Automatic prediction method, device and system for multi-target time sequence

Abstract

The invention discloses an automatic prediction method, device and system for a multi-target time sequence. The method comprises the following steps that data reading, data abnormal value automatic detection and data feature automatic conversion are completed through a DataManager assembly; the automatic adjustment of algorithm parameters and the automatic construction of a model are completed by using an AutoML_Core component; a Model(s) component inherited from the AutoML_Core component is utilized to complete automatic construction, calculation and management of a plurality of models from three dimensions of days, months and quarters; multi-target and multi-model evaluation and result management are completed by using an EvaluateModes component; the management of the model and the automatic selection of the optimal model are realized by utilizing a ModelSelectionManager component; and the market sales volume prediction is realized by utilizing an Estimator component. In the automatic prediction process, a ModelPipeline component and a TableTemplate component are used in a matched mode. Through mutual cooperation of a plurality of components, timely, accurate and automatic market prediction can be carried out on a plurality of products of an enterprise, and the enterprise can be assisted to make rapid and reliable decisions.

Description

technical field [0001] The invention relates to the field of product sales forecasting and decision-making, in particular to an automatic forecasting method and device for a multi-objective time series. Background technique [0002] With the development of market economy and the globalization of economy, the competition among enterprises is becoming more and more fierce. If enterprises want to gain advantages and win customers in the fierce competition, they must provide products to customers at the fastest speed and at the lowest cost. This makes timely and accurate product sales forecasts and the resulting reliable decisions a Key elements of modern business success. [0003] In the process of market sales, an enterprise will face the sales of dozens or hundreds of product lines internally, and the production of each product will face the problem of overproduction or underproduction. The direct problem is slow sales or the output does not meet market demand quantity. Un...

AI Technical Summary

Problems solved by technology

Qualitative forecasting methods include senior manager’s opinion method, salesperson’s opinion method, buyer’s expectation method, Delphi method, etc. These methods mainly rely on rich experience, and the prediction deviation is large and not accurate enough, and the timeliness of prediction is not timely enough. Unable to respond and change in a timely manner to unexpected situations, resulting in a lag in the company's strategic adjustment

Quantitative forecasting is an algorithmic method including time series, but the parameters of the algorithmic model need manual intervention to select, and before the model is built, complex operations such as data transformation are required. After a lot of manual intervention, the final To construct a model, if an enterprise has hundreds of products to be predicted, human intervention is required to carry out multiple repeated practices to complete data analysis and model construction, and the time period is relatively long

[0005] It can be seen that the above-mentioned existing time series-based forecasting methods obviously still have inconveniences and defects, and need to be further improved.

Images