基于扩展记忆粒子群-支持向量回归的短期电力负荷预测

Short-term power load forecast based on particle swarm optimization with extended memory and support vector regression

为了快速准确高效地预测短期电力负荷,提出了一种带扩展记忆的粒子群优化技术(PSOEM)和支持向量回归(SVR)相结合,以历史负荷数据、气象因素等作为输入的基于PSOEM-SVR的短期电力负荷预测方法。PSOEM比传统PSO收敛速度更快精度更高具有更强的寻优能力,用它来优化组合核函数SVR参数,减少了SVR参数设置的盲目低效性,获得较优的PSOEM-SVR预测模型。该模型的实例仿真预测结果表明该方法比BP神经网络具有更好的准确性和稳定性,平均绝对误差控制在1%以内。

To forecast short-term power load accurately, quickly and efficiently, a method based on the particle swarm optimization with extended memory （PSOEM） and support vector regression （SVR）is proposed for short-term power load forecast, taking the historical load and atmospheric data as model inputs in the paper. PSOEM has more extensive capability of global optimization than PSO owing to higher accuracy and convergence rate. In order to reduce blindness and inefficiency, PSOEM is used to optimize the parameters of the SVR with compounding kernels, and obtains an optimum PSOEM-SVR model to forecast the load. Example forecasting results of the PSOEM-SVR model show that this method can offer a better performance in accuracy and stability than BP neural net, and its average absolute error is within the range of 1%.