Fine-grained Photovoltaic Load Forecasting Method Based on Temporal Convolutional Network

Abstract

The invention relates to a fine-grained photovoltaic load forecasting method based on a temporal convolution network, and belongs to the technical field of smart grids. The steps are as follows: collect weather raw data, assign a floating-point value to each raw weather data according to the comprehensive solar radiation data collected at the photovoltaic site, and perform encoding processing. Other raw data expressed as floating-point numbers will not be processed to obtain the final weather characteristic data. Set the collection frequency and threshold of raw load data, and set the power window length by comparing the collection frequency and threshold. The front end is the input layer, after the residual block and the Flatten layer, and then through the fully connected layer, and finally the prediction result is obtained, which is the temporal convolutional network. Input it into the constructed time convolutional network to get the prediction result. The prediction effect is good, and the calculation can be parallelized based on the convolutional network, which greatly improves the training speed.

Description

technical field [0001] The invention relates to a fine-grained photovoltaic load forecasting method based on a temporal convolution network, and belongs to the technical field of smart grids. Background technique [0002] Because it is difficult to store a large amount of electric energy and the demand for electric power changes all the time, it is required that the power generation of the system should be dynamically balanced with the change of load. Improving the accuracy of load forecasting is conducive to improving the utilization rate of photovoltaic power generation equipment and the effectiveness of economic dispatch. Power load data has the characteristics of time series and nonlinearity. Based on its characteristics, the research on short-term power load forecasting models at home and abroad is generally divided into two categories. One is time series analysis methods, such as regression analysis, exponential smoothing model, Kalman filter method, autoregressive in...

AI Technical Summary

Problems solved by technology

However, due to the principle characteristics of the LSTM algorithm, it cannot perform parallel computing, the training speed is very slow, and it consumes a lot of computing resources.

Images