User-side Micro-grid Power Forecasting Method With Distributed Photovoltaic Access

Along with the new urbanization construction in our country, the power demand increases and creates the environment for distributed energy access. Taking distributed photovoltaic integration application through the user-side micro-grid is a realization method for new energy absorb locally, reducing carbon emissions and environmental pollution. Relevant national policy has planning towards the rapid development of distributed photovoltaic for a period of time in the future. State Grid Corporation of China also proposed policies making convenience and related technical support for distributed photovoltaic access. In order to make sure the operation stability and economy of micro-grid and distribution network with distributed photovoltaic access at user-side micro-grid, the micro-grid load forecasting technology and distributed photovoltaic power output forecasting technology need to be deep research combine with the application characteristics of user-side micro-grid.The power forecasting model is presented based on extreme learning machine with kernel, and using particle swarm optimization (PSO) algorithm to optimize the model parameters at off-line part. For the online power forecasting system, the proposed method ensures the prediction accuracy while reducing the model complexity, thus combining the off-line parameter optimization and online power forecasting in this power forecasting method. First discusses the current situation of power forecasting technology at home and abroad, then briefly discusses the related theoretical basis of extreme learning machine and PSO algorithm.(1) Considering the bigger load fluctuation of micro-grid, use the time-division training samples for parameters optimization, getting the optimal parameters for each moment in a day. To improve the operation efficiency of load forecasting system, only chooses the history data from high related time session in same day type to build the model. Taking respective load forecasting of four micro-grid that has average load of140kW to1300kW with different customer types for1month, the weekly prediction error is usually less than10%and the biggest is not more than15%. The micro-grid load may has a faster growth in a relatively short time, the parameters of forecasting model is periodic update in the research, and the original load forecasting accuracy after update can be maintained.(2) For the distributed photovoltaic power output forecasting, Taking training samples filtration mechanism based on attributes weight to decrease the building complexity of forecasting model. The forecasting method based on the low-cost weather recorder without numerical weather prediction, taking1month power output forecasting for the distributed photovoltaic system with dozens of kilowatt, the prediction error is about16%to18%. Meanwhile, the forecasting model can be simplified based on attribute weights to further reduce the computation time and having similar prediction accuracy. When the distributed photovoltaic has random dust overlaying or inverter partial failure, the forecasting model can keep the original accuracy and efficiency without human intervention or parameters update.