基于改进混沌理论和ACPSO-LSSVR的短期负荷预测

针对现有混沌理论中嵌入维数和延迟时间的选取难以达到最优,局域预测时邻近预测点的选取不够准确,提出了改进的混沌理论:采用改进的C-C法找到嵌入维数和延迟时间;邻近预测点的选取根据参考相点的演化趋势进行判断。针对最小二乘支持向量回归机LSSVR参数难以确定,提出ACPSOLSSVR:自适应混沌粒子群ACPSO一方面能根据群体早熟收敛程度和个体自适应值来调整惯性权重,另一方面能根据混沌变量的随机性和遍历性进行粒子的初始化,加快优化过程,防止局部极小。采用ACPSO来优化LSSVR的待选参数,提高负荷预测的精度。实例分析验证了该方法的可行性和实用性。

Short-Term Load Forecasting Based on the Improved Chaos Theory and ACPSO-LSSVR

In view of the existing chaos theory, the selection of the embedding dimension and delay time is difficult to determine, the choose of adjacent estimate point is not accurate enough. So an improved chaos theory is presented. On one hand, embedding dimension and delay time are determined via the improved C-C theory, the accurate embedding dimension is determined according to the forecasting result. On the other hand, the adjacent estimate points are determined according to the evolution trend of reference points. In view of the parameters of LSSVR is difficult to determine, ACPSO-LSSVR is presented. ACPSO can adjust inertia weight according to the premature convergence degree and the individual fitness. The initialization of the particles is made according to the randomness and ergodicity of chaotic variables. The parameters of LSSVR are optimized according to ACPSO, so the precision of load forecasting is improved. The example analysis proves the feasibility and practicability of the method.