Short-term load forecasting method based on artificial neural network improved training strategy

Abstract

The invention discloses a short-term load forecasting method based on an artificial neural network improved training strategy. The method comprises steps: an Elman artificial neural network is built, and pretreatment is carried out on original data to acquire a training sample; through setting different thresholds and weights, multiple groups of training results are acquired; by comparing the acquired multiple groups of training results, a forecast data set with the minimal error is recorded; initial hidden layer neuron numbers of the Elman artificial neural network are reset, and the training sample is substituted in the network for training; and an artificial neural network model corresponding to the group with the minimal validation sample error is selected as a forecasting model. A practice of multiple related day historical load data by inputting only one time point traditionally is changed into a practice of multiple related day historical load data by inputting multiple time points. When model forecasting is substituted, an intermediate value in a forecasting value sequence is only selected to serve as a target value for forecasting of the time, and influences on forecasting by edge effects and data fluctuation are avoided.

Description

technical field [0001] The invention relates to the technical field of power system load forecasting, in particular to a short-term load forecasting method based on an artificial neural network improved training strategy. Background technique [0002] With the continuous development of social economy and the rapid growth of urban power grid load, the requirements of modern production and life for power supply quantity and power supply quality have increased significantly. On the other hand, the load changes in the core area of ​​the urban power grid are complex and affected by many factors, which pose a challenge to the safe operation of the power grid dispatching department. How to quickly and accurately grasp the short-term load change characteristics of urban power grids and carry out accurate load forecasting is extremely important. [0003] Short-term load forecasting takes the continuous load change trend in the next few minutes, hours or days as the research object. ...

AI Technical Summary

Problems solved by technology

Although the above method has been proved to have achieved fruitful results in the field of short-term load forecasting, it has the following disadvantages: because the load in the urban core area is greatly affected by relevant factors, the load in different areas presents different regularities, and the robustness of the forecasting method is poor; most of the current research All studies are conducted on a single regional power grid, and the load characteristics of different urban areas are not systematically compared. There are problems such as a large amount of data to be processed and rough analysis of load influencing factors, resulting in long time-consuming load forecasting, poor load forecasting accuracy, and lack of Practicality

The main reason is that after the model is fixed, the number of neurons and the weight threshold of the model are fixed, and the adaptability of the model is reduced. Because its change law is different from the training data, after encountering new data, the prediction accuracy is prone to decrease.

Images