Published On: Apr 06, 2020

Power Quality (PQ) assessment requires simultaneous observation of multiple parameters and continuous monitoring of the respective data points. The fact that these data points could run into millions for each parameter is a key challenge in assessment of PQ. The use of manual or equipment-based compilation overview in data monitoring and classification poses a serious constraint due to the same reason. Also, the rule-sets for separating events that depict poor PQ from the good ones, between two sites, could be different. Site-specific PQ may really depend on the power profile of the site, conditions of electrical system infrastructure and several other local parameters. Even human-assisted analysis of PQ depends on drawing insights from the data by observing for patterns and their underlying connections.

Parallelly, PQ monitoring is fast evolving from just solving PQ issues post occurrence to real-time monitoring of power systems and intervention for their better performance. Naturally, this has led to a massive increase in the amount of data that is being collected which necessitates advanced software-assisted analysis. The ideal tool for PQ assessments should be able to describe the types of power quality variations and facilitate their characterizing for meaningful reporting. Finally, the tool needs to summarize the information in such a way so as to present useful and actionable reports. This is exactly where Artificial Intelligence has immense potential in driving efficient assessment of PQ through automated characterization of parameters and events.

INTRODUCTION

Thanks to advancement in technology today, collecting data regarding the flow of electric power has become quite easy. Of course, everyone is familiar with data-gathering about power consumption and the bills generated as a result. But another interesting application of data is to monitor the quality of power that reaches us. Occurrences of poor PQ events such as loss of supply whether temporary or prolonged, voltage dips or swells, harmonics, voltage transient etc have a deep and adverse impact on the industry. The performance of equipment is hampered, premature failure takes place, work stops, and costs escalate.

Conventionally, PQ is monitored by recording values of voltage and current every ten or fifteen minutes. Then the data is analysed either manually with traditional signal processing techniques or automatically by feeding it into computers. However, research is now considering the use of artificial intelligence and deep learning to continuously monitor power quality. This could also help recognize patterns of disturbances, detect any issues that were previously unseen and even predict possible digression from defined framework.

PQ DATA = BIG DATA

As data-gathering becomes more efficient and power networks grow and evolve, more and more data are being generated about flow and quality of power. Hence, reliance on automatic systems for analysis of data is increasing. At the same time, the vast amount of data generated is beginning to resemble Big Data. Digging into it could uncover various other details about power quality. In fact, we are somewhat familiar with predictive maintenance for production equipment. Power Quality Data is somewhat akin to that. In recent times, the number of sensitive loads such as power electronics in electrical networks continues to rise. Currently, the computer-analysed data is used to learn information on two aspects

1. Potential impact of power disturbances on equipment
2. The cause of disturbance in power quality

Assessing impact of power disruptions on equipment could be useful in many ways. Such information could help improve the ride-through capability of sensitive devices and setup protection systems in case of power failures. Similarly, accurately learning the causes of power disruption could help improve the quality of power distribution network (power lines, grounding systems etc.). This could help avoid such disturbances in the future. In fact, advanced data analytics can even predict possibility of loss or failure of an equipment functioning within a system.

Equally evolved methods of analysis are needed to understand how these sensitive loads behave during disturbances in power, and how to measure the impact of their behavior on PQ itself. Artificial Intelligence shows the way.

PQ FOR SENSITIVE LOADS, RENEWABLES DEMANDS DATA DRIVEN SOLUTIONS

Renewable power generators such as wind turbines and photo voltaic are double-edged swords. On one side, they could be the source of disturbances like reactive power drawn from grid affecting voltage profile or injecting harmonics, while on the other side, they are also incredibly sensitive to PQ disturbances that originate from other parts of the grid such as dips or swells in voltage, small disruptions in power supply. The voltage dip may result in a large voltage drop on the DC-link voltage of a converter part of a renewable electricity generator and consequently resulting in an unwanted trip when the under-voltage relay removes the RE generator from the grid. The voltage swell may result in tripping of overvoltage protection or power electronic switch failure.

Summary of PQ Issues, Characterization parameters and root causes

As technology advances, more and more new and sensitive equipment is being added to the power distribution grid. the new devices have different immunity thresholds than conventional devices. Developing new immunity requirements needs a closer look at power flow data. As technology changes, vast amount of data is collected, and developing capabilities to better analyse the data is the key to ensure uninterrupted power supply to the industry.

ARTIFICIAL INTELLIGENCE FOR PQ ASSESSMENTS

Broadly, here’s how applying Artificial Intelligence can be applied to measuring and assessment of PQ more effectively:

• Spotting the nature of the interference (nuisance tripping, malfunctioning etc.) owing to voltage disturbance
• Evidence for developing benchmarks for equipment modifications such as grid codes, standards for critical safety equipment etc.
• Frequency level monitoring of each type of PQ events to better understand the fundamentals
• Establishing co-relation between unrelated events such as harmonics induced due to inrush currents from LED lights
• Detect root causes by identifying origins of events electrical faults, device switching etc. through data patterns
• Extracting real-time trend analysis of the electrical system information (real-time trends, fault type and location, transformer sizes and impedance, load information, capacitor information, cable or line impedance, etc.)

Automated PQ data analysis, especially characterization of PQ events, is the most effective way to detect, extract and assess the potential PQ events in order to enhance corresponding power system.

PQ DATA COMPLEXITY NECESSITATES ARTIFICIAL INTELLIGENCE

As mentioned above, it has been understood that vast amount of data can now be collected about power quality.

For instance, a three-phase power quality monitor typically records 10-minute values for rms, harmonics (2nd to 50th order) and inter harmonics (0 to 39), for both voltage and current. Assuming three voltages and three currents it gives 25 million values per year per device. Further, each monitor samples typically with 25.6 kS/s over 6 or 8 channels. All this data creates ‘volume’ complexity in the PQ data monitoring system.

