Irinel-Sorin Ilie

Support Vector Clustering of Electrical Load Pattern Data

This paper presents an original and effective application of support vector clustering (SVC) to electrical load pattern classification. The proposed SVC-based approach combines the calculation of the support vectors, carried out by using a classical procedure adopting a Gaussian kernel, with a specifically developed deterministic algorithm to form the clusters. This algorithm exploits the meaningful location of the bounded support vectors (BSVs) to define the outliers, identifying the clusters in function of the distance of the non-BSVs to the BSVs. Its implementation is less computationally intensive than other existing approaches and the cluster formation is driven by a single user-defined threshold. Extended comparison to other clustering methods is included to show the effectiveness of the proposed approach in grouping multidimensional load pattern data into non-overlapping clusters. This effectiveness is confirmed by the calculation of various cluster validity indicators. In particular, the most successful tasks are the identification of the outliers and the more effective formation of small numbers of clusters with respect to other methods.

Reliability performance of smart grids with demand-side management and distributed generation/storage technologies

The future “smart grids” will feature significantly higher numbers of various distributed generation and storage (DG&S) systems, as well as a wide-scale implementation of different demand-side management (DSM) functionalities. In existing literature, related analysis mostly concentrates on the operation, control and planning aspects of “smart grids”, while their impact on, and potential for improving system reliability performance is less researched. The work presented in this paper offers a new methodology for assessing reliability performance of power systems with fully employed DSM and DG&S functionalities, allowing to quantify in the most realistic manner standard set of indices reported annually to the Regulators. The methodology extends conventional reliability assessment procedures to include equivalent network models, actual load profiles and empirical fault probability distributions in the analysis, allowing to make a clear distinction between the short and long supply interruptions. Different DSM and DG&S scenarios are considered, in order to validate the proposed methodology and demonstrate its accuracy and applicability.

Reliability performance assessment in smart grids with demand-side management

The paper discusses possible impact of demand-side management (DSM) functionalities on the improvement of reliability performance and formulation of novel reliability assessment procedures of future electricity networks (so called “smart grids”). Firstly, traditionally used continuity of supply metrics and indices are assessed for a given reliability test system without considering any DSM scheme. Differences in the results for reliability indices for test system modelled with bulk loads and system modelled with detailed network configurations supplying connected loads are quantified, emphasising the errors that occur when parts of the system are neglected during the estimations of reliability performance. Afterwards, the effect of DSM on the reliability performance of the same test network are analysed, in order to assess potential benefits of DSM to the network operators, particularly with respect to their annual performance reports. Finally, possible changes in used reliability metrics are discussed, as smart grids will allow to substitute standard lumped representation of system loads at higher voltage levels with a more accurate and detailed information on load demands and load structures, including estimated contribution of demand-manageable portion of system load to the total demand.

Realising the potential of smart grids in LV networks. Part 2: Microgeneration

This paper, which is the second part of a two-part series, considers the influence of microgeneration technologies on the overall network performance and quality of supply of low-voltage residential customers in future “smart grids”. The paper uses the network models and demand-side management (DSM) scenarios developed in the Part 1 paper to further assess changes in active/reactive power flows, system losses, voltage profiles and harmonic emissions due to the combined effects of implementing microgeneration, energy storage and DSM.

Operational characteristics of a 27-MW wind farm from experimental data

The diffusion of wind systems is emerging in several countries, with solutions of different type and size. The analysis of experimental results gathered on-site is essential to understand the characteristics of the wind systems and of their interconnection to the grid. The basic aspects of the experimental analyses concern availability of the wind resource, efficiency of the plants, and power quality of the grid connected wind system. This paper illustrates and discusses the results obtained from monitoring a 27-MW wind farm with SCADA systems and network analysers. A dedicated procedure for efficiency evaluation of the wind turbines is presented, whose results underline evident mismatches between the actual performance and the data provided by the manufacturer.

Realising the potential of smart grids in LV networks. Part 1: Demand-side management

This paper, which is part one of a two-part series, analyses the influence of demand-side management (DSM) on the overall performance of distribution networks, particularly with respect to the quality of supply of low-voltage residential customers. The paper uses detailed network configurations and load models to assess changes in active/reactive power flows, system losses, voltage profiles and propagation of harmonics due to changes in the load mix as a result of DSM actions. The Part 2 paper will consider the effect of microgeneration technologies on further improvement/ deterioration of network performance.

Probabilistic clustering of wind generators

Increasing number of wind farms (WFs) connected to power systems calls for efficient aggregate models so that large farms can be represented by only few equivalent wind turbines for steady state and dynamic system studies. A wind power plant consists of many small generators inside a wind farm. Future wind projects predict even greater number of wind turbines inside a wind farm for increased capacity. If each generator is represented individually this adds considerably towards calculation time for dynamic simulations. For this reason, wind farms are required to be modelled by few equivalent wind turbines which will reduce computation time. In this paper wind turbines (WTs) are clustered based on wind speed they receive using the Support Vector Clustering (SVC) technique. It was found that if set of clusters that reoccur several times during the year, as the best representation of the entire WF, can be obtained probabilistically, the highly frequent set then can be used to represent the wind farm for the entire year. This method can prevent time consuming way of choosing new set of clusters every time the wind speed (WS) and wind direction (WD) varies. The most probable equivalent set is also better than a single turbine equivalent model as the latter is good only at very high wind speeds (if same farm layout and site are considered) and would inaccurately represent the wind farm at other speeds.

Protections impact on the availability of a wind power plant operating in real conditions

This paper addresses a number of aspects referred to availability of the wind turbines operating in a wind power plant. Starting from the experimental data gathered on-site through SCADA systems, and from a recorded data indicating the operation of various types of protections, the causes of unavailability are investigated in terms of determining the impact of the different types of failures as represented by the tripping of the corresponding protections. The results obtained show that the variability of occurrence and duration of the protection tripping for the different wind turbines is relatively high. This impacts on the identification of suitable values and shapes of the failure parameters that can be used to build a reliability model of the wind power plant with probabilistic entries. For this purpose, some indications for constructing a probabilistic model of the protection operation are provided, also taking into account the possible simultaneous tripping of the protections due to mutually dependent events.

Probabilistic identification of turbines facing high and low wind speeds in a wind farm

Due to growing number of wind turbines (WT) within a wind farm (WF) scheduling of maintenance of turbines within the WF becomes challenging issue. Also, in instances when wind energy curtailments are required it can be difficult to decide which turbines should be shut down first. In this paper, a probabilistic methodology is presented to identify WT in the WF that face higher and lower wind speeds during the year. Since this calculation is dependant on WF layout, location of WF and position of WTs inside the WF probabilistic site analysis is performed, along with turbine clustering, after determining wind speed approaching each turbine by using a detailed wake effect model. The approach presented can be applied to a wind farm of any size and layout at any location.