System Analytics
Prof. Simon Watson and Dr.Henry Plank are exploring the systems associated with emerging digital technologies. The basic objectives of digital technology is intelligent decision making in complex and rapidly changing business environment, fast decision making in adaptive situation, improved accuracy in decision making, discovery of hidden intelligence from large pool of data, fast and correct transaction processing; support creation, storage, transfer and application of knowledge in an enterprise, support office automation and efficient management of resources (e.g. man, machine, materials, method and money) of an enterprise. An information system associated with digital technology can be classified into different categories such as transaction processing, decision support, group decision support, knowledge management, knowledge based office automation and business intelligence system. It is possible to analyze digital technology in terms of computing (e.g. centralized, distributed, local, global), data, networking (e.g. wired, wireless, Internet), application (e.g. features, modules, functions, application integration) and security schema.
dSaaS / DaaS : The basic objective of DaaS is to avoid the complexity and cost of running a database with improved availability, performance, price and flexibility. It gives the access to various types business intelligence solutions (through web) which include distributed database, data warehousing, data mining, business and web analytics, data visualization and business performance measurement applications. The pricing of dSaaS is based on the cost of hardware (e.g. data warehouse, servers), the cost of software (e.g. business intelligence solutions) and system administration cost (e.g. data centre administration, data base security, backup, recovery and maintenance). A consumer can lease a data storage space where it is required to measure different system parameters such as stored data (GB/month) and number of processed queries (per 10k requests / month) to compute the price of dSaaS / DaaS. The provider can offer quantity discount in case of group buying of storage space. The prices of DaaS / dSaaS are also determined by various QoS parameters such as connection speed, data store delete time, data store read time, deployment latency (i.e. the amount of latency between when an application is posted and ready to use) and lag time (how slow the system is). The pricing of dSaaS is also governed by the security and privacy of data. Some applications (e.g. education sector) require low level of privacy of data. Some applications (e.g. financial service, healthcare) need high level of security and privacy in data outsourcing and this involves high cost of computation and communication from the perspectives of statistical disclosure control, private data analysis, privacy preserving data mining, intelligent access control and query processing on encrypted data. The service provider should define a discriminatory pricing mechanism for dSaaS: high level of security and privacy of data demands high price and low level of security asks low price.
The price of dSaaS is a function of miscellaneous cost elements of a data center. A data centre or data bank is the collection of servers where the applications and data are stored. Data center consists of a set of servers and network architecture. The servers store the data from different organizations and network architecture facilitates the services to use, store, and update the data of the servers. The cost of administration of data centre includes several factors: initial development cost, operating cost, maintenance cost and cost associated with disaster recovery plan. The development cost includes the cost that requires making master plan, building infrastructure, buying hardware and software, making database and security schema. Operating cost includes the cost of energy, cooling system, system administrators, software license and network cost. Maintenance cost is the cost of maintaining the system which includes upgradation of hardware and software. One of the most challenging issues of data center management is the resource allocation strategy: how it is possible to cater the demand of the service consumers using minimum number of servers. It has an impact on the size, complexity and cost of data center. The data centre administrator can follow dedicated or shared server allocation strategy.
The price of dSaaS is also a function of energy consumption of cloud computing system in a data center. There are many open challenges of energy efficient design of computing systems and green IT covering the hardware, operating system, virtualization and data center levels. The basic objective of the cloud computing system design has been shifted to power and energy efficiency to improve the profit of the service provider. Energy consumption is not only determined by hardware efficiency, but it is also dependent on the resource management system deployed on the infrastructure and the efficiency of applications running in the system. Solar power electronics is an interesting option of green IT. Higher power consumption results not only high energy cost but also increases the cost of cooling system and power delivery infrastructure including UPS and power distribution panels. The consolidation of IT infrastructure should be done intelligently to reduce both energy consumption and performance degradation through improved power management. Energy consumption can be reduced by increasing the resource utilization and use of energy efficient cloud computing system.
