Demonstrating both energy conservation and comfortable environments through feed-forward AI air-conditioning control based on small-volume learning

We have made comfort prediction possible even with short-term data and demonstrated its effectiveness in the actual environment of the JR Shinjuku Miraina Tower

Nippon Telegraph and Telephone Corporation (NTT), East Japan Railway Company (JR East), NTT FACILITIES, Inc. (NTT FACILITIES), and NTT DATA Corporation (NTT DATA) jointly demonstrated the optimal air-conditioning control scenario calculation technology developed by NTT in the office lobby of JR Shinjuku Miraina Tower. As a result, we were able to reduce energy consumption by approximately 50% while maintaining a high level of comfort in the office lobby during the summer. This achivement is expected to contribute significantly to achieving carbon neutrality, which is the goal of both the NTT Group and JR East Group, on the basis of NTT's environmental energy visions, "NTT Green Innovation toward 2040"^*1, and the long-term environmental goals of the JR East Group, "Zero Carbon Challenge 2050."^*2

1. Background

Until now, air-conditioning control for common areas such as corridors, lobbies, and lounges in large-scale commercial and office buildings has been based on signals from sensors (feedback control) or the experience of building managers. However, with the growing momentum toward carbon-neutrality, it is difficult to take into account the time delay until air-conditioning control affects the indoor environment with the conventional technology. To take into account the time delay, data such as temperature, humidity, and human flow must be measured and analyzed over a long period of time, making it difficult to achieve a comfortable environment for visitors and tenants while reducing energy consumption in an actual building.
　Therefore, the NTT Group, which aims to achieve carbon neutrality by fiscal 2040, and the JR East Group, which owns a large number of buildings and is working to achieve ESG (environment, society, governance) management and improve the attractiveness of its buildings, have jointly conducted demonstration experiments aimed at both improving the attractiveness of buildings and optimizing energy consumption. In particular, we have worked together to optimize the operation of air conditioners, which account for approximately 50% of the energy consumption of buildings and strongly affect the comfort of visitors and tenants.

2. Outline and Results of Demonstration Experiments

In this experiment, we applied the optimal air-conditioning control scenario calculation technology developed by NTT to the air-conditioning operation of the office lobby in the JR Shinjuku Miraina Tower in summer. This technology was developed at the NTT Smart Data Science Center as part of the "Urban DTC"^*3 initiative to create new value in cities by reproducing the state of people, goods, and the environment and predicting the future. This technology uses the "4D digital platform™"^*4, which is a geospatial location information database that includes mobile objects. The roles of each company in this experiment are listed in Table 1, and the outline of the developed technology is shown in Figure 1.

This technology calculates the optimal air-conditioning control scenario on the basis of the following processing. In particular, the combination of computional fluid dynamics and machine learning enables comfort to be predicted from small amounts of data measured in a short period of time, and the use of deep reinforcement learning^*5 enables feed-forward control that takes into account the time delay before air-conditioning affects the indoor environment in large common spaces in buildings.

By applying the air-conditioning operation scenario calculated above, it was found that the amount of energy consumed by the air conditioner (chilled water heat) could be reduced by approximately 50 percent compared with the conventional air-conditioning operation on the same climate day, while the PMV stayed within a comfortable range. This is an important result demonstrating that the optimal air-conditioning control scenario calculation technology using deep reinforcement learning is useful for optimizing air-conditioning in actual buildings.

These results will be exhibited at the NTT R&D Forum^*7 to be held from November 16.

Figure 1: Overview of NTT's technology for calculating optimal air-conditioning control scenarios

3. Features of the Results

Figure 2: Energy Consumption of Air Conditioners (Chilled Water Heat) on Experimental and Reference Days

Figure 3: Change in mean PMV between the experimental and reference days

4. Future development

Going forward, the NTT Group and the JR East Group will jointly promote the expansion of these results to improve the attractiveness of buildings and reduce energy consumption. At the same time, NTT DATA and NTT FACILITIES will study the practical application of this service. In addition, the NTT Group and the JR East Group will take steps to achieve carbon neutrality, contributing to the Japanese government's goal of reducing greenhouse gas emissions by 46% by 2030 compared with fiscal 2013, as well as to satisfy the 2050 Carbon Neutrality Declaration.

Glossary

^*1 NTT group defined its Environment and Energy Vision "NTT Green Innovation toward 2040" on Sep 2021. Under this vision, NTT group aims to reduce 80% of greenhouse gas emitted by NTT group by FY2030 and achieve carbon neutrality by FY2040. NTT group contributes to the goal of Japanese government that aims to reduce greenhouse gas emission by 46% compared to FY2013 by 2030 and achieve carbon neutrality by 2050 through introduction of NTT's activity and technology to society.
https://group.ntt/en/newsrelease/2021/09/28/210928a.html

^*2"Zero Carbon Challenge 2050" is a long-term environmental goal that JR East has established in order to continue to be a corporate group that creates new value for society by improving its environmental superiority in the future.
https://www.jreast.co.jp/e/environment/pdf_2021/p066-077.pdf

^*3NTT, NTT Urban Solutions February 2, 2021
NTT Group's Digital Infrastructure for Realizing the Future of Community-Building: Development of Technology and Commencement of Demonstration Tests for Urban DTC™. https://group.ntt/jp/newsrelease/2021/02/02/210202a.html (in Japanese)
For more information about Urban DTC, visit:
https://www.ntt-review.jp/archive/ntttechnical.php?contents=ntr202101fa7_s.html

^*44D digital platform™ integrates various sensing data such as humans, things and environments in real time into high-precision spatial information, enabling fusion with various industries' platforms and the construction of future predictions. 4D digital platform™ would be one of the key elements of Digital Twin Computing, a part of NTT's Innovative Optical and Wireless Network (IOWN) initiative. We intend to leverage NTT R&D and NTT Group technologies and assets toward sequential commercialization beginning in FY2021, with future expansion through ongoing R&D efforts.
For more information, visit:
https://www.rd.ntt/e/4ddpf/

^*5Deep reinforcement learning: A method of machine learning using a neural network that reproduces the mechanism of human neurons and is characterized by its ability to learn optimal solutions through trial and error for a given environment.

^*6PMV (Predicted Mean Vote): A quantified measure of a person's thermal comfort. It is calculated from temperature, humidity, radiant temperature, wind speed, subject momentum and clothing, 0 is comfortable, positive in hot weather, and negative in cold weather. 75% of people feel comfortable if the PMV is within ±1.

^*7For more information about the NTT R&D Forum, which will be held from November 16, please visit the following site:
https://www.rd.ntt/e/forum/