基于主分量分析和遗传神经网络的电力负荷预测

针对中长期电力负荷预测受经济、人口、天气、政策的影响密切的问题，为了保证预测的准确性和快速性，应当将这些影响因素全部考虑进来作为预测模型的输入。首先通过主分量分析法在保证不丢失输入信息的情况下将输入的维数降低，然后使用遗传算法优化网络的权值和阈值，最后用L—M贝叶斯正则化BP算法训练网络，并与传统的只考虑经济因素的预测方法的训练结果进行了比较。通过《重庆统计年鉴》统计的数据仿真，结果表明本文提出的预测方法的预测精度更高。

Electric Load Forecasting Based on the Principal Component Analysis and GA Neural Network

With the problem of medium and long-term electric load forecast affected by economy, population, climate and policy, all these factors should be considered and regarded as inputs of the forecasting model. Firstly, the dimension of inputs is reduced by principal components analysis under the condition of not missing input message. Secondly, weights and thresholds are optimized by genetic algorithm. Thirdly, L-M Bayesian BP algorithm is used to train the network. The forecasting results are compared with that of method of only considering economy factors.