Bad Weather vs. Power Reliability Progress Florida Seeks to Normalize Reliability Indices

By Charles Williams, Progress Energy Florida, and David Kreiss,
Kreiss Johnson Technologies

Reliability indices gauge how well a utility uses incremental resources to benefit its customers. Reliability indices are also among the key factors by which regulators judge most utilities and determine whether they can justify rate increases. Perhaps most importantly, utilities use the indices to design T&D system upgrades and to calculate employee bonuses.

Click here to enlarge image

The disadvantage inherent in the indices most commonly applied—SAIDI (system average interruption duration index), SAIFI (system average interruption frequency index) and CAIDI (customer average interruption duration index)—is that the performance indicators they measure are heavily dependent on the weather. Lightning strikes, ice build-up on lines and wind-blown branches frequently trigger faults, which are the most common external causes of customer outages. Unfortunately, these forces of nature, over which the utility has no control, can wreak havoc on reliability indices.

If weather patterns were similar from year to year, their impacts on reliability index trends would cancel each other out. However, since that is not the case, the positive impact of enhancements made to a T&D system can be overwhelmed by weather-related outages during years of excessive lightning, wind or ice.

More than one utility has found itself in the unfortunate situation of having invested millions of dollars in new equipment to reduce the incidence and duration of outage, but, due to abnormal weather conditions in the same or next year, the utility might end up with lower reliability scores than before the investment was made. As a direct result, an anticipated rate increase proposed to pay for the investment might be turned down causing the entire organization and its shareholders to suffer. In the long run, customers may also feel the pain of poorly designed system enhancements resulting from the weather-skewed index calculations.

Given the need to have reliability indices accurately reflect system design, equipment condition and operating procedures, there has been a need for a methodology to normalize reliability by suppressing, but not eliminating, the effects of abnormal annual weather conditions. Fortunately, the technology is now available to do just that.

A Bad Weather Year

In 2003, Progress Energy Florida experienced the injustice that weather can inflict on reliability measurements. The utility’s service territory was hit by nearly 140,000 lightning flashes, a jump of 45 percent over the average number of annual strikes in the period 2000-02 (see Figure 1). Last year’s bad weather was not the first for the utility. Its personnel realized that reliability scores in Florida were strongly influenced by weather variations.

Click here to enlarge image

The utility’s Power Quality and Reliability Department decided to quantify the impact weather has historically had in the service territory. Personnel examined a 10-year reliability database (1990-99) to determine causal relationships between outages and weather factors. Statistical algorithms calculated the degree of correlation between outages and weather. Techniques such as maximum likelihood analysis were used to estimate unknown causes and allow proper analysis for drivers of these outages.

Click here to enlarge image

Many outage causes are undetermined by field personnel due to the difficulty in providing true fuse save operation for fuses in high-fault current locations. These outages, with no visible cause, are coded “unknown” in good weather and “storm/wind” during bad weather. Maximum likelihood analysis of these unknown and storm/wind- caused outages shows that 58 percent were probably caused by lightning, while 35 percent were related to animals. In Florida, lightning is obviously a key trigger of outages. Progress Energy Florida obtained historical data from lightning detection systems and performed studies of the correlation between monthly outage counts and lightning flashes for the entire service territory. With animal-related outages excluded, there was an extremely high correlation (See Figure 2).

Click here to enlarge image

Progress Energy Florida analyzed the outage frequencies by months and seasons and found three distinct patterns:

Click here to enlarge image

  • Summer Peaking. This includes lightning, primary cable, storm, tree, and defective equipment caused outages (see Figure 3);
  • Spring and Autumn Peaking. These outages are primarily animal caused (see Figure 4); and
  • Constant or Base Load. Outages from most other causes when combined form an almost constant level of outages through the year (see Figure 5).
Click here to enlarge image

The Power Quality and Reliability personnel then set out to normalize the indices to account for these weather-induced service fluctuations. The objective was to modify the SAIDI, SAIFI and CAIDI equations in a way that compares expected reliability performance with actual performance averaged over preceding years. This would enable a utility to factor in the T&D system’s response to weather, charting a more realistic performance trend relative to baseline operations.

