Study On Short-term Micro-grid Load Forecasting Based On IGA_PSO RBF Neural Network

As a good solution to the integrated and distributed power supply, micro-grid fits well with the demand of China’s power industrial development at present and is a potential selection to solve the problem of China’s power industry. Short-term load forecasting is an important part of micro-grid’s scheduling system, it’s prediction accuracy directly affects safety, economic and high-efficiency operation of micro-grid system. Therefore, this topic is of great significance to research the short-term load forecasting of micro-grid system.In this topic, the micro-grid system is based on an island. After an in-depth analysis about characteristic of load time series, short-term load forecasting model based on radial basis function neural network is established. In order to ensure the reliability of data, load data preprocessing is necessary, including: For the missing data, this topic use similar selection methods for completion, which avoids the impact on load forecasting accuracy by missing data; In view of the limitation of the abnormal data processing in one dimension, the topic adopts the horizontal and vertieal data smoothing method to proeess abnormal load data; The load data is normalized, which avoids saturation of neurons.In order to optimize the parameters of model, Particle swarm optimization(PSO)algorithm is introduced. Due to PSO algorithm has the problems of low convergence speed and sensitivity to local convergence, crossover、mutation in genetic algorithm and memory recognition 、 immune selection in immune algorithm are introduced to improve it. Then IGA_PSO algorithm is proposed and used to optimize the parameters of model.The experimental results show that the optimized prediction model has a higher precision, a faster convergence and a better stability.Considering the limitations of static prediction model, fuzzy control is introduced to improve it. The improved model uses the relative error and relative error rate as input for fuzz controller and modifying factor as output, then dynamic prediction model is established. The experimental results show that the model improved by fuzzy control has a higher precision and the relative error remain within 3%, so the forecasting system is more practical.Finally, as to make system has a better interactivity, web page of short-term load forecasting module isimplemented which is based on web framework of nodejs’ Express、MMDB Redis as caches and No SQL Mongo DB as data storage layer.