Photovoltaic Power Prediction And Time Shift Of Agricultural Greenhouse Load In Southern Xinjiang

Photovoltaic power generation has the characteristics of randomness,volatility and being affected by the weather.The capacity of microgrid in the shed is limited and the photovoltaic output power fluctuates greatly.The line voltage of microgrid is affected by the randomness of photovoltaic output and load change,resulting in large line voltage fluctuation.In this paper,a photovoltaic power prediction method is proposed.According to the prediction results,the power transfer strategy of greenhouse load is studied to improve the utilization rate of photovoltaic and reduce the influence of photovoltaic randomness and volatility on the line voltage in the shed.The specific research contents are as follows(1)The influencing factors of photovoltaic power generation are analyzed.The influence of irradiation intensity,temperature,wind speed and humidity on photovoltaic power is studied.The nonlinear correlation coefficient between each influencing factor and photovoltaic power output is obtained by using the cosine nonlinear correlation measurement method of elm excitation function.According to K-means clustering method,the training data are divided into three similar day types: sunny day,rainy day and cloudy day.(2)The network structure of BP(back propagation neural network),Elman and LSTM(long short term memory)prediction models is analyzed and the photovoltaic power prediction simulation is carried out.The simulation results show that the prediction model with single hidden layer network structure represented by BP has fast calculation speed and low prediction accuracy;The multi hidden layer prediction model represented by Elman has high prediction accuracy and slow operation speed;The prediction model with long-term memory represented by LSTM has high prediction accuracy and operation speed between them.(3)In order to improve the speed of photovoltaic power prediction,this paper studies the width learning(BLS)prediction model with single hidden layer and lateral learning ability;In order to improve the accuracy of photovoltaic power prediction,a cascaded structure is added to the BLS network to make it have the ability of time memory.Therefore,a deep cascaded width learning(DCBLS)photovoltaic power prediction model is studied.The simulation results show that the DCBLS prediction model is the best in terms of operation speed and prediction accuracy.(4)According to the results of photovoltaic power prediction,a time shifting strategy of agricultural greenhouse load is proposed.Firstly,the security structure system of greenhouse microgrid is designed,and then it is divided into different levels according to the load characteristics,and the load time shifting strategy is formulated.The simulation results show that the photovoltaic utilization rate is improved and the voltage fluctuation of microgrid in the shed is reduced after the load time shift.