Predictive Control For Solar Domestic Hot Water System Based On Machine Learning

Domestic hot water occupies relative large share of energy consumption in buildings,therefore it has become the key point in building energy-saving.The combination of solar and assisted heat source has the advantage of using solar to reduce energy consumption while at the mean time solving the problem of unstable and not timely hot water supply with better performance in demand response.As the system becomes more complex,the requirements for the control system is increasing.Predictive control has better performance by balancing supply and demand,but as demand side is closely related to user's behavior,the prediction of demand usually cannot achieve ideal results.To solve this problem,a machine learning me·thod for user's behavior prediction is carried out based on influencing factors analysis using real hot water consumption data from users.Related problems such as the judging criteria,unbalancing classes,machine learning algorithms,training dataset size,etc.were also discussed in this paper.A novel method to predict future hot water demand by predicting each individual's hot water demand separately and then sum up to calculate total hot water demand is proposed.The new method can increase hot water assurance up to 98.64% compared to traditional one,and therefore more suitable for using in predictive control.On the supply side,a method of using machine learning and quadratic curve fitting is carried out with less calculation and acceptable results.A predictive control method for solar hot water system is carried based on supply and demand side prediction,and related system model is built.Using systematic simulation,performance on hot water assurance,energy consumption and economy of different control strategy is calculated.The results show that using predictive control,24-hr hot water assurance can reach up to 98.64% while total energy consumption reduces by 46.45%,heat from solar reaches up to 44.75%,heat loss reduces by 22.04% and hot water supply cost reduces by 38.26%.This predictive control method can achieve performance improvement in many areas without much hardware change.