DATA DRIVEN RULE-BASED PEAK SHAVING ALGORITHM FOR SCHEDULING REFRIGERATORS

This paper proposes a novel method for the real-time prediction of photovoltaic (PV) power output by integrating phase space reconstruction (PSR), improved grey wolf optimization (GWO), and long short-term memory (LSTM) neural networks. The proposed method consists of three main steps. First, historical data are denoised and features are extracted using singular spectrum analysis (SSA) and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Second, improved grey wolf optimization (GWO) is employed to optimize the key parameters of phase space reconstruction (PSR) and long short-term memory (LSTM) neural networks. Third, real-time predictions are made using LSTM neural networks, with dynamic updates of training data and model parameters. Experimental results demonstrate that the proposed method has significant advantages in both prediction accuracy and speed. Specifically, the proposed method achieves a mean absolute percentage error (MAPE) of 3.45%, significantly outperforming traditional machine learning models and other neural network-based approaches. Compared with seven alternative methods, our method improves prediction accuracy by 15% to 25% and computational speed by 20% to 30%. Additionally, the proposed method exhibits excellent prediction stability and adaptability, effectively handling the nonlinear and chaotic characteristics of PV power.

Electricity peak shaving for commercial buildings using machine learning and vehicle to building (V2B) system

With the rapid development of Chinese colleges, the proportion of building energy consumption in total building energy consumption is rapidly expanding. The prediction of building energy consumption in colleges has become an increasingly important research topic in recent years. However, the most current building energy consumption forecasting models focus on the prediction research of the total energy consumption, but fewer about the perceptions of itemized building energy consumption. To solve this problem, an itemized energy consumption short term prediction model is given in this paper, which is based on long short-term memory (LSTM) neural networks. The input data include total history building energy consumption, local weather characteristics, date characteristics, and time characteristics. According to the results of experiments, an accurate prediction of lighting energy consumption and special energy consumption can be given by this model, which accounts for 96.51% of the total building energy. Moreover, the LSTM model can also achieve smaller computational complexity with little difference in prediction accuracy compared with other combined models.

This research introduces a novel machine learning-based strategy for generating supercapacitor (SC) reference current to optimize energy distribution in Battery Electric Vehicles (BEV) and Hybrid Battery Electric Vehicles (HBEV). A Long Short-Term Memory (LSTM) neural network is trained using real-world drive cycle data and exported in Open Neural Network Exchange (ONNX) format for real-time deployment within a Simulink-based control environment. This enables adaptive SC current prediction to dynamically offload high transient loads from the battery. The system is modeled using the Nissan Sakura EV and evaluated under EUDC and IM240 drive cycles. With SC support, the EUDC cycle exhibits a 21.3% reduction in battery peak current, 18.1% reduction in peak power demand, and 5.75% lower battery energy consumption. In the IM240 cycle, peak battery current is reduced by 33.5%, peak power by 31.6%, and energy consumption by 12.36%. These improvements validate the proposed LSTM-ONNX framework in reducing battery stress, improving thermal performance, and enhancing energy efficiency. Additionally, SC assistance leads to smoother traction motor torque, reduced current ripple, and optimized power delivery. A comprehensive model is developed and tested in Simscape to confirm the real-time applicability of this data-driven control strategy for electric vehicles.

The rapid proliferation of electric vehicles (EVs) has created unprecedented challenges for power grid infrastructure, particularly in managing the high-power demands of fast-charging stations. This research presents a novel approach to dynamic load balancing in EV fast-charging stations through machine learning integration with smart grid systems. The study develops and validates a predictive model that combines Long Short-Term Memory (LSTM) networks, Random Forest, and Gradient Boosting algorithms to optimize power distribution while maintaining grid stability. Using real-world data collected from 50 fast-charging stations across urban and suburban locations over a 12-month period, our proposed system demonstrates significant improvements over traditional methods. The results show a 27% improvement in load distribution efficiency and a 15% reduction in peak demand. Furthermore, the system achieves 92% accuracy in demand prediction and reduces charging waiting times by 8%. Grid stability metrics also improved substantially, with a 30% reduction in voltage fluctuations and a 20% improvement in power factor. The economic impact analysis reveals a 23% reduction in operational costs and a 35% improvement in energy utilization. The system's integration with smart grid infrastructure demonstrates robust scalability and adaptability to varying demand patterns. These findings suggest that machine learning-based load balancing can significantly enhance the reliability and efficiency of EV charging infrastructure while supporting grid stability. The proposed approach provides a promising foundation for future developments in smart charging systems, particularly in the context of increasing EV adoption rates and the growing need for sustainable transportation infrastructure.

