Real-time demand prediction method and device and electronic device

Abstract

The invention provides a real-time demand prediction method and device, and an electronic device. The method comprises the steps of receiving a demand prediction request of a client; wherein the demand prediction request carries a target time interval and a target position identifier, and the target position identifier comprises at least one sub-position identifier; reading target historical datacorresponding to the target time interval and the target position identifier from a preset offline database; wherein the target historical data comprises demanded quantities in different time intervals corresponding to the sub-position identifiers; inputting the target historical data into a demand quantity prediction model corresponding to the target position identifier to obtain a predicted demand quantity of each sub-position identifier in the target time interval; wherein the demand prediction model is generated by training a plurality of models including a graph convolutional neural network. According to the method, the prediction request of the user can be responded in real time, and an accurate demand prediction result is predicted through the demand prediction model trained by theplurality of models including the graph convolutional neural network.

Description

technical field [0001] The present application relates to the technical field of neural networks, and in particular to a real-time demand forecasting method, device and electronic equipment. Background technique [0002] Demand forecasting is one of the core issues of various services in various service industries. For online car-hailing services, an accurate demand forecasting algorithm can help the platform improve vehicle utilization, improve order quality, guide driver scheduling, and regulate traffic Planning, avoiding congested road sections, etc. Existing demand forecasts are usually based on convolutional neural networks, such as simulating a city as a large map, dividing it into multiple rectangular grids of the same size, and using Conv2D to perform convolution operations on these grids to extract the city’s Local invariance is used as features, and finally these features are input into the fully connected layer to predict the next moment. This prediction method c...

AI Technical Summary

Problems solved by technology

Existing demand forecasts are usually based on convolutional neural networks, such as simulating a city as a large map, dividing it into multiple rectangular grids of the same size, and using Conv2D to perform convolution operations on these grids to extract the city’s Local invariance is used as features, and finally these features are input into the fully connected layer to predict the next moment. This prediction method can only be applied to the above-mentioned rectangular grid. If the grid is divided into polygons such as hexagons, it is difficult. Naturally apply the convolutional network migration to the new grid map, the model loses its flexibility, and can only predict the demand at the next moment, it cannot respond to the real-time forecast of user demand feedback, and the forecast result is not accurate enough

Images