Normalizing for Weather

Adjusting reliability indices for weather first requires customer interruption counts to be corrected. Then the duration data remains unchanged, and the frequency (SAIFI) and overall reliability index (SAIDI) are adjusted to show the net reliability based on the weather data.

Click here to enlarge image

Prediction of customer interruptions adds another variable to the equation—the device type. A lightning flash may cause a transformer station outage with only one customer outage. The same lightning flash may cause a feeder outage, which results in outage to 3,500 customers. Inherently the correlation coefficient will be lower when this additional variable is included. The next step was to exclude animal-caused outages, since previous pattern analysis shows they are not weather-related. This resulted in a correlation coefficient to 0.819 as shown in Figure 6.

Attempts were made to correlate outages to both wind and precipitation data, but the correlations were poor. The available data for wind and precipitation is point sample data, taken at points such as airports where instrumentation is readily available. Thus, a localized thunderstorm, which significantly impacts reliability, is not captured by the wind measurements at a site 30 miles away.

Lightning data is measured by radio signal, which makes it an integrated measured data. All lightning (subject to system detection efficiency) is captured over the entire area. The data must be geographically filtered to correlate data only where the utility has asset exposure on the ground. The high correlation between lightning and outages does not mean lightning is the direct cause of the outages. However, it does indicate that lightning data is a good barometer of the types of violent weather that cause outages.

The methodology for weather normalization of reliability indices is to predict customers interrupted based on actual lightning data. Adding in actual animal outages gives a total of customers interrupted for the year. This is converted to a SAIFIW index. SAIFIW is then compared to actual total SAIFI for the year.

A SAIFI lower than the SAIFIW indicates the system response to weather has improved. A SAIFI greater than SAIFIW shows the system response to weather is deteriorating. These SAIFI values are then multiplied by the actual CAIDI values for the year to produce SAIDI and SAIDIW. The comparison of SAIDI and SAIDIW allows a true comparison of how the system performance is changing, by comparing actual performance to predicted performance based on weather.

Click here to enlarge image

Figure 7 shows the results of these weather adjustments for the Progress Energy Florida service area.

The trends clearly show reliability, as measured by SAIDI, is improving during recent years. The apparent reduction in SAIDI is real and not just the luck of good weather. This process can also be used to explain apparent reliability deterioration due to bad weather.

Collecting Accurate Data

Collecting the data required to perform the normalized calculations has been a time-consuming and difficult process, which is certainly one of the reasons it hasn’t been done in the past. But technology is now available to automate much of the collection and reporting process. Automation not only makes these calculations practical, but, because it is less subjective, it also adds a sense of legitimacy needed for official acceptance by state public utility commissions.

Before even considering a change to normalized indices, the PUCs will undoubtedly require the data collection methodology to be proven as accurate and repeatable. The challenge, therefore, is to devise a method for automatically identifying faults and determining if they are caused by lightning or other weather-induced disturbances.

Technology has been developed to analyze the non-operational data stored in automated substation devices. This software is primarily used to assist utilities in improving the overall efficiency of their transmission operations, but it can also be applied to automatically collect fault data and determine its cause.

The system establishes a live link to the Vaisala lightning database ( maintained in Tucson, Ariz. This database contains North American lightning strike details, including exact time and location, typically to within 500 meters. The software correlates each strike within a given area, such as the Progress Energy Florida service territory, with data from the utility’s fault records. Specifically, the system can analyze each fault’s waveform to determine its exact inception.

If the inception point coincides with a lightning strike recorded in the area, the fault’s cause can be traced to the stroke with a high degree of certainty—providing a critical input to the weather-normalized reliability calculation. The software can also pinpoint the fault’s location, which offers the added benefit of leading the repair crew directly to the damage site.

This methodology is accurate and repeatable, which should satisfy the PUCs, and it is fast and low-cost, which will satisfy utilities. The primary advantage of this technique is that it is automated. The system can be programmed to automatically notify an engineer every time a fault is caused by a lightning strike. The process takes minutes instead of the hours required for an engineer to do it manually.

