Method Towards The Dynamic Total Cooling Load Disaggregation Of Buildings Based On Sparse Coding Algorithms

The building cooling load is one of the most important indices to evaluate the practical energy efficiency of a building at the use stage.Total cooling load is employed to evaluate the entire thermal performance of building while the component of cooling load is used to analyze the contribution of building components to energy saving.Being different from the total cooling load,however,the component of building cooling load is difficult to be acquired for the lack of an effective measuring method.Thus,there remains a technical barrier to achieve the contribution of building components for energy saving in practice.Since each component of the building cooling load is generated dynamically in different heat transfer mechanisms by different kinds of heat disturbances,the variation characteristic of its continuous wave in time domain is different from each other.This paper proposes a concept to extract the cooling load components from the total cooling load by utilizing the difference in variation.A method of data disaggregation based on sparse coding is put forward,in which a non-negative dictionary learning algorithm is used to represent the characteristics of each component,and a non-negative sparse coding algorithm is adopted to disaggregate the total cooling load into its components.The simulated total hourly cooling load of reference building was disaggregated with the method and the estimated component of the hourly cooling load was obtained correspondingly.Compared to the simulated components,the disaggregation result shows an acceptably good agreement.The feasibility of the method was then validated.What's more,when selecting the building hourly cooling loads under different weather conditions and with different building-related parameters as the training data,since the hourly total cooling loads of buildings were well disaggregated,the training data of the method was verified to be available with different weather parameters and different building types.Furthermore,discussions which focus on the amount of training data and the running time of algorithms show a good efficiency of the method.