AI can predict with much greater accuracy than existing forecasting methods, including those using traditional machine learning methods.
With our AI based analytics, airports and airlines can draw on historical and current data to get a reliable handle on what is going to happen in the future. Armed with this foresight, it’s possible to make accurate, informed decisions about how best to deploy resources to match demand and keep passengers moving.
Tolerances in air travel are narrow. By the time operations have registered and reacted to a situation, it may already have had significant knock-on effects. But our AI makes it possible to get ahead of the game, opening the door to more robust resource planning and better customer service.
The Aurora-AI Approach
We have designed and deployed analytics that can forecast passenger, vehicle and baggage flows based on input such as flight schedules and passenger numbers.
Traditional modelling techniques are limited in their ability to make truly accurate predictions, because they are driven by pre-determined algorithms that can only give a partial picture.
Something as routine as an arrival gate change brings into play many factors which may be impossible to account for in the normal planning process, with everything from trolley availability to where passengers can take a comfort break impacting when they will turn up at immigration desks, luggage halls and taxi ranks
We have proven that AI can understand these complex relationships, without ever needing to be taught them specifically. What’s more, over time as additional data is processed, the AI learns new behaviours and constantly improves and adjusts the predictions.
Immigration Hall Flow
A key part of the passenger journey that greatly affects passenger perception of an airport is the ease of transition through passport control. Changes to arrival times are inevitable so it’s crucial to have accurate insight showing when surges are likely to occur so that mitigation can be put in place. Simply put, our AI solution can provide better management information to allow immigration staff to be deployed when they are needed. Passenger experience is improved because AI has been used to deliver accurate information to the point of decision.
Car Park Occupancy
For airport operators, getting accurate forward visibility of passenger movements can be incredibly valuable. It enables resources to be allocated appropriately and services to be adjusted to meet demand at peak times.
We have developed a dedicated AI that draws on various inputs to provide data on future car park occupancy levels. This predictive model provides key insights on particularly busy points in the day and can be used to optimise vehicle flow and manage staffing requirements.
Using AI to blend historical data with live changes
To achieve the high levels of accuracy in these predictions, historical information around passenger and vehicle numbers and flight information was combined to create a neural network tuned to the behaviour exhibited.
After training, the AI understood the effect of each subtle change and, when introduced to live data, could adjust the predictions in real time, throughout the day. In these cases, the AI was found to be considerably more accurate than the forecast method previously used.
We can gather information from the various systems which generate data relevant to the predictive task and add changes, live or simulated, to predict demand for services or assets and also provide an output that can be easily integrated into the existing information systems in daily use by staff.
Using reliable predictions generated by Aurora-AI can help Operations to confidently make the best decisions based on an up-to-date view of future flows.
Creating and Using AI based Predictive Analytics
Our simple, cloud-based service model enables speedy development of these highly accurate forecasts. Based around our research projects, the programme of development tests hundreds of different neural networks to determine which is the most appropriate for a specific prediction, ensuring the most efficient neural structure is chosen. These models are then deployed to our Cloud service to enable bespoke training for each problem, automatically.
Once the prediction requirement is defined, a representative sample of data is obtained and training commences. The accuracy of the predictions is monitored against some of the data reserved purely for testing. Once a positive correlation is shown, forecasts can be generated. Training the network continues as new forecasts are generated, meaning that the solution will continue to improve over time.
The Cloud service is offered as a Representational State Transfer (REST) API. We provide full details on the interface using Swagger, which can also be used to generate code samples to enable integration of the interface in the client application, covering most development languages. This service-based approach means clients do not need to deploy their own resources, saving cost and complexity.
Our strength is in provision of high-quality forecasts. These are simply integrated into existing dashboards so there is no need to change existing service provision – just benefit from the improved data that staff already understand. The predictions can be used to model scenarios as part of the Digital Twin, or make live operational decisions in response to changing events as they happen.
LinkedIn Company Page
Follow our company page and keep up to date.
Whitepaper: The Benefits of Infrared Face Recognition
Overcoming the limitations of image capture in natural light.