Power Engineering

Optimization of heating substation through neural network

For the local heating substation we deployed artificial neural network to provide advanced control and regulation of heating systems, leading to significant improvements in efficiency and environmental impact.

A local heating substation was experiencing inefficiencies in its energy management, leading to:

  • High operational costs
  • Inconsistent heat distribution
  • Increased carbon emissions

Currently used in heating substation controllers standard control algorithms with proportional (P), proportional-integral (PI) and proportional-integral-derivative (PID) operations ensure good control quality only in the case of cooperation with linear objects. Unfortunately, regulation processes are dynamic, which traditional algorithms cannot cope with. Regulation becomes slow and the system does not respond to emerging disturbances. This leads to incorrect operation, heat loss, and increased heating costs.

We were tasked with finding a solution that could enhance the efficiency of the heating substation, reduce costs, and improve environmental sustainability.

The Solution: Neural Network

Large changes in control parameters (this applies especially to flow heat exchangers, and buildings) prompted us to apply
adaptive control. The opportunity was to use an artificial neural network.

The main task of the neural network is to improve the quality of regulation in a wide range of system operation, without the need to change setting parameters. This applies to such indicators determining the quality of regulation, such as: period oscillation, the regulation time in which it is achieved set value, the number of oscillations during regulation, disturbance suppression rate, overshoot, maximum deviation, and much more. The main advantage of the neural network is adaptation to changing conditions, and this is achieved through continuous training and improving the effects over working time, without any additional, external actions.

We designed an algorithm to control heating nodes for central heating, domestic hot water and process heat, powered by centralized heat sources. Its key characteristic is improvement of the effects of action over time. For example, if the purpose of the building changes, the network after a short time will learn  new conditions without any additional external activities.

The neural network is an open system. The action is conditioned by a prior state, so there does not have to be a control deviation, for the network to be activated. A well-designed and trained network can obtain a new state of equilibrium by just one move of the actuator, which is not possible in traditional control systems.

Technologies Used

We leveraged Python to develop the neural network, using libraries like TensorFlow for its machine learning capabilities.
C++ was added for real-time system performance. For seamless communication between system components, we relied on gRPC and Protocol Buffers, ensuring fast data exchanges.

On the interface side, we built a sleek web-based dashboard using Vue.js with Vuetify, allowing operators to easily monitor and control the system. PostgreSQL served as our database for efficient storage of large datasets, critical for training and improving the neural network.

Testing Process

We performed extensive testing to guarantee the system’s reliability. pyTest was used for unit testing, ensuring each part of the neural network worked flawlessly. Integration was smooth thanks to Jest, confirming compatibility between sensors and controllers. For stress testing, Locust was used to simulate real-world conditions, making sure the system could handle heavy data loads.

Finally, once deployed, Sentry allowed real-time error tracking to continuously monitor and improve the system’s performance in the long run.

The effects of implementation

The local heating substation experienced significant benefits following the implementation of the neural network designed by IDO Cloud.

Energy Cost Reduction & Return of Investment

Monthly energy bills were reduced by approximately 25%, resulting in substantial cost savings over time. The system was recouped within 18 months.

Improved Heat Distribution

Consistent heat distribution was maintained, enhancing the comfort of end-users.

Lower Carbon Footprint

The optimized heating system led to a 20% reduction in CO₂ emissions, supporting environmental sustainability.

Client's Feedback

The management of the heating substation expressed high satisfaction with the results. They highlighted the ease of integration and the significant impact on both operational costs and environmental sustainability. Our self-teaching neural network not only met but exceeded their expectations, proving to be a valuable addition to their infrastructure.

This site is registered on wpml.org as a development site. Switch to a production site key to remove this banner.