论文部分内容阅读
催化裂化集总动力学模型是1种多参数、高耦联的复杂反应动力学模型。用经典的优化算法求解模型参数时,常常需要对模型作一些数学处理,而复杂模型的数学处理相对困难;此外,经典算法的求解结果也常常不尽如人意。为解决这类模型的参数估计问题,以老遗传算法为基础,提出1种以亲子竞争和最优个体保护策略相结合的新遗传算法。新算法采取全局交叉和自适应变异,既保证了最大范围搜索解空间、避免算法在计算初期就陷入局部最优,又能在后期对局部细致搜索,提高了计算精度;克服了老算法随机性大、容易陷入局部最优的缺点。为测试新算法的效果,首先用某多参数复杂模型做测试,结果证明无论是遗传代数相同情况下的计算精度,还是为了达到某一精度而要求的计算代数,新算法都优于老者。然后用于估计催化裂化提升管反应器集总动力学动态模型参数。最后,取工业实际数据验证模型参数,泛化结果表明模型预测值与实际测量值基本吻合,120组数据的平均相对误差为1.71%,证明新算法适用性较好。
The lump dynamics model is a multi-parameter and highly coupled complex reaction kinetic model. When using classical optimization algorithm to solve model parameters, it is often necessary to do some mathematical processing on the model, while the mathematical processing of complex models is relatively difficult. In addition, the results of classical algorithms are often unsatisfactory. In order to solve the problem of parameter estimation of such models, a new genetic algorithm based on the old genetic algorithm is proposed, which is based on parent-child competition and the optimal individual protection strategy. The new algorithm adopts global crossover and adaptive mutation, which not only guarantees the maximum range search solution space, avoids the algorithm falling into the local optimum in the early stage of calculation, but also searches for the detail locally in the later period, which improves the calculation accuracy; overcomes the randomness of the old algorithm Large, easy to fall into the shortcomings of the local optimum. In order to test the effectiveness of the new algorithm, the first test with a complex multi-parameter model shows that the new algo- rithm is superior to the old algebra both in terms of the computational accuracy under the same conditions of the genetic algebra and in the computational algebra required to achieve a certain accuracy. Then used to estimate the dynamic model parameters of the lump dynamics of the FCC riser reactor. Finally, the industrial actual data is used to verify the model parameters. The generalized results show that the predicted values of the model are in good agreement with the actual measured values. The average relative error of 120 data sets is 1.71%, which proves the applicability of the new algorithm is better.