针对局部邻域搜索方法搜索结果对初始解位置依赖大,难以摆脱局部最优解影响的问题,本文提出了一种带惩罚的逆梯度进化算法.该算法通过给当前位置适应度施加一个仅与在该点处停留时间正相关的惩罚以迫使该个体沿逆梯度方向移动,进而逃离当前局部极值点.同时为了防止出现"回跳"现象,引入禁忌邻域,禁止当前个体重回原先位置.相对于一般启发式算法跳出局部极值点的随机性,该算法通过惩罚实时构造填充函数以逃离当前局部极值点的机制具有一定的确定性因素,提高了算法的搜索效率.将该算法应用于换热网络优化问题上,分别对10SP1和10SP2两个经典算例进行验证,获得了优于已有文献的优化结果,表明该算法具有较强的跳出局部最优解能力.%The local neighborhood search methods depend greatly on the initial solution position and are usually subjected to the local optimal solutions. Therefore, an inverse gradient evolution algorithm with penalty is proposed in this paper. This algorithm employs a penalty positively related to the dwell time at the local optimal position to force the individual to move along the inverse gradient direction and then far away from the current local optimum. Meanwhile, to prevent 'rebound' phenomenon, tabooed neighborhood is introduced into the algorithm to prohibit the individual from moving back to its original position. As the filled functions are established by penalty in real-time, the mechanism of escaping the local optima in the proposed method is relatively deterministic rather than random in the heuristic methods, which improves the search efficiency for the individual. Finally, applying the algorithm to heat exchanger network synthesis problems, its effectiveness is verified by the typical 10SP1 and 10SP2 case studies. The obtained solutions are better than those in the literature, demonstrating the relatively strong ability of the proposed method to jump out of local optima.
展开▼
