Deep-Based Conditional Probability Density Function Forecasting of Residential Loads

Abstract

This paper proposes a direct model for conditional probability density forecasting of residential loads, based on a deep mixture network. Probabilistic residential load forecasting can provide comprehensive information about future uncertainties in demand. An end-to-end composite model comprising convolution neural networks (CNNs) and gated recurrent unit (GRU) is designed for probabilistic residential load forecasting. Then, the designed deep model is merged into a mixture density network (MDN) to directly predict probability density functions (PDFs). In addition, several techniques, including adversarial training, are presented to formulate a new loss function in the direct probabilistic residential load forecasting (PRLF) model. Several state-of-the-art deep and shallow forecasting models are also presented in order to compare the results. Furthermore, the effectiveness of the proposed deep mixture model in characterizing predicted PDFs is demonstrated through comparison with kernel density estimation, Monte Carlo dropout, a combined probabilistic load forecasting method and the proposed MDN without adversarial training.Index Terms-Residential load forecasting, conditional probabilistic load forecasting, deep mixture network, convolutional neural network, gated recurrent unit

A Short-Term Residential Load Forecasting Model Based on LSTM Recurrent Neural Network Considering Weather Features

“…In [20], a framework based on an LSTM recurrent neural network to forecast the residential load of 69 customers was proposed and MAPE was used to compare this with various benchmarks including state-of-the-art load forecasting and achieved satisfactory results but the researchers selected the data by pre-screening the users, which may have affected the results. Afrasiabi et al [21] proposed a direct prediction model of the conditional probability density of a residential load based on a deep hybrid network. It achieved high accuracy in both the prediction of a single household's power load and the prediction of an aggregated load of 3500 residential households.…”

Modulated Gabor filter based deep convolutional network for electrical motor bearing fault classification and diagnosis

“…10: Minimisation the loss function (binary/ multi-nominal cross entropy) and back propagation based on (10) and Adam algorithm. 11: Update learning weights based on (11) 12: Check the epoch number If epoch < = M, go to the 9, otherwise end the training process 13: End…”

Generating scenarios from probabilistic short-term load forecasts via non-linear Bayesian regression

“…Bayesian methods utilizing neural networks as non-linear approximators have been proposed in the literature [15]- [17]. However, beyond the issue of interpretability of neural networks, such models are inherently hard to train and are additionally sensitive to hyperparameters.…”