Research On Unsupervised Modeling Method For Real Time Soft Sensing Of Building Space Cooling Load

The central air conditioner has great energy conservation potential in running stage by adaptive control based on real-time load,and the accurate real-time load is the basis of providing energy on demand.Traditional load calculating method has some weaknesses such as complex model,difficult modeling,and poor real-time ability.On the other hand,load prediction method needs plentiful priori data,and is difficult to include dynamic factors,which make it has negative effects on accuracy and practicability.Accurate space cooling load is basis for quantitative analysis for on-demand supply and direct load control of air conditioning operation process.This thesis is aiming at developing a soft sensing modeling method of real-time cooling load for air conditioning control.To realize the real-time calculate of cooling load,the building space is treated as research object,the refrigeration is treated as typical conditions,from the view of exciting-response mode,with the research on mechanism analysis,theoretical modeling,numerical simulation and experimental verification,a room temperature response characteristics based space cooling load soft sensing model is build to supply reliable theoretical and technical support for energy conservation in air conditioning operation process.In the process of room cooling load soft sensing,the following questions need to be solved:(1)how realize the room cooling load soft sensing with incomplete sample data;(2)how mine the potential knowledge to improve the soft sensing accuracy;(3)how suppress noise in the modeling process.The main work and research contents are as follows:1.Model for soft sensing of room cooling load under temperature control mode is put forward.Some basic concepts are distinguished firstly,and then the formation mechanism of cooling load and basic principle of air conditioner is introduced.Under temperature control mode,the cooling load is not equal to the heat extraction of air conditioning system,but it still subjects the room energy balance equation.Based on the relationship among the cooling load,the heat extraction,the heat storage by the energy balance equation and the relationship between heat storage and space temperature deviation,a real-time cooling load soft-sensing model is first put forward,with heat extraction and room temperature as anxiliary variables and cooling load as primary variable.2.Frequency domain decomposition method based real-time soft sensing for room cooling load is put forward.The frequency domain characteristics of cooling load is analysised firstly,and it is found that in frequency domain the room cooling load distribution is concentrated on the frequency point with a period of one day and its second harmonic,third harmonic.Aiming at the bottleneck that data are incomplete for parameter identification,with the room cooling load characteristics,to decomposite the room energy balance equation in frequency domain,then the parameter identification can be realized which avoid the dependence on un-measureable real-time coong load.The parameters are used to realize the soft sensing for room cooling load.It is also proved that the frequency domain decomposition method cannot suppress noise,and the observation noise will add to the cooling load soft sensing value directly.3.Deep learning based room cooling load prediction model is put forward.The neural network gain extensive attention and widely application for its strong nonlinear fitting and generalization ability.By exploring deep architectures,deep learning approaches are able to discover the hidden structures and features at different levels of abstraction from data.This paper further do study on prediction of room cooling load with deep learning in order to minimize the error caused by manual modeling.It first introduces and successfully applied the deep learning into the real-time prediction of room cooling load under temperature control mode.4.Improved particle filter based real-time state estimation of room cooling load is put forward.The particle filter method is an information fusion method.An improved particle filter model is proposed in this article.The state transition function is deep learning prediction model.The measurement function is the rought measuremtn model obtained with frequency domain decomposition.The artificial fish swarms algorithm is introduced to overcome the particles impoverishment problem in particle filter with Re-sampling.It is first time to successfully realize the room cooing load soft sensing based on room temperature response under temperature control mode,from the perspective of incentives-response in control mode,with the improved particle filter technology.The temporal predictability is introduced as index to evaluate the unsupervised modeling results,which also proved the effectiveness of the proposed method.