Abstract Peak demand limiting is an efficient means to reduce the monthly electricity cost in cases where peak demand charge is a major factor. This paper presents an adaptive optimal monthly peak building demand limiting strategy based on exploration and exploitation tradeoff in threshold resetting. Two basis function components are developed, including a building load prediction model and an optimal threshold resetting scheme. The building load prediction model is built using the artificial neural network (ANN). The optimal threshold resetting scheme is developed based on the cost-benefit analysis, and the predicted building demands and/or actual building power uses. Three basic exploration-exploitation tradeoff schemes (i.e., the non-greedy, the greedy and the e-greedy schemes) are proposed for optimal threshold resetting. Monte Carlo simulation is conducted to analyze the impacts of the exploration-exploitation tradeoff scheme parameter on the demand limiting performance under uncertainties. The model validation results show that the ANN building load prediction model can achieve satisfactory accuracy with the average mean absolute percentage error (MAPE) of 5.7%. Case studies are conducted and the results show that the strategy based on the three proposed schemes can effectively reduce the monthly peak demand cost in different seasons. Monte Carlo simulation results show that the e-greedy scheme could achieve higher monthly net cost saving with better robustness when a large value of e is used in both winter and summer.

中文翻译：

---

基于勘探开发权衡的自适应最优月峰建筑需求限制策略

摘要 在高峰需求收费是主要因素的情况下，高峰需求限制是降低每月电力成本的有效手段。本文提出了一种基于阈值重置中探索和开发权衡的自适应最优月度高峰建筑需求限制策略。开发了两个基函数组件，包括建筑负荷预测模型和最佳阈值重置方案。建筑荷载预测模型是使用人工神经网络（ANN）建立的。最佳阈值重置方案是基于成本效益分析和预测的建筑需求和/或实际建筑用电量制定的。提出了三种基本的探索-利用权衡方案（即非贪婪、贪婪和电子贪婪方案）用于最佳阈值重置。进行蒙特卡罗模拟以分析不确定性下勘探-开发权衡方案参数对需求限制性能的影响。模型验证结果表明，人工神经网络建筑负荷预测模型的平均绝对百分比误差（MAPE）为5.7%，能够达到令人满意的精度。案例研究结果表明，基于提出的三种方案的策略可以有效降低不同季节的月高峰需求成本。Monte Carlo 仿真结果表明，当冬季和夏季都使用较大的 e 值时，e-greedy 方案可以实现更高的月净成本节约和更好的鲁棒性。