Research On Supper Short-Term Forecasting For Uncontrolled Remaining Load Of Micro-Grid

Accurate prediction of the uncontrollable remaining load is the basis of micro-grid energy management to achieve effective control with a significant impact of reliability, security and sustainable development to electric power system. However uncontrollable micro-power is an important component of distributed energy, which is mainly wind turbines and photovoltaic panels. The wind and solar power generation can’t be controlled effectively and artificially because of its constraint with environment. So, in order to ensure stable power supply micro-grid, the volatility of power generation must be considered when uncontrollable micro-power access to micro-grid. Meanwhile, from the direction of the economy, we need to consider the cost of micro-grid power generation, uncontrollable micro-power generation forecast and all electricity load. Then the microgrid power generation plan is formulated to meet the basic daily needs. Therefore, this paper research on forecast of the uncontrollable remaining load.Firstly, to ensure the reliability and validity of forecast data, making more accurate prediction, this paper propose an improved Grubbs detection algorithm. The variance of three previous data is amended with combination of graduation and Grubbs detection algorithm to avoid the situation of finding false anomaly or ignoring true anomaly in follow-up data. Next, aimed to solve the defect that traditional PSO algorithm easily fall into local optimum, a new CPSO algorithm is proposed which the chaotic variable is introduced into the PSO algorithm to achieve stable particle becomes active. Active particles tend to be more stable with the increasing of iteration. And active particles can increase their diversity, jump out of local optimum in solution space and enhance the capability of global optimal search. Finally, for the environmental needs of micro-grid uncontrollable remaining load forecast, this paper proposes average weighted Online Sequential Extreme Learning Machine(MM-OS-ELM) prediction model which the number of learning machine is confirmed by the new CPSO algorithm. In order to make historical and new training data play different roles, the weight of old and new data is added into Online Sequential Extreme Learning and averaging numbers of predicting outcomes to increase stability of predictions.Using the true load data and environment data from UCI standard test datasets to perform experimental tests and numerical examples and comparing with ELM, OS-ELM and OS-ELM-RMPLS algorithm. Experimental results show that the proposed remaining uncontrollable load forecasting accuracy and stability of the algorithm is superior to comparative algorithms, and it can provide reliable and effective basis for the remaining load forecasting micro-grid.