Computer Vision and Deep Learning for Electricity

---

Table of Contents

---

Computer Vision and Deep Learning for Electricity

Universal access to affordable, reliable, and sustainable modern energy is a Sustainable Development Goal (SDG). However, lack of sufficient power generation, poor transmission and distribution infrastructure, affordability, uncertain climate concerns, diversification and decentralization of energy production, and changing demand patterns are creating complex challenges in power generation.

According to the 2019 International Energy Agency (IEA) report, 860 million people lack access to electricity, and three billion people use open fires and simple stoves fueled by kerosene, biomass, or coal for cooking. As a result, over four million people die prematurely of the illnesses associated.

Artificial intelligence (AI) offers a great potential to lower energy costs, cut energy waste, and facilitate and accelerate the use of renewable and clean energy sources in power grids worldwide. In addition, it can help improve the planning, operation, and control of power systems. For example, intelligent grids embedded with an information layer, created through the widespread installation of smart meters and sensors, allow for data collection, which is used to build AI algorithms (Figure 1).

This series is about Computer Vision (CV) and Deep Learning (DL) for Industrial and Big Business Applications. In addition, this blog will cover the benefits, applications, challenges, and tradeoffs of using deep learning for electricity (power sector).

This lesson is the last in a 5-part series on CV and DL for Industrial and Big Business Applications 101.

1. Computer Vision and Deep Learning for Government
2. Computer Vision and Deep Learning for Customer Service
3. Computer Vision and Deep Learning for Banking and Finance
4. Computer Vision and Deep Learning for Agriculture
5. Computer Vision and Deep Learning for Electricity (this tutorial)

To learn about Computer Vision and Deep Learning for Electricity, just keep reading.

---

Computer Vision and Deep Learning for Electricity

---

Benefits

---

Efficiency

Deep learning algorithms learn by trial and error. When trained on energy patterns, the algorithms make electricity production efficient. For example, in Norway, Agder Energy partnered with the University of Agder to develop deep learning algorithms to optimize water usage in hydropower plants. Even though water seems like an endless source of energy, only a limited amount of it is available to produce electricity.

---

Reliability

AI can forecast electricity demand and supply, storing the excess solar/wind/hydropower generated in low-demand times and used in high-demand times. This can improve the reliability of solar and wind power by analyzing the massive amount of meteorological data and using this data to predict whether to gather, store, or distribute power. In addition, AI can further explore the grid before and after intermittent units are absorbed and learn from this to help reduce congestion and renewable energy curtailment.

Uncertain climate factors have also disrupted the reliability of electricity production. Generally, countries have reliable power production data collected over 40 years, which can be fed to deep learning models to forecast electricity production.

---

Security

Power losses and theft issues are widespread these times. The daily power consumption data collected through digital meters and fed into a single device connecting to an AI service identifies and clusters unusual patterns relative to customer profiles in similar areas. The data anticipate customer behavior and predict which customers will likely have informal connections to the power grid. Brazilian business magazine Exame named Ampla’s anti-theft system as one of the top innovations that address this problem.

---

Applications

---

Fault Prediction and Maintenance

Equipment failure is a common problem in the energy sector and can have serious consequences. Hence their identification and maintenance are crucial. For example, faults in distribution networks can occur due to lightning, insulation defects, sabotage, tree branches, and animal interference resulting in a short circuit. With AI’s help, sensors can monitor equipment and detect failures before they happen, thus saving resources, money, time, and lives.

These AI-based fault prediction systems can be coupled with drones to automate maintenance (Figure 2). First, the AI system can analyze weather and demand forecasts to generate drone flight schedules. Then, during its flight, the drone will capture the network’s high-resolution images, which will be sent to a cloud-based AI application to identify the health of the network and classify it as “functioning,” “malfunctioning,” or “sub-optimal functioning.” As an outcome, the AI application generates asset inspection reports that can be used for creating maintenance/repair work orders.

AI-powered Fault Detection solutions are economical as well as cost-effective. Such solutions help operators determine the overall system’s health state to take proactive measures to prevent a catastrophe. For example, Schneider Electric uses Microsoft’s machine learning capabilities to remotely monitor and configure pumps in the oil and gas field. As a result, early detection of a pump failure can save up to $1 million in repair costs.

Geothermal energy is another example of a rich amount of data used by predictive diagnostics to detect problems that can shut down power plants. Preventive measures such as chemical agent spray to avoid turbine shutdowns are optimized (quantity, composition, and timing) using the Internet of Things (IoT) and AI.

Long short-term memory (LSTM) networks, support vector machines (SVMs), and principal component analysis (PCA) are widely used in fault detection in power systems. For example, Wang et al. (2019) propose a stacked sparse autoencoder-based network with SVM and PCA to demonstrate its application to real-world data. They also use artificial neural network (ANN)-based methods to illustrate the model’s effectiveness when detecting the time and location of faults. Furthermore, Shafiullah and Abido (2019) propose a semi-supervised ML model consisting of a K-nearest neighbor (KNN) model and a decision tree model to handle labeled and unlabeled data. Finally, Helbing and Ritter (2018) evaluate the effectiveness of deep ANNs in wind turbine fault detection by detecting faults at an early stage.

