Value of human intervention in model-driven hydroelectricity production planning

Abstract

The decision-making in production planning is commonly supported by analytical models. Although the models are getting more precise and capable through advancements in technology and model-building, human intervention is still often required. However, human planners are subject to biases and cognitive limitations which may question the value of the human intervention. The previous research has found out that human intervention arises among others from system inadequacy and misaligned incentives between the model and the human planner. Moreover, decision-makers are found to intervene also by habit or for the feeling of control.

This thesis addresses human intervention in model-driven production planning in the field of hydroelectricity production. The monetary value of the discharge decisions is high due to volatile electricity prices and the task is challenging to human cognition. Moreover, uncertainties arise from the stochastic and seasonal prices, hydrological conditions and the delays of water flows between the consecutive plants. The operation of river systems is constrained by regulations related to environmental and recreational issues.

A case study was conducted on a hydroelectricity producer which uses an optimization model to construct a short-term production plans. Over 80% of the hourly recommendations obtained from the optimization tool were adjusted by the planners. The size of the adjustments was considerable; the size was on hourly level on average nearly 20%. The intervention resulted to a plan with less hour-to-hour changes and rounded quantities of discharges. The planner routinely intervened because of feasibility factors, e.g. too many ramp-ups for the control room or unnecessary generator startups. Moreover, external factors such as planned maintenance had a significant impact when present. In total, the planner and the model achieved similar prices. However, especially when the planner intervened for external factors alone or together with feasibility factors, the achieved price was lower. On the contrary, when they only intervened for market-driven reasons, the achieved price was slightly higher.

Since feasibility-related issues were the most common drivers for adjustments and these can be traced back to the optimization tool, the model could be improved to account for these. This would enable the planners to focus more on market-driven planning. Moreover, regular external factors, such as planned maintenance, could be included in the tool.