基于多储备池相关向量回声状态机和误差补偿的短期负荷预测研究

Research on Short-term Load Forecasting Based on Multi-reservoir Relevance Vector Echo-state Machine with Error Compensation

针对相关向量机核函数存在计算复杂度高、无法体现数据全部特点的问题,提出一种基于多储备池相关向量回声状态机和误差补偿的模型。多储备池具有稀疏性强、内部状态稳定的特点,采用多储备池加权组合代替核函数对数据进行高维映射,并通过稀疏贝叶斯理论求解模型参数。综合考虑多储备池初始化与训练过程的随机性,提取训练产生的误差并结合多种影响因素构建误差补偿环节,对预测结果进行修正。根据某地电网实际负荷数据进行仿真验证,结果表明该模型具有良好的精度与稳定性。

Aimed at the problem that the kernel function of a relevance vector machine has a high computational complexity and cannot reflect all the characteristics of data,an error compensation-multi-reservoir relevance vector echostate machine(EC-MrRVESM)is proposed.The multi-reservoir has characteristics of strong sparsity and stable internal state.The weighted combination of multi-reservoir is used instead of the kernel function to map the data in high dimension,and the model parameters are solved by the sparse Bayesian theory.Considering the randomness of the multi-reservoir initialization and training process,the error generated by training is extracted and an error compensation link is constructed by combing a variety of influencing factors,thus correcting the prediction results.The simulation results of the actual load data of one regional power grid indicate that the proposed model has satisfying accuracy and stability.