Software-as-a-Service (SaaS): SaaS is an application hosted on a remote server and accessed through web; it can be business service or customer oriented service. The basic objective is to reduce software licensing cost and improve productivity by using sophisticated applications. The pricing strategy of SaaS is based on pay-as- you-go basis; not dependent on number of licensing period and licensing users as in case of direct software procurement. Another concept is software plus service where an enterprise uses a locally hosted software application and additionally uses SaaS through cloud for a specific type of application. Using the existing software paradigm, the consumer purchases a software package and license by paying a one- time fee. The software then becomes the property of the consumer. Support and updates are provided by the vendor under the terms of the license agreement. This can be costly if the user is installing a new application on hundreds or thousands of computers. SaaS, on the other hand, has no licensing. Rather than buying the application, the consumer pay for it through the use of a subscription based on number of concurrent users and only pay for what is used.
The computation of subscription fee can be stochastic pricing or simple cost based pricing. The price of SaaS depends on specific business model of the service provider. Suppose, a service provider develops in-house software products. Another service provider buys COTS from third-party vendor based on number of licensed users and licensing period and provides SaaS to the consumers. There may be restriction of number of concurrent users and different subscription rate of SaaS in second case.
This pricing strategy should also consider cost of upgrading software application; the provider may offer incentive for upgrading applications. In case of security software pricing, there may be different alternative strategies to manage network security: (i) consumer self-patching where no external incentives are provided for patching or purchasing, (ii) mandatory patching, (ii) patching rebate and (iv) usage tax. For proprietary software, when the software security risk and the patching costs are high, a patching rebate dominates the other strategies. When the patching cost or the security risk is low, self-patching is the best option.
Stochastic risk based pricing mechanism considers several risk factors and optimizes the expected net present value of revenue subject to maximum acceptable risk of the provider. In this case, the service provider does not give much focus on cost accounting model or profit margin but tests the price sensitivity of the customers experimentally or through trial and error method. The provider does not have any precise perception about the demand of the new software products. But, it follows dynamic risk based pricing based on assessed risks and competitive intelligence. For in-house software development, software cost is a function of efforts on feasibility study, requirement analysis, system design, program design, coding, testing and modification following waterfall / v-process / spiral / proto- typing / incremental delivery model. The service provider estimates effort for a specific SDLC model and then selects an optimal profit margin.
Infrastructure-as-a-Service (IaaS): A cloud computing infrastructure consists of different types of elements: clients (e.g. mobile, PDA, laptop, thin and thick), the data center and distributed servers. Thin clients are less costly than thick clients. A growing trend in the cloud computing is virtualization of servers. In a virtualized environment, applications run on a server and are displayed on the client. The server can be local or on the other side of the cloud. Software can be installed allowing multiple instances of virtual servers which run on a physical server. Full virtualization is a technique in which a complete installation of one machine is run on another. It allows the running of different and unique operating systems. Hardware-as-a-Service (HaaS) simply offers the hardware required by a consumer. Cloud computing is a business model of delivering IT resources and applications as services accessible remotely over the Internet rather than locally. IaaS supports remote access of computer infrastructure as a service.
Cloud computing supports elastically scaling computation to match time varying demand. But, the uncertainty of variable loads necessitate the use of margins i.e. the servers that must be kept active to absorb unpredictable potential load surges which can be a significant fraction of overall cost. There are challenges of minimizing margin costs and true costs for IaaS. The provider should not adopt a fixed margin strategy; the margin should be load dependent. The margin required at low loads may be higher than the margin required at high loads. Secondly, the tolerance i.e. the fraction of time when the response time target may be violated need not be uniform across all load levels. It is really challenging to achieve optimal margin cost while guarantying desired response time for IaaS.
The pricing strategy of IaaS is based on the cost of servers, storage space, network equipment and system software like operating systems and database systems. The price of IaaS is basically a subscription fee for a specific timeline. Now the question is how to compute this subscription fee. The rate should be fixed based on the cost of hardware and software, target revenue and profit margin. The service provider may adopt a profit maximizing pricing strategy or revenue maximizing pricing strategy within reasonable, stable target profit margin. The profit margin is a dynamic variable; it should be set intelligently according to competitive intelligence and quality of service. The quality of service is measured in terms of computing time. For small firm or individual service consumer, the provider can set a fixed price per unit time; there may be SLA but there is no scope of negotiation of price. Large PSU can negotiate with the service provider to set a rational price for fixed timeline.
