Financial big data system based on online deep learning and market operation value quantification

Abstract

The invention discloses a financial big data system based on online deep learning and market operation value quantification. The system includes a person flow stay perception module, a shopping mall BIM space module, a shopping mall value super index evaluation module, a shopping mall value super index evaluation data set storage and instant training module, and a financial big data and shopping mall BIM visualization module. A subjective operation value quantized value is fitted through passenger flow data and business volume personalized data, an optimized convolutional neural network is trained in real time for processing a large amount of data sampled during operation, and real-time quantification of the operation value of a shopping mall is realized. According to the invention, fairness from objective data to subjective rating is ensured, data desensitization is realized, network training is convenient, and real-time monitoring and evaluation of managers are facilitated.

Description

technical field [0001] The invention relates to artificial intelligence deep learning and financial big data technical fields, in particular to a financial big data system based on online deep learning and quantification of shopping mall operation value. Background technique [0002] With the rapid development of online shopping, the economy of physical stores is showing a downward trend. How to scientifically and reasonably evaluate the value of physical shopping malls has attracted more and more attention from the market. Evaluating the value of shopping malls can help investors make decisions for investment analysis, strategic analysis and value-based management, and can also help management authorities effectively improve business decisions. The goal of financial management is to maximize the value. Therefore, whether each business decision of the mall is feasible depends on whether this decision is conducive to increasing the value of the mall. [0003] At present, the...

AI Technical Summary

Problems solved by technology

However, this method cannot reasonably correlate spatial information, and it is difficult to perform reasonable ratings for higher-level visual tasks based on spatial analysis technology, and it is impossible to perceive reasonable features and convey them as expected representations during training.

The target detection method based on the bounding box has the following problems: there is no method of combining CNN classifier and trajectory perception, the flow direction cannot be described according to the captured data under the timestamp, and the flow direction feature cannot be added to the hierarchical elements; due to the occlusion of the bounding box It is impossible to accurately describe the position of the human body, that is, it is impossible to describe the position of the key parts of the person (feet, pelvis). When blocked, the bottom surface is in the wrong position, so it is difficult to form a reasonable trajectory, and trajectory statistics cannot be performed

[0007] 2. The retraining of the network based on sequence prediction requires slow start of learning rate, annealing and other dynamic learning rate methods, which require high experience of the implementer, and are prone to problems of overfitting or too many neuron failures;

[0008] 3. The data obtained by conventional statistical methods lacks regularity in time series and space, and relies too much on off-the-shelf computer vision tasks and open source technologies, ignoring the original privacy features, and cannot be desensitized well during processing, resulting in data security issues

Images