ENERGY INTELLIGENCE WITH EMIR SYSTEM (Patented Service and Method)

ABSTRACT
Internet use, domestic energy consumption for electricity, renewable energy production, consumers' education and their interaction are focused in this paper. The increasing number and speeds of internet connections can be seen as the tool for measuring electricity and consumers' behaviour on energy consumption, giving an added value to consumer access and interactive approach to energy and energy related solutions. An increasing number of consumers from all levels of the society, cultures, lifestyles and social status have continuous internet access through ADSL connections. Those consumers will be targeted in order to collect and analyse their acceptability in new energy technologies.

The aim of this system and method is to develop an internet-based methodology that brings together energy consultants, domestic consumers and renewable technologies. Consumers' interaction with energy consultants will be a valuable step into implementing efficiently EU policies, focusing on the introduction of informative billing through end-use energy efficiency directive and the use of renewable energy sources. Furthermore, the involvement of utility companies [6] will be valuable in order to manage efficiently electricity production and to plan demand side management and energy efficiency programs. Energy-related market will be enforced and benefited through targeted information, behavioral changes and innovative accessibility internet tools.

MATHEMATICAL MODEL
A typical electricity demand model [10], [11] that we wanted to capture in our model includes four very important EPIs (Energy Performance Indicators) metrics: elasticity, seasonality, mean reversion, and stochastic growth.

" Load Elasticity: We assume electricity demand to be completely inelastic (i.e. independent of market clearing price). This may appear to be a strong assumption, but in the current state of deregulation, few end users actually observe real time price movements.

" Load Seasonality: Seasonality is a major driver for electricity demand. We observe seasonality over the daily, weekly, and yearly cycles.

" Load Mean reversion: One can observe temporary spikes in electricity demand, often induced by extreme weather conditions. However, these spikes are not sustainable and demand reverts back to normal levels within a few days.

" Stochastic growth: Growth in electricity demand is driven in part by trends in the overall economy. The growth is therefore hard to predict over longer time horizons, and must be considered stochastic.

TECHNICAL ARCHITECTURE
In order to show and describe the functionality of the Matlab Server Pages (MSP), a web-based Energy Decision Support System was used in order to integrate Matlab functions and tags to the energy analysis procedure. A modern Decision Support System (DSS) can be defined as computer-based tool, or a more complex Information System structure, used to support and generate decision-making and problem solving.

In a typical EIS architecture [3], [6], the EIS server hardware and software located at the EIS service provider's physical site record interval data via the Internet. The EIS receives these data from signals dispatched by meters installed in a customer's building, or directly communicates with meters. The EIS users can access the server with a password, and access the archived energy data either in real-time or in hourly, or daily updates from anywhere via a web browser.
This web-based functionality can be enriched with many add-on services in order to create a complete Information System that will act as the basic "ad-hoc broker" between free customers and energy providers, in the future opened Energy market. Two different environments were involved in the development of the web-based evalua-tion tool interface: hyper-text markup language (HTML) standards with the addition of some java server pages (JSP) under J2EE specifications, and advanced Matlab code [5], which permits the dynamic mathematical development of the evaluation model described above. The Matlab web server toolbox [5], [7] was effectively used by combining java language, to produce dynamic Matlab server pages [7].

The communication was pre-served, through internal java objects, which are incorporated in the apache tomcat server. Remote Matlab method invocation was achieved by this way and by combining CGI-based POST methods to the Matlab web server; a very powerful Matlab application server was produced. The developed system allowed the creation of compact programs, called Matlab beans and the execution of those interconnected beans. The application, using web GUI facilities sends data, through java calls and post methods to the Matlab beans and vice versa. The system support parallel processing algorithms and form the first simulations that took place in Medialab (NTUA), matlab parallel toolbox was used, in order to parallelize norm distance calculation. By using Valiant algorithm, internal package blocks where minimized statistically.
The overall performance output was more than positive, allowing the algorithm to be executed concurrently to any energy customer, using special screen-saver programs.


