We use AI to predict air conditioning demand and optimise air conditioning operation and control.
We aim to achieve significant cost reductions.

THE AI CONTROLLER WILL BE RETROFITTED TO THE CONTROL EQUIPMENT WHILE MAKING USE OF THE EXISTING AIR CONDITIONING SYSTEM. BY USING ARAYA'S AUTONOMOUS AI TECHNOLOGY TO PREDICT THE REQUIRED AIR CONDITIONING CAPACITY (DEMAND) AND MATCH HEAT PRODUCTION TO DEMAND AND ELECTRICITY TARIFFS, WE AIM TO ACHIEVE SIGNIFICANT COST SAVINGS (AROUND 20%).

Air Conditioning Optimization

Solution for District Heating and Cooling (DHC) and Large Heat Source System

WE OFFER THE FOLLOWING SERVICES TO DISTRICT HEATING AND COOLING (DHC) AND LARGE HEAT SOURCE OPERATORS, INDIVIDUALLY FOR EACH BUILDING. (WE CAN ALSO PROVIDE CENTRAL AIR CONDITIONING FOR SOME BUILDINGS.

Air Conditioning Optimization

AUTONOMOUS AI FOR OPTIMAL CONTROL

AUTONOMOUS AI IS A METHOD IN WHICH THE AI ITSELF LEARNS BY TRIAL AND ERROR WHAT ACTION (AIR CONDITIONING CONTROL) IS BEST (COST SAVING).

Air Conditioning Optimization

When the AI takes action (e.g. turning on/off the number of heat production machines in operation), the state of the air conditioning environment (e.g. changes in heat storage and cold supply) changes, and the AI is rewarded if a good result is achieved (e.g. lower heat production costs).
▶︎The AI learns by trial and error to get more rewards.

TWO ADVANTAGES OF USING AI FOR AIR CONDITIONING CONTROL

Advantage 1: Use of a wide variety of input data

Weather forecast data and data obtained from IoT devices and surveillance cameras can be used as control inputs for more detailed control.

air conditioning

Advantage 2: Data-driven parameterisation

Multiple parameters such as set temperature and operation time can be predicted based on past and present data instead of empirical rules, and detailed optimum settings can be made.

Air Conditioning Optimization

Experiment 1: Reduction of annual heat production costs by approx. 20 % for large heat sources

Facilities where the demonstration tests were carried out

Demonstrations were carried out with the following large heat sources
Heat production machines: around 5
Heat storage tank: Yes
Customer: Commercial facility (approx. 60,000 m2)
Cooling water demand (daily average): approx. 200 GJ in summer, approx. 30 GJ in winter
Discounts are available for reduced usage during peak hours.
・Annual heat production cost: 20-40 million yen

Scope of the demonstration

Simulation of heat production only (demand forecasting is done using rule-based algorithms in the existing system)
Verification of heat production cost savings in the simulation

Results of the demonstration

The annual heat production costs could be reduced by about 20% (on simulation).
The cost savings are particularly high in summer, when heat production is avoided at more expensive times of the year.

Season Percentage reduction
Reduced electricity costs (winter) 5 to 10 per cent
Reduction of electricity costs (summer) 15 to 22.5

DEMONSTRATION EXPERIMENT 2: HIGHLY ACCURATE DEMAND FORECASTING IN DISTRICT HEATING AND COOLING (DHC)

Facilities where the demonstration tests were carried out

Demonstrations were carried out in the following district heating and cooling (DHC)
Heat production machines: around 5
Heat storage tank: Yes
Customers: Commercial facilities and office buildings (total: approx. 70,000 m2)
Cooling water demand (daily average): approx. 300 GJ in summer, approx. 30 GJ in winter
Annual heat production cost: 50-100 million yen

Scope of the demonstration

Forecasting of demand only (the forecast is displayed on the screen and instructions for heat production are given by the operator on a rule-based (manual) basis)
Forecasting with an error within the capacity of one heat production machine (approx. 1000MJ/30min)

Results of the demonstration

It was possible to predict demand with a very high accuracy at a practical level (on simulation).
This allows the operators on site to give instructions for heat production without delay.

Item Prediction accuracy after 30 minutes Prediction accuracy after 1 hour Prediction accuracy after 24 hours
Chilled water system load heat value 99.7% 99.8% 91.6%