fixed in the program and be reused by beginner. It is benefit to model reusability, maintenance and update. Because the combination of genetic algorithm and atrifical neural network in the paper is a general frame, it is suitable for model optimization of power plant unit of different type.