---

Efficient Decision-Making

Intelligent devices like Alexa, Google Home, and Google Nest can help customers interact with their thermostats and other appliances to monitor their energy consumption. In addition, AI systems equipped with automatic meters can optimize a home’s energy consumption and storage. For example, it can turn off appliances when energy is expensive/limited and store in batteries whenever it is in abundance (solar energy).

By utilizing grid data, smart meter data, weather data, and energy use information, AI can study and improve building performance, optimize resource consumption, and increase comfort and cost efficiency for residents (Figure 3). In addition, it can enhance the electricity forecast demand and generation, which will be helpful in the transition toward renewables, as their productivity is often dependent on inconsistent factors like weather, wind, water flows, fossil fuels, etc. Furthermore, AI-based forecasts combined with energy storage infrastructures can reduce the need for such backup systems.

Since the usage of solar systems in households and industries is increasing significantly, consumers will become a part of a distributed network, act as producers and contribute to power generation. In such a distributed ecosystem, AI can predict optimal times for distributed generation to contribute to the grid rather than draw from it (Figure 4).

Furthermore, in deregulated markets like the United States, AI can help consumers by suggesting an energy provider based on their energy source preferences, budget, and consumption patterns. For example, Carnegie Mellon University researchers have created a machine learning system called Lumator which takes customers’ preferences and consumption data. The data contains information on the different tariff plans, limited-time promotional rates, and other product offers and provides recommendations for the most suitable electricity supply deal.

As customers’ habits keep changing, the system automatically switches to new energy plans when better deals become available without interrupting the supply. These solutions can help increase the share of renewable energy by assisting consumers in converting their preference for renewables into realized demand for it and can be used to signal to producers the level of consumer demand for renewable energy.

---

Energy Trading

Trading energy is different from trading commodities because it has to be delivered instantly. This is a massive challenge for energy traders. Load forecasting helps energy traders and regional providers to calculate electricity generation and energy pricing. AI algorithms can make trading more efficient by predicting the energy demands and providing real-time energy prices, which can help traders to make the right selling and buying decisions. These algorithms can make more accurate predictions by consuming more data from micro weather conditions, meter level consumption data, and social media postings like Twitter (Figure 5).

When asked about the future of AI and machine learning in the energy market, Larisa Chizhova, Principal Data Scientist at Inspired, answered: “I am certain that we are witnessing a transition from manual to completely automated electronic trading in the energy market right now, similar to what happened in the financial sector 10-20 years ago. And the use of AI and machine learning plays a major role in this transition. We, as humans, can draw only very few conclusions from the data we see, and we easily miss important trends. Finding trading signals in the enormous amount of trading, weather, and load data is a task for machines to crunch.”

Studies have been done on LSTMs and SVMs to achieve load forecasting at the individual household level. However, due to volatile user behavior, the method utilizes the historical load and appliance measurement to deliver better results than the existing classical methods. Another class of techniques called the bottom-up hierarchical approach proposes leveraging the new perspective of smart meters. K-means clustering is first used to group customers by the similarity of their energy usage behaviors. Deep neural networks (DNNs) are then used for different clusters, and finally, the aggregated load is obtained by adding up the electricity consumption from all clusters.

Probabilistic forecasting provides more information on the uncertainties of future electricity consumption. Initial probabilistic forecasts are based on recurrent neural networks (RNNs). The copula model is employed in the next stage to obtain a multivariate distribution. The method can deliver accurate forecast distributions; moreover, the generated scenarios would benefit the energy aggregators and traders.

Together with peer-to-peer (P2P) blockchain technology, AI can help galvanize electric vehicle (EV) charging stations’ deployment (Figure 6). On the EV owner’s side, AI technologies can aid in selecting the suitability and management of charging stations. In contrast, on the charging station’s side, blockchain technology-based platforms can enable the discovery of charging stations and smart contracting. AI will initiate the charging process by requesting the blockchain-based charging platform. This request contains EV location, preferred charging time, etc. Simultaneously, charging stations will keep sending queries to the trading platform for new demanding requests. After a match, the bidding process is started and repeated until the price matches. Finally, AI displays the route map on the consumer’s mobile/tablet screen. A smart contract executes between the EV’s owner and charging station, and its record gets stored on the blockchain platform.

---

Losses Due to Informal Connections

Losses due to informal electricity connections are another challenge for the power sector. Theft and fraud of electricity costs as much as $96bn every year globally, with as much as $6bn yearly in the United States alone. AI can analyze discrepancies in customer usage patterns, payment history, etc., to detect such informal connections. Furthermore, when combined with automated meters, they can improve monitoring.

Brazil, for example, is suffering from high electricity theft issues. The University of Luxembourg has created an ML algorithm that can detect abnormal electricity usage by analyzing information from electricity meters. The algorithm yielded good results when applied to information over five years from 3.6 million Brazilian households.

