Applied analytics for Cyber Physical Systems

From the National Institute of Science and Technology definition of a Cyber Physical System:

“Cyber-Physical Systems or “smart” systems are co-engineered interacting networks of physical and computational components. These systems will provide the foundation of our critical infrastructure, form the bases of emerging and future smart services, and improve our quality of life in many areas.”

How do we know this?

The broad answer to this question lies in the field of analytics.

Analytics makes use of data, information technology, statistical analysis, quantitative methods, mathematical and computer-based models to assist the discovery of insight (typically patterns) so that we can make decisions based on fact. Business Intelligence, which has been a popular application for organisations to purchase, is an example of a function that requires data analytics for it to operate.

Other examples of domains suitable for analytics include:

• Customer Relationship Management (CRM);
• Financial and marketing activities;
• Supply chain management;
• Human resource planning and monitoring;
• Pricing decisions;
• Sport science strategies – “marginal gains”

Business analytics

There is an explicit link between business analytics and the profitability and revenue generation of a business, as well as the strength of return to shareholders. The activities of analytics enhance the comprehension of data, enabling a business to remain competitive. At its most basic, analytics facilitates the creation of informative reports to assist decision making.

There are four categories of analytics:

1. Descriptive analytics – using historical data to understand the past; “what has happened?”
2. Diagnostic analytics – using data to find the root cause of an event; “what made it happen?”
3. Predictive analytics – evaluating historical performance to produce models that can predict behaviour in the future; “what will happen?”
4. Prescriptive analytics – using optimisation techniques to direct actions and simplify decision making; “how do we make it happen?”

For example, a retail organisation clears seasonal stock with a sale. The question is:

“when do we reduce the price and by how much to maximise profits?”

Descriptive analytics examines the historical data for similar products and reports prices, units sold, advertising campaigns, etc.

Diagnostic analytics identifies key combinations of events and activities that produced recognisable behaviour.

Predictive analytics creates a model to present a number of possible future scenarios.

Prescriptive analytics finds the best sets of pricing and advertising to maximise sales revenue.

Descriptive analytics simply tell the consumer what is happening and make visible relationships, such as where a breakeven point might lie in relation to the current results. Such systems don’t instruct a manager as to what to do, and are the core of traditional Business Intelligence functions for some time. Typical examples are reporting and Online Analytical Processing (OLAP) dashboards, data visualisation via reporting, etc.

Decision modelling

Central to the prediction of future performance is the construction of decision models. A model is often a mathematical abstraction or representation of a real system, idea or object, that captures the most important features of reality. It can be described in writing or verbally, but for the purposes of automation is best described mathematically.

 This model is used to understand, analyse and facilitate decision making from data that will contain controllable variables (decision variables) and uncontrollable variables.

Predictive decision models often incorporate uncertainty to help inform managers so that they can analyse risk. The aim is to predict future behaviour based on different scenarios, and is thus shaped by the imperfect knowledge of what will happen, otherwise referred to as uncertainty. Risk is associate with the consequences of what actually happens.

The models use algorithms for regression analysis, machine learning and neural networks, all of which are mature and have been tested in many different domains. One business example of predictive modelling is that of marketing. Used in conjunction with descriptive analytics visualisations, marketeers can interpret the outputs from predictive and prescriptive analytics so that they make the most informed decisions possible.

Since we want to identify the “best” solution, we need to find the values of decision variables that minimise (or maximise) cost/profit, etc. This is know an optimisation and is represented by an objective function. The constraints of the model represent the limits of the domain to be modelled, and the optimal solution is the values of the decision variables at the minimum (or maximum) point.

Prescriptive analytics

Commonly referred to as advanced analytics, the predictive models are used to present an optimum set of activities. For instance, an organisation that uses scarce resources, or whose business involves perishable goods, will be interested in identifying the activities that can best manage its operations.

Analytics-led activities enable us to tackle complex problems by providing individualised solutions. Each of the products and services modelled can be organised around the needs of individual stakeholders, and is suited to scenarios where the value of interactions with stakeholders is high.

These approaches become more important as the volume of interaction between stakeholder agents increases; one pertinent example is the rise in IoT devices that are all individual actors in a complex CPS.