PQ assessment may include processing of multitude of data sources starting from remote energy meters, transmission system operation, smart grids, power system protection, forecasting and even weather data. Processing of such heterogeneous data creates the ‘variety’ complexity for the data.

The rate of incoming data such as smart meters provide data by the month, day or by the hour. But PQ monitors with standard settings could collect up to 23 samples with a rate of 28 kS/s across 6 to 8 channels. This creates complexity in terms of handling the data ‘velocity’ complexity in assessment of PQ.

Last but not the least, all this data may not be consistent. There could be lost packets, inconsistency, errors and more in the data collection. These too must be recognized and overcome in the Artificial Intelligence and could be characterized as the ‘veracity’ issue related to PQ data.

Together, the volume, variety, velocity and veracity complexities in the PQ data create a strong case for Artificial Intelligence based solution in characterization of PQ data.

ARTIFICIAL INTELLIGENCE BASED PQ DATA CLASSIFICATION

The use of Artificial intelligence (AI) in PQ assessment essentially involves automation of human thinking activities including decision-making, problem-solving, learning, perception, projecting and reasoning for the resolution of complex problems.

Neural Networks for classification of PQ disturbances

Neural network is a highly promising tool for classification of PQ data. It is a nonlinear, data driven self-adaptive method, and already well proven for several other real-world classification tasks in the industry, medical and scientific applications. Like in Human body, the learning ability of the neural networks allows pattern recognition by experience. The learning is developed during the training phase when various examples are presented to neural networks. PQ disturbances may not be exactly similar between two sites but tend to have similar patterns. This is where the neural network-based classification comes handy.

Support Vector Machine (SVM) Based Classification for Voltage Sags

SVMs are a special kind of neural network with better accuracy when it comes finding non-linear boundaries – which in simple terms is to minimise the misclassification probability of data for unseen patterns. Studies have shown SVM based algorithm have been able to classify up to 98.8% of the voltage sags, accurately.

Rule-based Expert Systems Classification for voltage dips

To formulate an expert system – a rule-based system for PQ disturbance classification system is built based on a knowledge base composed using a form of expert knowledge. Rule-based expert systems are driven by “if…then” logic and can be leveraged for classification of PQ events such as voltage dips by rules defining the estimation of the amplitude.

Fuzzy Expert System Based Classification

The fuzzy logic system is dependent on the rules specified by the know-how of human experts. Typically, the power systems data is highly uncertain. Effective power disturbance monitoring requires pattern classification. The systems can be used for classification of PQ parameters such as voltage sags, through a complex analysis of the noise in the power network. Several other applications of the system are under research.

Genetic Algorithm Based Classification Genetic Algorithm (GA) for individual PQ disturbances

A probabilistic method, based on the biological evolution, is highly useful in accurate capture of unique and salient characteristic of each PQ disturbance. There are several successful approaches based on this method which consists of a pre-processing unit in combination with genetic algorithm for classifying the power system fault disturbances.

The above list is only indicative of new and prevalent systems for using Artificial Intelligence in characterization of various PQ parameters. Further to this, several hybrid methods are constantly evolving along with totally new approaches. Artificial Intelligence is a science that is being applied in every sphere of life from warfare to customer support. It’s greater use and application in mapping PQ disturbances is critical to advance the science of improving PQ.

CONCLUSION

Systematic procedures for evaluating PQ concerns must include the entire system from transmission to end use facilities. PQ events that show up in the end user facility may emanate outside the facility’s electrical network. At the same time PQ measurements must bring in site specific insights. The PQ data analysis solution therefore should be able to process a range of data from variety of instruments and formats. While continuous data monitoring enables consummation of time trends and statistics, characterization of individual events to make sense of the health of the system calls for Artificial Intelligence solutions. These solutions, enabled by deep learning, can teach themselves through iterative procedures. It not just reduces the near impossible human effort that would otherwise be required to analyse the data but is also able to drive continual learning from the ever-expanding body of data and analysis generated historically.

The main task of PQ analysis involves detection, identification, recognition and classification of various types of PQ disturbances. Automatic power quality monitoring and artificial intelligence-based classification methods will be therefore vital to PQ assessment. In the next few years, as these techniques evolve, it will be at the core of reporting and dashboards of PQ assessments for industrial, commercial and retail facilities. In fact, it would not be wrong to say that the successful shift of Customer and Utility focus to ‘quality’ of power, depends on the successful adaptation of Artificial Intelligence techniques in PQ assessment.

REFERENCES

1. Interpretation and Analysis of Power Quality Measurements – https://www.dranetz.com/wp-content/uploads/2014/02/powerquality-measurements-interpretation-analysis-1.pdf
2. Power Quality Disturbance Detection Using Artificial Intelligence: A Hardware Approach – https://www.researchgate.net/publication/232614981_Power_Quality_Disturbance_Detection_Using_Artificial_Intelligence_A_Hardware_Approach
3. Adaptive Method for Power Quality Improvement through Minimization of Harmonics Using Artificial Intelligence – https://www.researchgate.net/publication/315109673_Adaptive_Method_for_Power_Quality_Improvement_through_Minimization_of_Harmonics_Using_Artificial_Intelligence
4. Comparison of Different AI Techniques for Power Quality Improvement using Statcom – https://pdfs.semanticscholar.org/b532/1c380cffda1d020316464f1e28e7ab550a4e.pdf
5. An artificial immune system with continuous-learning for voltage disturbance diagnosis in electrical distribution systems – https://dl.acm.org/doi/10.1016/j.eswa.2016.03.010