HYPERCUBIC GRID AND CLUSTERING COMPLEXES (Patented)
The method for analysing energy data and moving towards a decision (patented method and system by two patents so far) uses a different version of a clustering algorithm and a hypercubic grid [2] in order to cluster energy pages and energy measurements in a distributed way, not only by an energy count analysis but using a relevance distance calculation method from an optimal value, which is called hypercubic centroid. The probable energy pages or measurements that will be used to measure normed distances form a surrounding grid in a multidimensional binary space. For each possible energy measurement that is going to be assessed, we choose and construct an attribute vector describing some attributes that we have to take into account in order to decide if the page is relevant with a speci?c query. After the formation of the attribute vector, we assign weights (wi) to the attributes to distinguish the most important and we choose possible optimal values (optimal energy pages) that are described by various optimal attribute vectors. These optimal values are the centroids. After the formation of the centroids, we start to measure metric distances (calculation of p-norm) from each centroid and all the possible pages that we have gathered (energy crawlers) through internet (identical to indexer). This can be achieved by a hypercubic parallel network with moving agents, where the centroids are the vertices of a multidimensional hypercube. With this method, we create relevance tables for each centroid and the total system is called Hypercubic Knowledge Grid (HNG). The algorithm generates the knowledge grid and by using search algorithms we can access the relevance tables and rank output data according to a specifc query.

The above algorithm will be also used for Data bases analysis and SQL query optimization. The above mentioned method, helps us a lot in order to assign different optimal values to the various vertices of the hypercube and take these values as optimal centroid solutions for a multidimensional data mining application that has as a result, ranked sets of information retrieval queries. In order to perform a distributed data mining and clustering algorithm inside a hypercubic network, we need to ensure that parallel communications between vertices follow some structured rules and that ranked output data will be available.

A hypercube with dimension N, (GH (N)) is a network grid having 2N binary vertices which are mapped on a binary space. The distributed agent hypercubic scheme that will be used in order compute the normed energy distance between an optimal centroid solution and a multidimensional energy vertex that represents a possible query to the energy search engine, will follow the above structure and a method of the probabilistic routing algorithm of Les Valiant [2]. We will de?ne a probabilistic random vertex permutation Pi, using a uniform distribution. In each vertex there exists an agent that will measure the n-th dimensional topological norm between the vertex and the optimal centroid. The above mathematical procedure is being used in order to calculate all the topological distances between centroids and relevant vertices, by using hypercubic routing and distributed agents. The results of the calculations form a uniform multidimensional table which is called Energy Relevance Table. This table represents the degree of relevance of each vertex from the surrounding grid comparing with a unique optimal centroid (hypercubic node), which can express an optimal solution, an optimal value or a suggested value to a problem or an Energy query [8]. In each cycle the algorithm will compute the normed distance between each local vertex (optimal centroid) that the agent resides in each cycle and the surrounding grid of possible solutions from the query. A multidimensional relativity matrix will be created with all the normed distances between the centroid vertex and the surrounding grid. After the determination of the local matrix the agent moves to the next vertex, defined by the routing algorithm.
The computational complexity of the above hypercubic-based algorithm was computed by using the Chernov born. We will have to add the computational load for calculating the p-th dimensional norm for each vertex, which of course depends on the dimension of the surrounding grid (ie. Dimension N).
As it was mentioned above, relativity measurement between an optimal centroid and a probable value from the surrounding grid, can be achieved by measuring the metric distance between these two vertices.

CASE STUDY of EMIR
The system (http://fermat.medialab.ntua.gr/emir) was tested with real energy data, from DESMIE (http://www.desmie.gr) using the previous year's energy data repository for the National Energy Consumption databse 2004-2005, using the National Load (MWh) and the extracted SMP (Euros/MWh). The system has a web-based Graphical User Interface, that lets the consumer or the producer to control the directed queries on the energy database. There is a functional menu with all available functionalities. The user simply selects the option and by pressing a key, the system generates useful graphs (Figs) and various customised measurement tables with statistical data.
Also, special measurements from Greek Home consumers where used, in order to analyse and correlate energy behaviours along time. Energy data where measured successfully from various electrical devices inside the house, using PLC modems. This gives advantage to the system, since many statistical and stochastic correlations between home devices and load can be made, in order to extract correlated energy behaviours.

