基于自适应扰动量子粒子群算法参数优化的支持向量回归机短期风电功率预测

Short-Term Wind Power Prediction Based on Support Vector Regression Machine Optimized by Adaptive Disturbance Quantum-Behaved Particle Swarm Optimization

智能电网的建设和大规模风电接入电网对短期风电功率预测精度提出了更高的要求。为了克服支持向量回归机(support vector regression machine,SVR)依赖人为经验选择学习参数的弊端,在量子粒子群优化(quantum-behaved particle swarm optimization,QPSO)算法中加入自适应早熟判定准则、混合扰动算子和动态扩张收缩系数,提出了自适应扰动量子粒子群优化算法(adaptive disturbance quantum-behaved particle swarm optimization,ADQPSO),并使用ADQPSO优化选择SVR的学习参数。实例研究表明,ADQPSO算法全局寻优能力强、鲁棒性好、计算耗时短,利用ADQPSO优化得到的SVR参数,可有效提高模型的预测精度;与反向传播神经网络(back propagation neural network,BPNN)和径向基神经网络(radial basis functionneural network,RBFNN)相比,提出的ADQPSO-SVR能够提高短期风电功率预测的准确性和稳定性。

A higher accuracy of short-term wind farm output prediction is required due to the construction of smart grid and grid-connection of large-scale wind farms. To remedy the defect of support vector regression machine （SVR） that the learning parameter selection of SVR depends on factitious experiences, adaptive disturbance quantum-behaved particle swarm optimization （ADQPSO） algorithm is proposed by adding adaptive premature criterion, mixed disturbance operator and dynamic expansion-contraction coefficient in quantum-behaved particle swarm optimization （QPSO） algorithm, and ADQPSO algorithm is used in optimized selection of learning parameters for SVR. Case study shows that the proposed ADQPSO algorithm possesses such advantages as good global search ability, strong robustness and high computation efficiency, and applying the ADQPSO algorithm to the optimization of the obtained learning parameters of SVR the accuracy of short-term wind power prediction is higher than those by back propagation neural network （BPNN） and radial basis function neural network （RBFNN）.