Energy Informatics最新文献

Background

Power grid automation is critical for maintaining the stability and reliability of electrical grids. A major challenge in power grid management is identifying and isolating faults quickly and accurately to avoid widespread disruptions. Traditional fault detection and isolation methods rely on rule-based diagnostics, which frequently struggle for speed, precision, and adaptability to changing fault conditions. As power grids become more complex, intelligent algorithms are critical for improving the efficiency of fault localization and isolation.

Problem statement

Conventional fault management methods, like rule-based and heuristic methods, have limitations in both accuracy and real-time adaptability. To address these issues, this study proposes and assesses two intelligent algorithms: the Fault Localization Algorithm (FLA) and the Fault Isolation Algorithm (FIA). Unlike conventional methods, FLA incorporates machine learning methods to improve fault detection, whereas FIA provides an optimized isolation strategy, decreasing operational delays and reducing power disruption.

Methodology

The FLA algorithm uses a Support Vector Machine (SVM) classifier to predict fault locations based on key variables like voltage, current, frequency, line impedance, and meteorological conditions. The FIA algorithm then uses the FLA output to evaluate fault severity and select the best fault isolation strategy. This approach combines an SVM-based fault localization algorithm with a decision-tree-based fault isolation strategy to guarantee quick and accurate fault identification, reducing system downtime and enhancing fault resolution efficiency. The proposed system is validated with the PowerGrid Fault Localization Dataset (PGFLD), which contains practical power grid fault data.

Results

Experimental findings show that the proposed FLA algorithm achieves 92% accuracy, outperforming traditional techniques like Decision Tree (85%), KNN (82%), and Logistic Regression (78%). Furthermore, FIA achieves 95% accuracy, outperforming current rule-based (89%) and heuristic (85%) methods. These findings show a significant enhancement in fault detection accuracy and isolation effectiveness, which reduces false positives and improves power grid resilience.

Conclusion

This study provides an innovative method of power grid fault management that employs intelligent algorithms for fault localization and isolation. The use of SVM for fault localization and decision-tree-based fault isolation improves fault detection accuracy while reducing operational delays. The proposed methods improve grid resilience and offer actionable isolation tactics, making them extremely effective for contemporary power grid automation.

Multiple methods have been developed and implemented to reduce dependence on fossil fuels and conserve electricity. However, accurately predicting electricity consumption is essential before reducing it. Forecasting building electricity consumption has become increasingly critical, as buildings account for 39% of global electricity consumption. Among these, campus buildings are particularly energy-intensive. In this study, we used Monte Carlo (MC) simulations—trained on each leaf that generated by the regression tree (RT) algorithm—to predict the electricity consumption of Widya Dharma University Pontianak (UWDP)’s campus building. Unlike traditional approaches that rely on the mean of samples within a leaf, our method incorporates their likelihood. Since RT algorithms are prone to overfitting, training each leaf individually is expected to mitigate this issue. The data were collected by measuring hourly electricity consumption on one floor of the UWDP campus building over several months. The proposed MCRT prediction algorithm achieved an accuracy of 91.61%, with a Root Mean Square Error of 3.49 and a Normalized Root Mean Square Error of 0.09.

The transition to sustainable energy increasingly relies on robust communication infrastructure to monitor, control, and optimize energy distribution. Supervisory Control and Data Acquisition (SCADA) networks manage these processes, transmitting sensor data and control commands. However, integrating (smart) communication systems into an ageing existing communication infrastructure introduces vulnerabilities to cyber-attacks, such as false data injection and man-in-the-middle attacks. Although recent advancements in Intrusion Detection Systems (IDS) for SCADA networks show potential in detecting domain-specific threats, testing has largely been confined to simulations due to the nature of critical infrastructure. This paper presents two real-world case studies using actual grid data, where a process-aware IDS solution is tailored to specific network topologies. The result effectively detects various cyber-attacks, including those targeting critical devices like transformers. This work marks a crucial step toward practical deployment, emphasizing the need for a gradual transition from simulation to real-world validation to ensure the safety and reliability of critical grid infrastructure.

The prediction of new energy generation is challenging due to its intermittency and uncertainty. To solve this, we propose a framework combining an optimized multiscale convolutional neural network (MSCNN) and long - short - term memory network (LSTM). MSCNN improves feature extraction with dynamic scale selection and deep residual modules. LSTM captures long - term dependencies better using bidirectional processing and attention mechanisms. We also introduce a fuzzy decision support system (FDSS) to handle prediction uncertainty. Our model outperforms ARIMA, SVM, Gradient Boosting, CNN, and RNN in hourly, daily, and weekly predictions. It also excels in uncertainty quantification and generalization, offering strong support for accurate new energy generation prediction and dispatch.