CONCLUSIONS
So it can be said that traditionally most utility companies classified their customers according to a few electrical parameters and some commercial codes. In the liberalised electricity market, there is a strong need for classifying the electricity customers on the basis of indicators able to characterise their true electrical behaviour. A possible scenario of the interactions among customer and supplier could be the following:

" the customer comes to the supplier, states its type of activity and is assigned/free to choose a starting tariff; the supplier monitors the customer for a specified period (e.g., 3-6 months) and establishes a reference pattern for its load diagram;

" the supplier fits the reference pattern to one of the customer classes already defined and identifies the appropriate tariff;

Hence, the supplier performs a continuous monitoring of the daily load curves of all customers, periodically updates the reference patterns and the composition of the customer classes by automatic clustering and adjusts the tariffs applied to each customer class such as to maximise its foreseeable profits in the respect of possible price caps.
Efforts to put some order in the tools to analyse the load diagrams have been produced for quite some years. We can mention the systematic approach used in UK [12], according to which several subclasses are defined within each major class of customers, for each of them a different tariff being assigned. This approach is backed by .some extensive field measurement campaigns that span over two decades. A rather similar approach has been implemented in Taipei, together with a comprehensive survey system.

The load diagram associated to each average customer is the load profile of the corresponding customer class. The economical aspects related to the possible tariff diversification for the various customer classes are investigated by using the load profiles for providing suggestions on possible market strategies seen from the point of view of the electricity utility. Traditionally, most utility companies classified their customers according to a few electrical parameters and some commercial codes. In the liberalised electricity market, there is a strong need for classifying the electricity customers on the basis of indicators able to characterise their true electrical behaviour.
So it can be said that traditionally most utility companies classified their customers according to a few electrical parameters and some commercial codes. In the liberalised electricity market, there is a strong need for classifying the electricity customers on the basis of indicators able to characterise their true electrical behaviour.

References

[1] E.M.I.R. Project (Energy Management & Intelligent Reporting) : http://fermat.medialab.ntua.gr/emir

[2] V. Nikolopoulos "Analyse et Simulation des methodes de routage dans la topologie d'hypercube" memoire, Ecole Polytechnique, promotion X99, 2002

[3] RTE, France, "Responsable d'equilibre - Regles et contractualisation"

[4] Lekatsas, "Financial Analysis of Energy Systems", Chap.4 Pupl. ÔÅÅ 2000

[5] The Mathworks Inc., 1999. MATLAB Web Server. Natick, MA

[6] MAVIR website http://www.mavir.hu "Code of Commerce"

[7] V.Nikolopoulos, V.Loumos, "A web-based Energy Decision Support System for dynamic knowledge energy management and automatic intelligent reporting (EMIR)" Paper under preparation, 2007

[8] Newman MEJ. "The structure and function of complex networks", Siam Rev 2003;45(2):167-256.

[9] Steyvers M, Tenenbaum JB, "The large-scale structure of semantic networks: Statistical analyses and a model of semantic growth" Cog. Sci 2005;29(1):41 78.

[10] M.Emoult and F.Meslier, "Analysis and forecast of electrical energy demand". RCE, No.4, 1982.

[11] Skantze, P., Chapman, J., Ilic, M.D., "Stochastic Modelling of Electric Power Prices in a Multi-Market Environ ment", Transactions of IEEE PES Winter Meeting, Singapore, January 2000.

[12] S.V.Allera and A.G.Horsburgh, "Load profiling for the energy trading and settlements in the UK electricity markets", Proc. DistribuTECH Europe DA/DSM Conference, London, 27-29 October 1998

[13] Ali Kizil, "MSP matlab server pages", http://msp.sourceforge.net/

[14] V. Nikolopoulos, "EMIR - Energy Management and Intelligent Reporting" Bulletin of Electrical and Mechanical Engineers Bulletin, No 382, 1 December 2005 "

[15] V. Nikolopoulos, V.Loumos, ""A Web-based system for optimal Energy Sources Management, through Ontologies and Semantic Clustering" Technical Chamber of Greece (TEE) Research Conference 2006, Athens, Greece

[16] V. Nikolopoulos, Vassili Loumos, "A Web-based Information System for Optimal Energy Sources Mamagement through Ontologies and Hypercubic Clustering ", Energy 2006 International Conference, Athens Greece

[17] V. Nikolopoulos, Vassili Loumos, "A Web-based Information System for Optimal Energy Management",T.E.E. Energy Minimization Research Day 2006, Academy of Sciences & NTUA

[18] V. Nikolopoulos, Vassili Loumos "Integrated Ontological Model of web-based Energy Management through Semantic Hypercubic Grid", V, 2nd Panhellenic Conference of PSDMH in Athens, May 2007

For more information contacts at: info@intelen.gr