Incentive compatibility plays a significant role in IaaS pricing, it is important to analyze the significance of incentives for network infrastructure investment under different pricing strategies: congestion based negative externality pricing and the flat rate pricing. A lack of proper infrastructure investment incentive may lead to an environment where network growth may not keep pace with the service requirements. It is really complex to compute maximum capacity that IaaS provider will be willing to invest under different pricing schemes. Optimal capacity of IaaS is determined by different factors: per unit cost of capacity of network resources, average value of the user’s requests, average value of the user’s tolerance for delay and the level of exogeneous demand for the services on the network. It is hard to determine whether time based pricing is more profitable than flat rate pricing. IaaS consumers always try to identify whether average stream of the net benefits realized under congestion based pricing is higher than the average net benefits under flat rate pricing. IaaS provider may adopt different types of pricing strategies at different points of time but the service consumers may control their demand of IaaS service adaptively to avoid the increase in cost.
Platform-as-a-Service (PaaS) : PaaS supplies all the resources required to build applications and services completely from the web without any download or installation of any software in the clients. The price of PaaS can be negotiated for a specific project. There can be different types of project environments such as application-delivery-only-environment (e.g. security and on demand scalability), standalone environment and add-on-developmental-environment (e.g. subscriptions of add-on SaaS application are bought). The price of system software can be charged as a subscription fee based on number of concurrent users and usage period. The pricing of PaaS is also governed by the complexity of platform services which may include application design, development, testing, deployment, hosting, geographically dispersed team collaboration, web service integration, database integration, security, scalability, storage, state management and versioning. The developers, project managers, and testers can access the development and testing softwares of the service provider through web; but lack of interoperability and portability may be a critical issue in PaaS. The price of PaaS is determined by the complexity of interoperability between the systems of the service provider and service consumer.
Virtual and Augmented Reality : There are three types of reality technologies: virtual reality (VR), mixed reality (MR) and augmented reality (AR). These reality technologies are sophisticated, creative and powerful tools to offer a complete computerized digital experience through artificial intelligence, computer vision, computer graphics and automation. A virtual entity may not exist physically but created by software in a digital environment. Augmented reality is an enhanced version of the real-world by overlaying our existing reality with an additional layer of digital information, which can be viewed through smartphones or smart glasses (ARSGs). Mixed reality facilitates the merger of and real-time interaction between, digitally rendered and real-world data and objects through MR headset. Virtual reality is characterized by generating real-time, immersive and interactive multi- sensory experiences situated in and artificially induced by a responsive three- dimensional computer-generated virtual environment - usually paired with advanced input and output devices.
The expert panel are analyzing a smart grid in terms of self-healing network. The basic building blocks of a self healing network are computationally efficient state estimation algorithms that can predict voltage and phase at different nodes of a smart grid in real-time given the current and predicted energy demand and supply of the prosumers. Distributed coordination is important for automated voltage regulators, voltage control and balancing demand and supply during recovery of faults. It is really challenging to develop automated and distributed active network management strategies given the uncertainty of demand and supply at different levels in the smart grid, fault correction mechanisms, self healing strategies, cause- effect analysis on various types of faults (e.g. overload, over current, earth fault, short circuit, over voltage, under voltage, over frequency, under frequency, automatic voltage regulation). An active network configures the topology automatically, sends control signals to individual customers to adjust generation and also load control, automatically correct faults and self-heals the smart grid.
A self-healing mechanism should maintain the stability of a distribution network, perform accurate and timely monitoring and control of the prosumers; big data analysis for multiple actors and sensors, micro-level measurement and predict the future state of smart grid. It can adopt a set of active network management techniques based on distributed intelligence in the self healing network for fast recovery from faults. In case of voltage drift, automatic action is necessary on the transformer to reestablish correct voltage levels. It is essential to balance the mismatch between supply and demand to avoid blackout situation. Essential need of a self healing mechanism is that various components of a smart grid should be able to communicate for voltage regulation and control of generation capacity and load demand.
Artificial intelligence can be applied to a smart grid in various ways such as knowledge based expert system for knowledge acquisition, inference engine, knowledge base, applications, fault diagnosis; real time alarm handling and fault analysis (AHFA); voltage control such as voltage collapse monitor (VCM), reactive power management (RPM), combined active and reactive dispatch (CARD), power system protection for protective relay setting, phase selection, static security assessment, condition monitoring, scheduling and maintenance of electrical transmission networks and intelligent system for demand forecasting. It is interesting to apply various types of soft computing tools for smart grid system performance analysis. Adaptive fuzzy system is used for fuzzy reasoning, defuzzification and function approximation on imprecise and uncertain data. Artificial neural network can be useful for intelligent data mining, neuro-fuzzy control, neuro expert power system and evolutionary computing (e.g. neuro GA).
