Research on the Application of Time Convolution Series in Futures Price Forecasting

Abstract

Aiming at the problem of anomaly detection of time series data with unbalanced distribution among classes, a detection method based on depth convolution neural network is proposed. With the increasing trading scale of the futures market, it covers more and more economic and financial fields, and the volatility of the futures market is more and more intense, which constantly presents many complex phenomena that cannot be explained by other classical financial theories. Investors predict stock prices based on statistical analysis or simple machine learning methods, but because the stock market is a complex nonlinear dynamic system, these methods have huge limitations. By using the existing data and the chart obtained by the game master software for prediction analysis, the advantages of several technical indicators have been effectively complemented, and the accuracy of prediction has been improved. On the basis of the concept of attribute set and attribute measure, aiming at the problem of pattern recognition, this paper proposes a recognition method, establishes an attribute mathematical model, and combines it with other methods, which has been gradually applied to artificial intelligence, neural network and other fields. It has been successfully applied in the prediction of futures price trend. Building a futures price forecasting model to reveal the inherent law implied in futures market price index and show its evolution mechanism can improve the control ability to deal with economic and financial risks, and provide an objective and rigorous basis for regulatory authorities to formulate relevant policies, which is also the significance of this study.

Improved Algorithm with YOLOv5s for Obstacle Detection of Rail Transit

“…4, the KM-Decoupled Head, a decoupled detection head, follows a specific process. First, a 1×1 convolution [29] is applied to the input feature map to reduce the number of channels and parameters generated. Subsequently, the output feature map is split into two branches to address the classification and localization tasks separately.…”