Power Market Load Forecasting on Neural Network With Beneficial Correlated Regularization

In day-ahead market (DAM), load serving entities (LSEs) are required to submit their future load schedule to market operator. Due to the cost computation, we have found the inconformity between load accuracy and cost of power purchase. It means that more accurate load forecasting model may not lead to a lower cost for LSEs. Accuracy pursuing load forecast model may not target a solution with optimal benefit. Facing this issue, this paper initiates a beneficial correlated regularization (BCR) for neural network (NN) load prediction. The training target of NN contains both accuracy section and power cost section. Also, this paper establishes a virtual neuron and a modified Levenberg-Marquardt algorithm for network training. A numerical study with practical data is presented and the result shows that NN with BCR can reduce power cost with acceptable accuracy level.