Presently, the most popular scheme to detect abnormal usage patterns in electricity is to collect power consumption data from smart grids, upload them to a centralized database, and analyze them through intelligent algorithms. The prevalent anti-power-stealing algorithm includes clustering, which detects areas with dissimilarity in supplied and billed power. A typical load curve extracted by K-means clustering realizes the load forecasting (Figure 7).

However, electricity monitoring data is dynamic. Hence, the data analysis difficulty lies in finding the abnormal data from the constantly updated dynamic data flow to predict the theft users accurately. RNNs are effective for monitoring and analyzing time-series dynamic data flow to produce a neuron output by combining the current status data with the previous status data of the system.

Naturgy, a Spanish utility, developed an AI-based solution to detect non-technical losses (NTL) in its power supply system. The solution built a “predictive model” from historical campaigns and assigned a score to all the customers so that the utility could inspect cases with high NTL possibilities. Moreover, inspections were designed with 36-50% accuracy — much higher than manually planned inspections.

---

Challenges

Like other sectors applying AI technology, the power sector also faces challenges (e.g., governance, transparency, security, safety, privacy, employment, and economic impacts).

---

Data Protection and Security

We constantly share data with centralized authorities when connected to a digital system. Any system breach can leak confidential information that can be used against the consumer (Figure 8). A successful cyberattack can be as damaging as a natural disaster. The world’s first successful attack happened in Ukraine in 2015, leaving thousands without power.

Cybersecurity is becoming increasingly essential to protect AI-enabled electric grids from leaking customer data. The growing threat from hacking has become an ordinary matter of significant concern, mainly because smart metering and automated control represent close to 10% of global grid investments, equivalent to $30 billion a year dedicated to digital infrastructure.

---

Power Consumption

Processing large amounts of data consumes energy on its own. Therefore, when using AI-based systems in power plants, it is crucial to analyze how to design energy-efficient and climate-neutral data centers. In addition, engineers should keep in mind the physical proximity of data centers and renewable energy generation plants and the postponement of power-intensive computing operations when less power is available.

---

Lack of Knowledge, Data, and Transparency

In rural and low-income areas, the accessibility of cellular and digital technologies is limited. Since intelligent meters rely on constant data communication, AI-equipped power systems’ reliability is not guaranteed. Furthermore, low communication means insufficient data for machine learning models to learn from, making them susceptible to inaccurate data.

Moreover, while the applications of AI in the power sector are multiple, there is a need to educate the AI industry more deeply on the aspects of the power sector (e.g., the regulatory restrictions on using cloud-based applications in the power industry).

---

Summary

Because of uncertain climate concerns, changing demand patterns, diversification, and decentralization of energy production, the energy sector faces many challenges. The adoption of Artificial Intelligence (AI) can help the energy sector lower its costs and waste and accelerate the use of renewable energy in the following ways:

• Fault Prediction and Maintenance: With AI’s help, sensors can monitor equipment and detect failures before they happen, thus saving resources, money, time, and lives. These AI-based fault prediction systems can be coupled with drones to automate maintenance.
• Efficient Decision-Making: Intelligent devices like Alexa and Google Home can interact with customers’ thermostats and other appliances to monitor their energy consumption. In addition, AI systems equipped with automatic meters can optimize a home’s energy consumption and storage.
• Energy Trading: AI algorithms can make trading more efficient by predicting the energy demands and providing real-time energy prices, which can help traders make the right selling and buying decisions.
• Losses Due to Informal Connections: AI can analyze discrepancies in customer usage patterns, payment history, etc., to detect such informal connections. Furthermore, when combined with automated meters, they can improve monitoring.

However, AI in the energy sector comes with its own challenges.

• Data Protection and Security: Any system breach can leak confidential information that can be used against the consumer.
• Power Consumption: Processing large amounts of data consumes energy on its own. Therefore, when using AI-based systems in power plants, it is crucial to analyze how to design energy-efficient and climate-neutral data centers.
• Lack of Knowledge, Data, and Transparency: Since intelligent meters rely on constant data communication, AI-equipped power systems’ reliability is not guaranteed. Furthermore, low communication means insufficient data for machine learning models to learn from, making them susceptible to inaccurate data.

I hope this post helped you understand the benefits, applications, challenges, and tradeoffs of using deep learning in electricity. Stay tuned for another lesson where we will discuss deep learning and computer vision applications for oil and gas.

---

Consultation Services

Do you need some help with your industrial application of AI? Learn more about our consulting services.

---

Citation Information

Mangla, P. “Computer Vision and Deep Learning for Electricity,” PyImageSearch, P. Chugh, R. Raha, K. Kudriavtseva, and S. Huot, eds., 2022, https://pyimg.co/p3ncy

---

@incollection{Mangla_2022_CVDL4Elec,
  author = {Puneet Mangla},
  title = {Computer Vision and Deep Learning for Electricity},
  booktitle = {PyImageSearch},
  editor = {Puneet Chugh and Ritwik Raha and Kseniia Kudriavtseva and Susan Huot},
  year = {2022},
  note = {https://pyimg.co/p3ncy},
}

---