Successful implementation of this system requires four pieces of data: precise fault time, record of the fault current, the spatial coordinates of the faulted circuit or apparatus, and the lighting strike data. As mentioned, the time-stamped lightning data comes from Vaisala. Many utilities will need to add GPS-synchronization to their substation automation equipment so fault inceptions are precisely timed. And, to pinpoint the fault location, the utility must have its transmission network georeferenced and mapped on a GIS.

Of course, not all weather-related faults are caused by lightning. Wind-blown tree limbs and ice build-up on lines can be just as devastating to transmission operations as lightning. Progress Energy Florida is now working on viable solutions to identify and track these types of faults as well. The software already can handle inputs of wind or temperature data, but the challenge is collecting the data.

One solution is to install sensors throughout the service territory. These sensors would record wind, temperature, precipitation and other weather data continuously at known locations and feed this data back to the software. The system would then cross-reference fault times and locations with instances of wind gusts or low temperatures to determine the probable cause of the problem, just as it would with lightning strikes. Vaisala is preparing products and services to directly address this requirement.

To compute reliability indices, engineers must enter each fault’s cause, and not all faults are caused by weather. Animals, for instance, are a major cause of system faults. Modern information technology can automate this time-consuming fault-identification process. Neural logic can be programmed into the system so it can teach itself to identify probable fault causes based on other external data factors. For instance, if a fault occurs in a heavily wooded area in the spring during relatively high temperatures, clear skies and no precipitation, the system can narrow the origin down to an animal crossing a line or perhaps an equipment failure.

Using this technique, a semi-automated method can be developed with the IT software. The engineer can choose from a list of possible fault causes—ones he believes may be the cause—and then have the system remember this scenario using neural logic. The system then can be trained over time to identify the most likely source of a non-weather related fault.

Making the Change

The importance of reliability indices cannot be overstated. They reflect the quality of service utilities provide to their customers as well as the condition of their T&D infrastructure. While the absolute value of a reliability index is useful, the trend of that index over time holds even greater value.

The task to normalize for weather and improve reliability index accuracy does not require extensive resources or manpower. Fault data is generally available due to the explosive growth of substation IEDs. At least one substation automation system provides the ideal platform for the tasks of fault identification, analysis and reporting as well as directly embedding the normalization algorithms. Precise weather data via direct access is made available through Vaisala.

Utilities, investors and customers have too much at stake in the calculation of reliability indices to allow unforeseen weather factors to detract from precise measurement of the utility’s ability to respond to customer needs. Now that the methodology and technology are available to normalize these indices taking weather into account, the industry should move toward accepting these new formulas and techniques.

Charlie Williams graduated from the University of Florida with a B.S.E.E. and has been employed for the past 33 years in the distribution engineering department at Florida Power (now Progress Energy). As principal engineer, distribution asset performance, he is responsible for reliability analysis, design, development and optimization of reliability programs, power quality consulting and problem investigations. He is an IEEE senior member and is a past chairman of the Overhead Distribution Committee of the Southeastern Electric Exchange. Charlie has been a licensed Professional Engineer in the State of Florida since 1975. Charlie can be contacted at charles

David Kreiss is president of Kreiss Johnson Technologies, a developer of substation data acquisition and analysis software for the utility industry based in San Diego, Calif. Kreiss can be contacted at or 858 535-2088.


  • The Clarion Energy Content Team is made up of editors from various publications, including POWERGRID International, Power Engineering, Renewable Energy World, Hydro Review, Smart Energy International, and Power Engineering International. Contact the content lead for this publication at

Previous articlePOWERGRID_INTERNATIONAL Volume 9 Issue 5
Next articleCONSOL Energy, FirstEnergy to evaluate advanced coal technologies
The Clarion Energy Content Team is made up of editors from various publications, including POWERGRID International, Power Engineering, Renewable Energy World, Hydro Review, Smart Energy International, and Power Engineering International. Contact the content lead for this publication at

No posts to display