Machine Learning Models for Smart Cities

Smart city development will require the collection of large amounts of data and the ability to process, reason and make decisions about that data.

Artificial intelligence (AI) and machine learning (ML) will help make it possible to create an urban landscape that enables safe, efficient, convenient and self-optimizing traffic eco-systems, while dealing with highly increased complexity. As cities become “smarter”, data collected from sensors regarding energy consumption, traffic, sanitation, will all increase at a scale that makes it difficult for certain types of tasks to be done well by humans alone, or would be unthinkable without the aid of automated system.

“City authorities need to handle an overwhelming treasure of information and data,” explained Anita Mathieu, AI portfolio management in digital labs at Siemens Mobility, Intelligent Traffic Systems. “Humans can never track, analyze and understand these massive streams of data in which ‘needles in a haystack’ need to be found.” She said they need partners who understand the data that is produced by their field devices, data from fleets in their network as well as the influence of other relevant data like weather and then they need partners who turn traffic data into value.

Sohrob Kazerounian, senior data scientist at Vectra, a San Jose, California, provider of technology which applies AI to detect and hunt for cyber attackers, said AI can also be used to tackle problems like predicting traffic patterns, smart-grids, energy prediction and services for which allocation can be done in response to real-time conditions. “These are all application areas that are not simple to achieve–or at least, to achieve well–without the aid of algorithms that can learn and adapt on their own,” he said.

Kazerounian noted the primary challenges facing ML in the area of smart cities would be privacy, efficacy, and necessity. “The data collection required to make smart-cities work, could become a privacy nightmare,” he explained. “Should cities, for instance, have access to the geo-location and movement data of people in that city? If so, how can it be anonymized? Can it be maintained securely? Who is responsible if there are data breaches? The list of questions goes on.”

When it comes to efficacy, Kazerounian said it will be important to know what, if any, regulatory mechanisms will be in place to determine how well AI systems that are deployed really perform. “In the same manner as autonomous vehicles, we will have to start asking who is responsible or culpable, when AI systems fail?” he said.

Vishwas Shankar, research director for mobility and associate fellow of automotive and transportation at Frost and Sullivan, noted between 2020-2030 there would likely be a growth in AI and ML applications in transportation. “Be it smart mobility as a part of the smart city framework or the smart city infrastructure itself, AI and ML adoption will become more widespread,” he said. “From a road or city standpoint we are expecting first adopters to be traffic signals, followed by road intersections.”

Shankar said the role of companies like Siemens, or other tech companies, is now expanding within the topic of “digital twins” that allow for a simulation of a real time environment and allow AI and ML developers to play around dynamically to arrive at best cases for optimized solutions for adoption. He noted a number of legacy systems on mainframes and servers are being moved to the cloud, which would allow for cloud enabled analytics capabilities that could power AI and ML technologies at a rate required to move at the speed of traffic. “Machine learning and AI can function well in silos,” he explained. “The challenge is in bringing the siloes of excellence to come together to talk to each other. However, the high cost of adoption, and piloted operations makes widespread integration challenging.”

Mathieu noted other challenges include how AI and ML would handle brown-field and legacy field components, as well as legacy equipment in the field that is either not sending any data or only limited data. “There will also be a need to provide the adaptation of standard AI methods to meet all citizens’ and city’s requirements, and interdisciplinary capabilities are needed and need to work closely with each other, which includes AI experts and traffic experts data sharing across different legal entities,” she noted. Mathieu added that key success factors have been to work closely together with city authorities and operators on very specific city-related issues in order to bring the specific value that the customer requires, pointing to projects in London and Hamburg, among other “confidential” projects.

Kazerounian also noted that despite the importance of AI and ML for smart cities, one question that remains is the extent to which the two technologies are necessary, when certain areas may benefit just as much from simple solutions. “Automated subway systems may not need much by way of AI but could probably reduce a lot of the need for smart prediction of vehicular traffic,” he said.

As with all technological changes, Kazerounian said municipalities, and society at large, should recognize that these are likely to be incremental, rather than seismic changes. “Cities will begin to increase technologies at their own pace, aside from already abandoned experiments like the Google smart-neighborhood in Toronto,” he said. “As such, we need to not only think about how these technologies work in their own right but, ultimately, how well they integrate with the technologies we have today.”