Heat pump system is widely used in district heating because of its high energy efficiency and economic benefits. Because of the complexity of large heat pump systems, achieving optimal operation in practical project remains challenging. Model predictive control is a promising method for operation optimization. This paper represents a data-driven optimization framework based on model predictive control for optimal district energy supply. Four sub-models of the framework including heating load prediction model, heat pump performance model, main pump performance model and indoor state prediction model were built via machine learning and fitting methods. The particle swarm optimization algorithm was adopted to determine the optimal operation strategy. The framework’ effectiveness was verified by the operation data of a district energy system in Shanghai by means of both offline and onsite validation. The optimization results under low load conditions show that the energy-saving ratio can reach 9.43%, correspondingly, 12.37% for medium load conditions and 4.50% for high load conditions. In onsite validation, the Mean Absolute Percentage Error and Root Mean Square Error of heating load prediction are 6.72% and 104.21 kW, respectively. Compared with the day under similar weather conditions, the framework helps operators to obtain 12.9% energy-saving ratio in practice.

热泵系统因其高能效和经济效益而被广泛应用于区域供热。由于大型热泵系统的复杂性，在实际项目中实现优化运行仍然具有挑战性。模型预测控制是一种很有前途的运行优化方法。本文提出了一种基于模型预测控制的数据驱动优化框架，用于优化区域能源供应。通过机器学习和拟合方法构建了该框架的四个子模型，包括热负荷预测模型、热泵性能模型、主泵性能模型和室内状态预测模型。采用粒子群优化算法确定最优运行策略。该框架的有效性通过上海某区域能源系统的运行数据通过离线和现场验证的方式得到验证。低负荷工况下的优化结果表明，节能率可达9.43%，相应地，中负荷工况为12.37%，高负荷工况为4.50%。在现场验证中，热负荷预测的平均绝对百分比误差和均方根误差分别为 6.72% 和 104.21 kW。与类似天气条件下的一天相比，该框架帮助运营商在实践中获得了12.9%的节能率。高负载条件下为 50%。在现场验证中，热负荷预测的平均绝对百分比误差和均方根误差分别为 6.72% 和 104.21 kW。与类似天气条件下的一天相比，该框架帮助运营商在实践中获得了12.9%的节能率。高负载条件下为 50%。在现场验证中，热负荷预测的平均绝对百分比误差和均方根误差分别为 6.72% 和 104.21 kW。与类似天气条件下的一天相比，该框架帮助运营商在实践中获得了12.9%的节能率。