To achieve adaptive and accurate ultra-short-term power generation forecasting in power systems, this study proposes a novel prediction method combining Sparrow Search Algorithm (SSA) with Long Short-Term Memory (LSTM) networks. The methodology involves the following steps: (1) Collecting historical ultra-short-term power generation data from photovoltaic systems, where outlier detection and data cleaning are performed using horizontal processing methods; (2) Applying Pearson correlation analysis to identify key meteorological factors significantly influencing power output as feature inputs; (3) Developing an Adaptive Sparrow Search Algorithm (ASSA) by dynamically adjusting the quantities of discoverers and followers in traditional SSA; (4) Optimizing LSTM network parameters through ASSA to enhance prediction accuracy. The experimental results demonstrate superior performance with Root Mean Square Error (RMSE) values of 0.075, 0.088, and 0.089 for sunny, cloudy, and variable weather conditions respectively. The corresponding Mean Absolute Percentage Error (MAPE) values are 0.21 MW, 0.52 MW, and 0.13 MW, while Absolute Error (AE) values reach 0.17 MW, 0.46 MW, and 0.18 MW. These findings confirm the method’s effectiveness in achieving precise ultra-short-term power generation forecasting across diverse weather conditions.

Amid the escalating global climate change issue, environmental design and thermal management have emerged as research focal points. Leveraging virtual reality (VR) technology’s immersive and interactive advantages, this study developed an integrated system for green building environment design and thermal energy management. By analyzing green building design needs and exploring VR’s architectural application potential, a thermal energy management system integrating sensor, data processing, and intelligent control technologies was constructed. Field experiments in a green building project demonstrated that VR technology enhanced energy consumption prediction accuracy by 20% and shortened the design cycle by 30%. The system also achieved a 15% reduction in energy consumption, optimizing thermal management. This research not only created an efficient integrated system but also proposed energy - saving strategies, promoting green building sustainability.

In the context of global response to climate change and promoting energy transformation, the rapid popularization of electric vehicles and the widespread application of clean energy have become important components of modern power systems. However, the charging demand of electric vehicles brings new challenges to regional power grids, especially those that rely on clean energy, due to its uncertainty and randomness. This study examines the impact of EV charging demand on the control efficiency of clean energy-based regional power grids. Using real grid data and time-series simulation, we develop a dispatch optimization framework incorporating a master-slave game model based on wind power output distribution. We simulate EV charging patterns, renewable fluctuations, and uncertainties in user behavior and station availability. The results show that unmanaged charging increases peak load by up to 20%, while optimized strategies like Time-of-Use (TOU) pricing, Direct Load Control (DLC), and Vehicle-to-Grid (V2G) reduce the peak-valley gap by 15%, improve renewable energy consumption by 12%, and lower curtailment. These findings offer valuable insights for EV integration and clean energy planning in regional grids. The results show that at a 30% EV penetration rate, the peak charging demand may lead to a 20% increase in the regional grid load, and by optimizing the charging time, the peak-valley load difference can be reduced by 15%. In addition, a reasonable charging strategy can help improve the utilization rate of clean energy, maximize the consumption of wind power and photovoltaic power generation, and reduce dependence on fossil fuel power generation.

Global climate change and greenhouse gas emissions have become significant challenges requiring urgent solutions. As an energy-intensive sector, the low-carbon transformation of the civil aviation industry plays a critical role in achieving China’s “dual carbon” goals. In this study, taking China’s civil aviation industry as a case study, a multi-scenario dynamic coupling model is constructed based on the LEAP (Long-range Energy Alternatives Planning) model. Using historical data from 2000 to 2024, carbon emissions from 2025 to 2060 are quantitatively simulated and the dynamic feedback mechanisms among demand, energy, technology, and policy modules are systematically analyzed. The results indicate that under dual policy and technological drivers, especially in the scenarios of low-carbon policy and technological progress, carbon emissions in 2050 could be reduced by over 50% compared to the baseline scenario. This study provides a scientific basis for formulating precise low-carbon policies, optimizing the energy structure of civil aviation, and promoting the research and development of new aircraft.

Traditional power supply decision methods rely on fixed and rigorous mathematical models, which are difficult to accurately capture the characteristics and changing patterns of new loads, resulting in low prediction accuracy. Therefore, a decision model for guaranteeing power supply is constructed based on an improved proximal policy optimization algorithm, to study the intelligent guarantee power supply decision method. The experimental results show that the stability of the proximal policy optimization algorithms is generally high in all scenarios, especially in fault or anomaly scenarios and low load demand scenarios, which exceeds 110%. Its loss value decreases with the increase of training iterations. At 60 iterations, its loss value reaches the optimal value of 100 and then tends to stabilize. The research results indicate that the intelligent power supply strategy has good feasibility. This decision method helps to improve the stability, efficiency, intelligence level, and ability to respond to emergencies of the power grid.

To achieve the rapid unsupervised learning of multi-source information, this paper studies a multi-source information integration method for the “dual carbon” smart monitoring center based on the improved federated learning. To solve the problem of rapid integration information from many sources in the “dual carbon” smart monitoring center, a multimodal federated learning framework is built on the basis of the traditional federated learning. The generator and discriminator of the conditional generative adversarial network model are used to distinguish between the generated pseudo-samples and normal samples, and the multi-source information is obtained unsupervisedly. Based on the global distribution, the fast integration is achieved by using the passive distillation method of federated data. At the same time, the stochastic gradient descent is used to enhance the learning rate, improve the learning ability of the model, and promote the unsupervised fast fusion. The experiment shows that this method can effectively integrate the multi-source information, display the spatial status of carbon emissions and enterprise energy production data. The integrated information has high completeness and entropy value, and is accurate and applicable in the multi-source information integration of the “dual carbon” smart monitoring center.