Stochastic Optimization Of Integrated Energy System Based On Probability Density Prediction

Integrated Energy System(IES)has the characteristics of flexible operation and scheduling,green,low-carbon environmental protection,and is of great significance for solving the problems of energy shortage and environmental crisis.With the rapid development of wind and photovoltaic renewable energy technology,large-scale access of renewable energy has further enhanced the characteristics of low-carbon and environmental protection of the system.However,it is inevitable that the access of renewable energy makes the system need to respond to changes in end-load demand on the one hand,and increase the impact of renewable energy volatility on the system on the other hand.The stable operation of the system is subject to both "source-load" pressure.To solve the above problems,based on the historical load and weather data of a real campus,consider designing an integrated energy supply station for the campus,and make improvements from the two aspects of prediction and optimization respectively to guide the stable operation of the energy supply station.The main contents of this thesis are as follows:(1)According to the historical energy demand of a campus and considering the climate characteristics of this region,an integrated energy supply station with compressed air energy storage(CAES)was designed to meet the campus’ s cooling,heating and power load needs.The integration of CAES into IES makes the thermoelectric coupling characteristics of the system further enhanced and the efficiency of energy cascade utilization further improved.At the same time,the mathematical model and working mechanism of wind turbine,photovoltaic generator,internal combustion engine,CAES and air-source heat pump(ASHP)are analyzed.The above system design and analysis provide model basis for subsequent prediction and optimization.(2)Aiming at the historical park load and weather data,a hybrid probabilistic forecasting model based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)-long short term memory(LSTM)network-gaussian process quantile regression(GPQR)is studied in this thesis.The above models are implemented based on the idea of "decomposition-point prediction-probabilistic forecasting",which can obtain the probabilistic forecasting interval of "source-load" multidimensional data.The experimental results prove that the model proposed in this thesis has high prediction performance in both point prediction and probabilistic prediction,and the prediction results can provide data basis for the subsequent stochastic optimization.(3)Aiming at the influence of "source-load" uncertainty on the stable operation of the system,a day-ahead scheduling optimization method based on stochastic optimization is studied.Day-ahead scheduling optimization firstly generates typical scenes by using K-means clustering method on the basis of probabilistic forecasting interval,so as to reflect the characteristics of "source-load" data uncertainty,and then configs the capacity of key equipment in the system.Then,on this basis,the above data is input into NSGA-Ⅱ method to realize stochastic optimization.In order to prove the authenticity of the day-ahead scheduling optimization method proposed in this thesis,a comparative experiment is carried out.Experimental results prove that under real load and weather conditions,compared with other scheduling optimization results,the accuracy of operation scheduling results of key equipment proposed in this thesis can be improved by up to 26.7%.The above optimization method can guarantee the stable operation of the system,improve the system’s renewable energy absorption rate,and avoid the influence of large-scale fluctuations of key equipment on the energy supply stability.