The basic process of load forecasting

The key to load forecasting is to collect a large amount of historical data, establish a scientific and effective forecasting model, use effective algorithms, take historical data as the basis for a large number of experimental studies, summarize the experience, and constantly amend the model and algorithms to truly reflect the pattern of change of the load. The basic process is as follows. After the initial sorting, but also for the data used in data analysis preprocessing, that is, the smoothing of abnormal values in the historical data and the missing data addendum, for abnormal data, the main use of horizontal processing, vertical processing methods.

The horizontal processing of data that is, when analyzing the data, the load data before and after the two times as a benchmark, set the maximum range of variation of the data to be processed, when the data to be processed exceeds this range, it is considered bad data, using the average value of the method of smoothing the changes; vertical processing of data that is, in the pre-processing of the load data to take into account the small cycle of 24h, that is, the load should have a similarity of the same time of the day on different dates, the load should have a similarity of the same time of the day. Vertical processing of the data, i.e., the load of the same moment on different dates should be similar, and the load value of the same moment should be maintained within a certain range, and for the correction of bad data out of the range, it is the average value of the load of the last few days of the moment of the data to be processed. Load forecasting model is a generalization of the trajectory of the statistical data, the forecasting model is diverse, therefore, for specific information to choose the appropriate forecasting model, which is a crucial step in the process of load forecasting. When the prediction error is too large due to improper model selection, it is necessary to change the model, if necessary, can also be used at the same time several mathematical models for computing, in order to compare and choose.

After selecting the appropriate forecasting technology, the establishment of load forecasting mathematical model, forecasting work. As from the development of the law of change that has been grasped, does not represent the future law of change, so to analyze the new factors affecting the forecasting object, the forecasting model to determine the forecast value after the appropriate correction.