Research On Short-term Power Load Forecasting Based On Cloud Computing And Smart Algorithms

People needs high quality of electricity with the development of the society, so the power sector has to provide the user of reliable and high quality of electricity. But there is no equipment can store electric power with a large scale, we have to make a balance between the power supply and load. The short-term power load forecasting is the foundation of security and stability of the power grid, so using the historical load data to predict the load of the next few days is worth studying. A lot of research show that the power load is chaotic, it's difficult to have an accurate forecasting with conventional methods. But we can restore the high dimensional hidden information in the one-dimensional time data sequence by reconstructing phase space, thus we can have an accurate power load forecasting. This paper build an accurate power load forecasting model by reconstructing phase space of the historical data at first, then combine with the ELM and SVM. For the big load data in the power system, this paper designed two parallel algorithms in the Spark platform to speed up the historical data processing. The experimental results show the effectiveness and feasibility of these models. This paper mainly studied the following contents:This paper gave a short-term load forecasting method combine the phase space reconstruction and ELM through the load data have chaos characteristics. To build an accurate short-term load forecasting model, reconstruct the phase space of the historical load data samples by calculating the phase space reconstruction parameters, so the new samples reflect the variation of load characteristics more accurately. The simulation results show that this method have more precise forecast ability compared with traditional forecast methods.Combining the phase space reconstruction with SVM in short-term load forecasting. A focus random search optimization algorithm is given for input parameters optimization of the model by using the mapping problem of linear kernel function instead of the original nonlinear problem, using this method for short-term power load forecasting, The simulation results show that this method have more precise forecast ability compared with traditional forecast methods.For the big load data in the power system, this paper designed two parallel algorithms in the Spark platform to speed up the historical data processing. Building a Spark computing cluster with one master node and seven data nodes using the equipment in our laboratory. Using laboratory equipment built to contain a master node, the Spark of the seven data nodes compute cluster, the simulation results show that these methods have faster processing speed compared with the commonly in huge amounts of data processing.