Search for: All records

A novel multivariate deep causal network model (MDCN) is proposed in this paper, which combines the theory of conditional variance and deep neural networks to identify the cause-effect relationship between different interdependent time-series. The MCDN validation is conducted by a double step approach. The self validation is performed by information theory - based metrics, and the cross validation is achieved by a foresting application that combines the actual interdependent electricity, transportation, and weather datasets in the City of Tallahassee, Florida, USA. 

Abstract: Load forecasting plays a very crucial role in many aspects of electric power systems including the economic and social benefits. Previously, there have been many studies involving load forecasting using time series approach, including weather-load relationships. In one such approach to predict load, this paper investigates through different structures that aim to relate various daily parameters. These parameters include temperature, humidity and solar radiation that comprises the weather data. Along with natural phenomenon as weather, physical aspects such as traffic flow are also considered. Based on the relationship, a prediction algorithm is applied to check if prediction error decreases when such external factors are considered. Electricity consumption data is collected from the City of Tallahassee utilities. Traffic count is provided by the Florida Department of Transportation. Moreover, the weather data is obtained from Tallahassee regional Airport weather station. This paper aims to study and establish a cause and effect relationship between the mentioned variables using different causality models and to forecast load based on the external variables. Based on the relationship, a prediction algorithm is applied to check if prediction error decreases when such external factors are considered.