Animals and Vegetation

Some of the most common causes of power outages include vegetation and animals interfering with electrical equipment. Problems caused by vegetation include trees or tree limbs falling on power lines and roots growing into feeder cables. Squirrels, birds, snakes and other animals interacting with power lines, transformers, or other equipment can also easily cause an outage.

Predicting where problems with vegetation and animals are likely to occur is not as daunting as it may sound.  Vegetation interference can be detected both through smart readings, image analysis, and past maintenance records. The same image recognition technology that is driving the development of self-driving cars can be used to recognize trees in satellite and other images, giving an indication of where vegetation may be near an aerial line. Changes in power and voltage along a power line may also indicate that vegetation is becoming a problem. Also, maintenance records contain valuable information about where vegetation has been a problem in the past. All of this information can be integrated to predict areas of the network which may be prone to problems.

The computer vision technology allowing automated detection of objects such as trees in images is well developed. Since 2012, the problem of automatic image recognition has been attacked most successfully by one type of algorithm, convolutional neural networks. The imagenet competition illustrates how revolutionary the technique of convolutional neural networks was for the field of computer vision. The annual contest involves taking a set of images, classifying the object in the images into 1000 classes, and drawing a box around the classified item. In 2010 and 2011, before the advent of convolutional neural networks, the winning entries had misclassification error rates of 28.2 percent and 25.8 percent respectively. In 2012, a group from the University of Toronto won with an error rate of only 16.4 percent, crushing the efforts of the other teams. The method used to win the competition, a convolutional neural network, has become the standard method for attacking image recognition problems.

Further refinements of convolutional neural networks over the past 4 years have yielded increased accuracy. In this year’s imagenet competition, Microsoft research used a 150 layer neural network to classify a set of 100,000 images with an error rate of less than 3.5 %. Computer vision technology has been used by driverless vehicles to identify pedestrians, other vehicles, and signs quickly and accurately. It has been used for medical applications such as automatically detecting tumors in x-rays or the development of retinopathy from an eye exam photo. By applying computer vision to satellite and street-level images and comparing them with smart sensor readings from the same area, regions of potential interference of tree branches with aerial lines can be identified before major problems occur.

Animal interference with power lines and other electrical equipment is another common cause for power outages. At first, it may seem difficult to predict something as random as a squirrel jumping onto the aerial lines. However, there are things that can be done to predict the probability of animal interference in different areas. The problem is closely correlated with vegetation in the area.  Mathematical models exist for predicting the density of animals such as squirrels given the distribution and type of vegetation in an area.1 Pratter et. al formulated a multiple regression model indicating a relationship between forest density and squirrel density for a squirrel population in Arizona. The model, based on various measures of forest cover accounted for 89% of the variation in squirrel density over the region. Similar models of tree cover and types of vegetation in a region could predict the density of squirrels in the region.  Combining that information with information about past instances of squirrel interference and locations where squirrel guards have been installed could be used as a preventative measure.

In order to integrate the diverse information involved in predicting an outage, it’s necessary to combine data from multiple sources including sensor readings, weather data, maintenance records, and satellite imagery. GridCure is leading the way in data integration and predictive modeling solutions allowing utilities to access, visualize, and understand their data with a single easy-to-use interface. Our predictive modeling modules make sophisticated use of disparate data to answer a diverse range of questions; each module answers a specific question or group of related questions, and we’re able to stack our modules on top of one another allowing for a customizable and flexible solution. We currently offer several asset health predictive maintenance modules, as well as anomaly detection and technical and nontechnical loss modules. Always looking to expand our product offering to encompass the most pressing utility needs, we’re very interested in developing a vegetation module to address the many outages caused by animal and vegetation interference.

Interested in learning more about what we’re up to?  Sign up for our newsletter here!

 

1 “Landscape Models to Predict the Influence of Forest Structure on Tassel-Eared Squirrel Populations” John W. Prather, Norris L. Dodd, Brett G. Dickson, Haydee M. Hampton, Yaguang Xu, Ethan Aumack, Thomas D. Sisk, Journal of Wildlife Management, 7(03) (June 2006) pp. 723-731.