结合小波变换与改进SSA优化小波神经网络的电力负荷预测

Power Load Prediction Combined Wavelet Transformed and Improved Sparrow Search Algorithm Optimizing Wavelet Neural Network

摘要:

电力负荷预测是输电网络扩展和规划及合理电力调度的关键手段。针对电力负荷时间序列的非线性和复杂性特征,提出结合小波变换与改进麻雀搜索算法优化小波神经网络的电力负荷预测模型ISSA-WNN。设计改进麻雀搜索算法ISSA对小波神经网络的关键参数初值寻优,有效解决梯度调参易陷入局部最优及对参数初值敏感的不足,提升模型学习能力。对标准麻雀搜索算法SSA改进,引入Logistic-Tent混合混沌种群初始化、发现者/警戒者自适应更新、跟随者可变对数螺旋更新和高斯-柯西混合变异策略提升算法寻优能力。利用小波变换对电力负荷样本分解与重构,降低负荷时序的无序性和波动性,在此基础上构建新的电力负荷预测模型ISSA-WNN。实验结果表明,与标准小波神经网络模型WNN和标准麻雀搜索算法优化小波神经网络模型SSA-WNN相比,预测模型ISSA-WNN的平均绝对百分比误差和均方根误差指标值平均可以降低18.42%和21.21%,其拟合能力更强,预测性能更加稳定。

Abstract:

Power load forecasting is a key means for transmission network expansion,planning and reasonable power dispatch.According to the nonlinear and complex characteristics of power load time series, a power load prediction model ISSA-WNN is proposed, which combines wavelet transform and improved sparrow search algorithm to optimize wavelet neural network. The improved sparrow search algorithm is designed to optimize the initial value of the key parameters of the wavelet neural network, which can effectively solve the problem that the gradient parameter adjustment is easy to fall into local optimum and sensitive to the initial value of parameters, and improving the model learning ability. The standard sparrow search algorithm is improved by introducing Logistic-Tent hybrid chaotic population initialization, discoverer/watcher adaptive update, follower variable logarithm spiral update and Gauss-Cauchy hybrid mutation strategy to improve the optimization ability of the algorithm. The wavelet transform is used to decompose and reconstruct the load sample to reduce the disorder and volatility of the load time sequence. On this basis, a new power load prediction model ISSA-WNN is constructed. The experimental results show that compared with the standard wavelet neural network model WNN and standard sparrow search algorithm optimizing wavelet neural network model SSA-WNN, the average absolute percentage error and root mean square error index values of the prediction model ISSA-WNN can be reduced by 18.42% and 21.21% on average, with stronger fitting ability and more stable prediction performance.