Decision Support System

The so called decision support system unites the error detection routine that decides about the occurrence of a malfunction in a PV system, the footprint algorithm that detects causes for a system failure, data bases that store long-term surveillance data of a PV system, and the notification system that informs the operator. It is the central system that manages all information. Its central element is the footprint algorithm developed at the Fraunhofer Institute of Solar Energy Systems that will be described here in more detail.

Development of the footprint algorithm

For the development of the footprint algorithm monitored data of several PV systems have been analysed for the occurrence of system malfunction periods. The aim of this analysis is to identify typical error patterns.

From the database of the German 1000-roofs programme hourly mean values have been extracted. The analysis concentrated on two monitored signals, the irradiance and the produced AC power.

Displaying scatter diagrams (Figure 2) and using a normalised presentation of the

produced AC power allowed in a first try the detection of the following errors:

• String error (four of the eight strings of the PV system were disconnected for a few days). The number of strings disconnected could be derived successfully.

• MPP tracking error. Within a three-days period, an excursion of the MPP-tracking system of the inverter could be detected in one of the days. The analysis of the power production pattern on hourly base is a pre-condition for detecting this error source.

• Snow coverage. Difficult to detect within one day. A snow cover may disappear within days, causing the AC power production continuously approaching from nearly zero to the expected values over days.

Figure 2: Example: String error in a system from the D-1000 data base. The scatter diagram above indicates the occurrence of the failure; the figure below shows the decrease in power production (normalised) on the time axis. With the correct assumption of 50% of disconnected strings, the ‘virtual’ correction of the data (blue boxes) is more close to the average production profile than with the assumption of one string more or less disconnected (n+1, n-1).

This analysis has shown so far that for the allocation of different error sources hourly values of irradiation and AC power are necessary. Daily mean values are not sufficient. Furthermore, the uncertainties of satellite-derived irradiance values demand more effort in averaging processes in the footprint method to reduce the expected errors.

The footprint algorithm

The footprint method is divided into two steps. The first step contains a pre-sorting algorithm that prepares the calculated and the monitored yields to take the errors from the satellite data into account. The second step is the identification of the error source.

In general, normalised signals will be considered: P_sim / P_mon = simulated power / monitored power; P_mon / P_inst = monitored power / installed power.

Since the individual calculated yield values with hourly time resolution are expected to be provided with large errors, the approach in the pre-sorting of data is as follows:

The signals P_sim / P_mon will be sorted in intervals with an interval average value P*. The interval average P* shows in general a smaller variance than the variances of the individual signals. Thus, P* exhibits more stability and allows an improved detection of errors.

The intervals are defined in two nearly independent domains: The signals are sorted into a capacity domain and into a time domain. In the capacity domain, the intervals are fractions of P_mon / P_inst, and the time domain consists of hourly intervals. For both domains, the interval averages are determined. The different spatial distribution of the interval averages in both domains is a pre-condition to detect the error source in the subsequent footprint algorithm. Figure 3 illustrates the sorting into intervals in the capacity domain.

Three averaging periods are considered: One day (the past day), the last seven days and the last 30 days. Thus, for each interval in both domains, there will be three interval averages calculated according to the considered periods. An increase in accuracy of the signal pattern is expected with this approach.

Figure 3: Signal sorting in the capacity domain.

First tests of this approach were made using monitored data from:

a) A grid-connected PV system installed at Oldenburg University. For this system, monitored and simulated yield values including the standard deviations due to the irradiation uncertainties have been submitted.

b) A grid-connected PV system installed at a secondary school building in Freiburg. For this system, the simulated yields were determined with a simple model and standard deviations of the signals were estimated.

For both systems, trouble-free operation periods were used for the first test of the approach. Figure 4 shows the interval averages in both domains for the 30 days period from data of the system at Oldenburg University.

Figure 4: Interval averages from the 30-days period of the test system at Oldenburg University. For this period, a slight systematic lower production than expected in the upper power range, mainly in the afternoon hours, can be detected.

Figure 5: Error pattern for the 30-days test period of the Oldenburg University PV system (extracted from interval averages as shown in figure 3).

With the described preparation of the signals it is possible to reduce the system behaviour to more simple error pattern, as shown in Figure 5. These error pattern may then be compared with pre-defined error pattern for specific system malfunctions.

An example for a pre-defined error pattern for shading is given in Figure 6. Probability weights are distributed according to the expected appearance of the error. The probability for this error increases as the real system behaviour follows this specific error pattern.. The method may use additional input values (e. g. clear sky index, sun position).

A promising approach was developed to detect system errors on the one hand and to distinguish between system errors on base of pre-defined footprint tables on the other hand. An advantage of the method is that decisions will be made preferably on base of interval averages instead of unstable individual signals. In addition, a lack of individual signals for short periods due to server, network or other problems will not affect the procedure seriously (Wiemken and Heydenreich, 2004).

Figure 6: Example: Pre-defined error profile (footprint-table) for shading. Probabilities are distributed to the yellow marked circles; the blank circles contain probabilities of zero.

Summary

In the different parts of the PVSAT-2 project good progress has been made in developing a comfortable PV system surveillance.

The development of the footprint method so far has been a successful effort in error detection. The part of the method described in this paper gives a first view on the functionality. The entire error detection routine and also the footprint algorithm will be able to consider more errors than described here.

The PVSAT-2 procedure will be validated in a one year field test in Germany, the Netherlands, and Switzerland.

e PVSAT-2 project is supported by the 5th framework programme of the European Community under the contract number NNE5-2001-00571.

Models for PV/Wind Hybrid Power Generating System

Mustafa Engin, Dr., Solar Energy Institute, Ege Mes. Yuk. Okulu, Ege University, Bornova,

35100, Izmir, Turkey

Metin Colak, Dr., Department of Electrical &Electronics Engineering, Faculty of
Engineering, Ege University, Bornova, 35100, Izmir, Turkey

Configuration of the PV/wind hybrid energy system depends on the wind speed and solar irradiation at the considered site. The cost-effective, reliable design and appropriate operation of the hybrid systems is important. The design problem of the hybrid system is non-linearity due to the non-linear component characteristics. This non-linearity problem was solved using genetic algorithm by Seeling-Hochmuth and Marrison (1997. Beyer and Langer, (1996), determine the size of the hybrid system using limited meteorological parameters. Kellogg et al. (1998) developed a simple iterative technique that is based on energy balance, for sizing PV-wind hybrid energy system. Green and Manwell, (1995), Morgan et al. (1997) presented software tools that assess hybrid system performance for pre-determined system configuration. Habib et al. (1999) defines cost ratios between the PV system, wind power systems, and storage for cost-effective hybrid system design.

In this paper a hybrid system mathematical model was proposed. The hybrid system model contains simple mathematical models for each individual element of PV/wind hybrid power generating system. The models for PV cell, wind turbine and battery are based on model description found in the literature. The other components models for PV/wind hybrid system, namely charger, inverter, converter, load and controller are based on electrical and electronics knowledge. Proposed hybrid system model can be used to size cost-effective hybrid power generating system configuration with highest reliability. The implementation is done using MATLAB — SIMULINK, a simulating program. Simulating hybrid system model can also be used for predicting the performance of hybrid power generating systems. A comparison is made between model simulations and measurements that taken from PV/wind hybrid power generating system settled at Solar Energy Institute for lighting.