[Paper Review] A Deep Learning Framework for Short-term Power Load Forecasting

Tinghui Ouyang, Yusen He+3 more|arXiv (Cornell University)|Nov 30, 2017

Energy Load and Power Forecasting10 references57 citations

TL;DR

The paper proposes a data-driven deep learning framework for short-term grid load forecasting that uses Box-Cox transformation, tail-dependence analysis via parametric copulas, and a deep belief network, evaluated on a Texas urban area dataset for day-ahead and week-ahead forecasts, outperforming several baseline models.

ABSTRACT

The scheduling and operation of power system becomes prominently complex and uncertain, especially with the penetration of distributed power. Load forecasting matters to the effective operation of power system. This paper proposes a novel deep learning framework to forecast the short-term grid load. First, the load data is processed by Box-Cox transformation, and two parameters (electricity price and temperature) are investigated. Then, to quantify the tail-dependence of power load on the two parameters, parametric Copula models are fitted and the threshold of peak load are computed. Next, a deep belief network is built to forecast the hourly load of the power grid. One year grid load data collected from an urbanized area in Texas, United States is utilized in the case studies. Short-term load forecasting are examined in four seasons independently. Day-ahead and week-ahead load forecasting experiments are conducted in each season using the proposed framework. The proposed framework is compared with classical neural networks, support vector regression machine, extreme learning machine, and classical deep belief networks. The load forecasting performances are assessed by mean absolute percentage error, root mean square error, and hit rate. Computational results confirm the effectiveness of the proposed data-driven deep learning framework. The prediction accuracies of both day-ahead forecasting and week-ahead forecasting demonstrate that the proposed framework outperforms the tested algorithms.

Motivation & Objective

• Motivate accurate short-term power load forecasting in modern grids with distributed energy resources.
• Develop a data-driven framework combining transformation, tail-dependence modeling, and deep learning for hourly load forecasting.
• Assess forecast accuracy across seasons for day-ahead and week-ahead horizons.
• Compare the proposed framework against classical neural networks, SVR, ELM, and classical DBN.
• Quantify performance using standard metrics (MAE, RMSE, hit rate).

Proposed method

• Apply Box-Cox transformation to load data.
• Investigate two parameters (electricity price and temperature) as factors.
• Fit parametric copula models to quantify tail-dependence and determine peak-load thresholds.
• Construct a deep belief network to forecast hourly load.
• Conduct day-ahead and week-ahead forecasting experiments across four seasons.
• Benchmark against classical neural networks, support vector regression, extreme learning machines, and classical deep belief networks.

Experimental results

Research questions

• RQ1 Does the proposed framework improve short-term (day-ahead and week-ahead) load forecasting accuracy compared to traditional models?
• RQ2 How do price and temperature influence load through tail-dependence, and how does this affect peak-load thresholds?
• RQ3 Are forecast performances consistent across seasons for hourly load on a real urban-area dataset?
• RQ4 What are the relative merits of Box-Cox transformation and copula-based tail modeling in this context?

Key findings

• The framework demonstrates improved forecast accuracy over the tested baselines in both day-ahead and week-ahead tasks.
• Tail-dependence analysis with copulas informs peak-load thresholding within the forecasting pipeline.
• Seasonal experiments show the proposed approach remains effective across four seasons.
• Compared methods include classical neural networks, SVR, ELM, and classical DBN, with the proposed framework outperforming them on the evaluated metrics.