Power System Short-term Load Forecasting Based On Neural Network Research

Short-term power system load forecasting relates to the power system scheduling and safety in production. With the rapid development of the power industry market process, the precision of forecasting of Short-term load forecasting of power system has a direct effect on economic benefit of industry department. Thus to find effective method has important meaning for enhancing forecast precision. This paper used a popular method---neural network to forecast short-term load of power system.For only one model cannot totally reflect the changing rules and information of power load, a combined NN forecasting model is constructed in this paper. This model is composed of AM-NN sub-model which bases on the additional momentum algorithm and QN-NN sub-model which ground on Quasi-Newton algorithm. By means of time-variant comprehensive weight, the model compromises two sub-models. Simultaneously, weather factors are led in the model. Adopting the strategy of rolling optimization makes this model be better at generalization and convergence. The combination model is used in the load forecasting simulating and testing in power grid, the results prove that it can effectively improve forecasting accuracy.At the same time, this paper constructed a short term load forecasting model based on improved PSO-IRNN neural network. On the basis of BP neural network, the increased deviation units and feedback unit make the network has the ability to store previous information, this model fits the rule which the short-term load forecasting has a strong correlation between the two adjacent periods.Meaning time the added deviation units can be more convenient to change the network outputs and join the experience and knowledge of the forecasters. It also can improve the prediction accuracy and save time. This paper used dynamic inertia weight adaptive learning factor and the quadratic term to improve the PSO optimization. The combination model is used in the load forecasting simulating and testing in power grid, the results prove that it can effectively improve forecasting accuracy.