DMOPSO based Multivariate Linear Regression Model for Electricity Consumption Prediction from Smart Meter

The Particle Swarm Optimization (PSO)algorithm is very useful for solving a high range of complexstatic and dynamic multi-objective real world problems. Thispaper presents a new challenge: Dynamic Multi-ObjectiveParticle Swarm Optimization based Multivariate LinearRegression model (DMOPSO-MLR). The DMOPSO-MLR isused to predict the electricity consumption for the incomingyear. Firstly, the weight vector is initialized randomly as aswarm of particles. Then, the iterative loop of optimizationprocess is started to determine the best weight vector that willbe multiplied with the real data for predicting electricityconsumption. In addition, the weight vector aims to ensure thegood convergence to minimize the difference between predictedand realistic data. This smart prediction system is providedwith a dataset of real historical half-hourly energy readingsfrom 3248 smart meters during twelve months. The proposedsystem is evaluated using the total year and monthly RelativeAbsolute Error. According to the quantitative and thequalitative results, the DMOPSO-MLR is proved as a goodprediction method for electricity consumption.