Let us do technical analysis on evolution of AI in a smart grid system. A knowledge based expert system captures the knowledge of human expert in a specific domain. It uses the knowledge for decision making and appropriate reasoning of complex problems. The expert system needs a knowledge structure in the form of production rules, frames and rules. A knowledge base is a form of database containing both facts and rules. Let us present here examples of few rules.
o THEN open I.
B = Z /* NOT (A). B + A. NOT (B) = Z*/An expert system (ES) can function based on knowledge of power system operation which may or may not be complete. An ES performs several functions such as knowledge acquisition and inference engine [16]. Data mining algorithms analyze SCADA data. This component acquires new facts or rules. Inference engine performs several functions such as verification, validation, cause-effect analysis, sequence control for rule firing, data processing, meta knowledge management, forward and backward chaining. But, ES may have some limitations from the perspectives of inappropriate representation of knowledge structure. Expert systems can be used in power system analysis
Planning : AC/DC network design, power plant management, capacity planning;
Operation: alarm processing, fault diagnosis, forecasting. Maintenance scheduling, demand side management, reactive voltage control;
Analysis : Control system design, power system protection and coordination;
AC load flow analysis : Input data includes network parameters, connections, loads, maximum active and reactive power output. It minimizes a set of
objective functions subject to a set of constraints such as network laws, plant loading limits, busbar voltage limits, line loading constraints and security.
Case : Self-healing smart solar grid
North American power grid, one of the greatest engineering innovations of 20th century is now 50 years old and needs a smart self healing grid to incorporate renewable energy sources, reduced number of power outages and reduced carbon emissions [44]. North America has already experienced a number of power outages. In 2012, the occurrence of hurricane Sandy caused power outages in twenty four state of USA and shut down of schools and offices. In 2003 a blackout occurred for two days throughout North-eastern and Midwestern parts of USA and United States and the Canadian province of Ontario. The cause of the blackout was a software bug in the alarm system at a control room of the First Energy Corporation in Ohio. In 2011, New England experienced a Halloween snowstorm that put millions of people in the dark for several days. The following section presents a self healing mechanism [SHM] for the smart power grid.
Self Healing Mechanism [SHM]
Agents : Service consumers (B)[e.g. smart home, smart building, industrial load, solar pump for agriculture, microgrid], Service provider (S);
Structure
Smart power grid comprising of power generation, transmission and distribution system, generators, transformers, transmission lines, loads, switchyards, microgrids comprising of AC / DC sources and loads, renewable energy sources (e.g. PV panels) and energy storage system;
fully automated power delivery network that monitors and controls a set of nodes, supports a bidirectional flow of electrical power and information among the power plants, loads and all intermittent points;
Scope:
ensure stability, reliability, consistency and improved efficiency during normal operating conditions;
self-recovery during human error or natural disaster;
enable better integration between conventional grid and renewable energy sources;
mitigate the impact of power outages;
ensure fewer blackouts for shorter periods;
Constraints : time, cost, technology;
Strategy: Select a set of intelligent strategic moves rationally.
System : AI enabled expert system
define and configure expert system in the form of knowledge base, knowledge acquisition system, inference engine, workplace or memory, justifier, user interface, knowledge refining system and consulting environment;
develop self-stabilizing and self-organizing distributed network management algorithms;
call computationally efficient state estimation algorithms that can predict voltage and phase at different nodes of a smart grid in real-time given the current and predicted energy demand and supply of the prosumers;
distributed coordination for automated voltage regulators, voltage control and balancing demand and supply during recovery of faults;
automated and distributed active network management strategies given the uncertainty of demand and supply at different levels in the smart grid, fault correction mechanisms, self healing strategies, cause-effect analysis on various types of faults;
Configuration of the network automatically, sends control signals to individual customers to adjust generation and also load control, automatically correct faults and self-heals the smart grid.
Security
verify security intelligence through automated or semi-automated system verification.
Verify security intelligence of a smart grid at various levels such as L1, L2, L3, L4 and L5.