Research on strategies to improve model accuracy based on incomplete time series data

Aiming at the problem of inaccurate model prediction caused by incomplete time series data, an improved denoising autoencoder was proposed to supplement missing data. Firstly, the convolutional neural network is added into the encoding and decoding of the denoising autoencoder for sequence analysis. The missing values are completed by making full use of the spatial correlation in the time series and used to build models. Secondly, real photovoltaic data are used to train and test the model. In the training process, instances with random missing values are used as the verification set, and the model with the best generalization ability is selected. Then, the performance of the training strategy against instances with missing values of different granularity is tested, which proves the generalization and robustness of the algorithm. Finally, under the unified standard, the model accuracy of this method is improved by 32.64% based on the original model, which verifies that the algorithm improves the model accuracy after data filling, and verifies the feasibility and efficiency of the proposed missing value filling algorithm. The data filling algorithm in this paper greatly improves the problem that the performance of the prediction model deteriorates due to data loss caused by various reasons in photovoltaic power stations.

I. Introduction

At present, there are two main data prediction model methods., one is to use statistical methods to predict; the other is based on the background of the rapid development of big data and artificial intelligence technology, using data-driven machine learning methods to make predictions. For the sake of efficiency, prediction accuracy and practicability, the latter is more practical in current research. However, under this general trend, how to improve the accuracy of the prediction model is a problem that needs to be solved. The solutions to this problem include include optimizing model construction and improving the quality of data cleaning. Therefore, this paper aims to improve the low accuracy of the prediction model for incomplete time series data.