Load Prediction under Accelerated Urbanization

Abstract

Electricity load forecasting plays a significant role in power system. Among all load forecasting tasks, high granularity cases, i.e., industrial and residential level are acknowledged much more challenging than the aggregated forecast. Heterogeneous high granularity loads with different patterns make the performance of load forecasting frameworks volatile, which brings difficulty for model design and model selection. To this end, we explore the predictability for different loads in terms of various indicators. Specifically, we measure the predictability for various electric loads from internal load structure and external prediction accuracy using different forecasting models. We further adopt the notion of job-housing ratio and construct synthetic data using various loads with predefined portions to simulate electric loads for different job-housing cases, which provide a new perspective to study the load predictability of the transformed power systems under accelerated urbanization process. Numerical study shows that different electric loads have various degrees of predictability. Meanwhile, job-housing ratios make a significant difference in the performance of prediction models and the effects on the predictability are quite varied when combining different industry loads.