适应度反向学习的平衡灰狼算法及其应用

Balanced grey wolf algorithm for fitness back learning and its application

针对传统灰狼优化算法位置更新时勘探与开发失衡,收敛速度慢且陷入局部最优的问题,提出一种改进的灰狼算法(balanced grey wolf algorithm based on fitness back learning,BGWO),引入非线性控制参数,增强算法前期勘探能力,加速收敛;在种群迭代阶段采用重心反向学习的最优适应度权重更新策略,平衡算法的勘探与开发。16组基准函数测试结果表明,改进后算法能自适应跳出局部最优,在加快算法收敛速度的同时提高全局收敛能力与精度。将BGWO应用于PV型旋风分离器粒级效率GBDT(gradient boosting decision tree)的建模,提高了GBDT的精度,模型相关系数0.980,均方误差0.00079,BGWO-GBDT与GBDT、PSO-GBDT和GWO-GBDT相对比,建模精度和稳定性明显提高,验证了BGWO的有效性。

In view of the imbalance between exploration and development,slow convergence and local optimization in the location update of the traditional grey wolf optimization algorithm,an improved grey wolf algorithm based on fitness back learning(BGWO)was proposed.Nonlinear control parameters were introduced to enhance the early exploration ability of the algorithm and accelerate the convergence.In the population iteration stage,the optimal fitness weight update strategy of barycenter back learning was adopted to balance the exploration and development of the algorithm.The test results of 16 benchmark functions show that the improved algorithm can adaptively jump out of the local optimum and improve the global convergence ability and accuracy while accelerating the convergence speed of the algorithm.Applied to the GBDT(gradient boosting decision tree)mode-ling of particle size efficiency of PV cyclone separator,BGWO improves the accuracy of GBDT,with a model correlation coe-fficient of 0.980 and a mean square error of 0.00079.Compared with GBDT,PSO-GBDT and GWO-GBDT,BGWO-GBDT significantly improves the modeling accuracy and stability,and verifies the effectiveness of BGWO.