Distributed Neural Network Training Server for Building Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current environment control systems in buildings face challenges in efficiently generating and improving predictive models of neural networks for effective environmental control, particularly in automating the generation of samples for training and allowing for improvements during the operational phase.
Innovation Solution
A training server is introduced that receives training data sets from multiple environment controllers, determines reinforcement signals based on execution metrics, and updates the neural network weights using a training engine. This updated predictive model is then transmitted back to the environment controllers for improved environmental control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional training methods are used to generate predictive models, then the model can be trained with sufficient samples, but the process requires significant manual effort and time for sample generation
Solution Approach 1:
The system enables automated self-training by allowing environment controllers to autonomously generate training samples from their operational data and transmit them to the training server, eliminating the need for manual sample collection and preparation while improving training efficiency
Solution Approach 2:
The system collects and stores operational data from environment controllers in advance during normal operation, preparing training samples before they are needed for model training, thus reducing the time required when training actually occurs
2Manufacturing precision
If the predictive model is updated frequently to improve accuracy, then the environmental control becomes more effective, but the complexity of the training system increases
Solution Approach 1:
The training system is segmented into distributed components where multiple environment controllers independently contribute training samples to a central training server, allowing frequent model updates without requiring complex centralized coordination or increasing overall system complexity
Solution Approach 2:
The training server implements a universal training framework that can process diverse training samples from multiple environment controllers using the same neural network architecture and training algorithm, enabling frequent accurate model updates without increasing system complexity
Data Source
AI summary
Interactions between a training server and a plurality of environment controllers are used for updating the weights of a predictive model used by a neural network executed by the plurality of environment controllers. Each environment controller executes the neural network using a current version of the predictive model to generate outputs based on inputs, modifies the outputs, and generates metrics representative of the effectiveness of the modified outputs for controlling the environment. The training server collects the inputs, the corresponding modified outputs, and the corresponding metrics from the plurality of environment controllers. The collected inputs, modified outputs and metrics are used by the training server for updating the weights of the current predictive model through reinforcement learning. A new predictive model comprising the updated weights is transmitted to the environment controllers to be used in place of the current predictive model.