基于特征筛选的VMD-MIC-SSA-Informer短期负荷预测

VMD-MIC-SSA-Informer short-term load prediction based on feature selection

对短期负荷进行准确、快速的预测有利于电力系统的安全稳定运行,故提出一种基于特征筛选的Informer短期负荷预测模型.针对短期负荷序列波动性大、随机性强的特点,利用变分模态分解(Variational Mode Decomposition,VMD)将原始负荷序列分解为不同频率的本征模函数(Intrinsic Model Function,IMF);考虑气象、地理、电价等影响因素,采用最大信息系数(Maximal Information Coefficient,MIC)对各IMF进行特征筛选;针对传统深度学习模型无法并行化、预测精度低的问题,应用了前沿的Informer模型进行短期负荷预测,同时利用了新颖的麻雀搜索算法(Sparrow Search Algorithm,SSA)进行模型参数优化.以西班牙区域级负荷数据为实例,同多层感知机(Multilayer Perceptron,MLP)、长短时记忆神经网络(Long Short-term Memory Neural Network,LSTM)和门控循环单元(Gated Recurrent Unit,GRU)进行横向与纵向实验对比,结果表明,提出的模型预测精度更高,平均绝对百分比误差低于1%.

Accurate and fast prediction of short-term load is beneficial to the safe and stable operation of the power system,so an Informer-based short-term load prediction model was proposed.For the characteristics of high volatility and randomness of short-term load sequence,the original load sequence was decomposed into intrinsic mode function(IMF) components with different frequencies by using variational modal decomposition(VMD);considering the influencing factors such as meteorology,geography and electricity price,the maximum information coefficient(MIC) was used for feature selection of each component;for the problems that traditional deep learning models cannot be parallelized and have low prediction accuracy,the cutting-edge Informer model was applied for short-term load forecasting,while a novel sparrow search algorithm(SSA) was utilized for model parameter optimization.The proposed model was compared with multilayer perceptron(MLP),long short-term memory neural network(LSTM) and gated recurrent unit(GRU) in cross-sectional and longitudinal experiments using regional level load data in Spain as an example.the results show that the proposed model has higher prediction accuracy with an average absolute percentage error of less than 1%.