The electrification of residential heating systems, crucial for achieving net-zero emissions, poses significant challenges for low-voltage distribution networks. This study develops a simulation model to explore the integration of heat pumps within active building systems for community heating decarbonisation. The model optimises heat pump operations in conjunction with thermal energy storage units to reduce peak demand on low-voltage networks by using real-time measured electricity demand data and modelled heat demand data for 76 houses. The study employs an algorithm that adjusts thermal storage charging and discharging cycles to align with off-peak periods. Three scenarios were simulated: a baseline with unoptimised heat pumps, a fixed threshold model, and an active building model with daily optimised thresholds. The results demonstrate that the active building model achieves a 21% reduction in peak demand on the low-voltage substation compared to the baseline scenario; it also reduces the total electrical energy consumption by 12% and carbon emissions by 17%. The fixed threshold scenario shows a 16% improvement in peak demand reduction, but it also shows an increase in energy consumption and emissions. These findings highlight the potential of active buildings to enhance the efficiency and sustainability of residential energy systems, marking a significant step toward decarbonising residential heating while maintaining grid stability.

A new hybrid model for wind speed forecasting combining long short-term memory neural network, decomposition methods and grey wolf optimizer

The purpose of the NETL Project was to develop a diverse combination of distributed renewable generation technologies and controls and demonstrate how the renewable generation could help manage substation peak demand at the ATK Promontory plant site. The Promontory plant site is located in the northwestern Utah desert approximately 25 miles west of Brigham City, Utah. The plant encompasses 20,000 acres and has over 500 buildings. The ATK Promontory plant primarily manufactures solid propellant rocket motors for both commercial and government launch systems. The original project objectives focused on distributed generation; a 100 kW (kilowatt) wind turbine, a 100 kW new technology waste heat generation unit, a 500 kW energy storage system, and an intelligent system-wide automation system to monitor and control the renewable energy devices then release the stored energy during the peak demand time. The original goal was to reduce peak demand from the electrical utility company, Rocky Mountain Power (RMP), by 3.4%. For a period of time we also sought to integrate our energy storage requirements with a flywheel storage system (500 kW) proposed for the Promontory/RMP Substation. Ultimately the flywheel storage system could not meet our project timetable, so the storage requirement was switched to a battery storage system (300 kW.) A secondary objective was to design/install a bi-directional customer/utility gateway application for real-time visibility and communications between RMP, and ATK. This objective was not achieved because of technical issues with RMP, ATK Information Technology Department’s stringent requirements based on being a rocket motor manufacturing facility, and budget constraints. Of the original objectives, the following were achieved: • Installation of a 100 kW wind turbine. • Installation of a 300 kW battery storage system. • Integrated control system installed to offset electrical demand by releasing stored energy from renewable sources during peak hours of the day. Control system also monitors the wind turbine and battery storage system health, power output, and issues critical alarms. Of the original objectives, the following were not achieved: • 100 kW new technology waste heat generation unit. • Bi-directional customer/utility gateway for real time visibility and communications between RMP and ATK. • 3.4% reduction in peak demand. 1.7% reduction in peak demand was realized instead.

In April 1992, the Energy Information Administration (EIA) released data on 1989 and 1990 electric-utility demand-site management (DMS) programs. These data represent a census of US utility DSM programs, with reports of utility expenditures, energy savings, and load reductions caused by these programs. In addition, EIA published utility estimates of the costs and effects of these programs from 1991 to 2000. These data provide the first comprehensive picture of what utilities are spending and accomplishing by utility, state, and region. This report presents, summarizes, and interprets the 1990 data and the utility forecasts of their DSM-program expenditures and impacts to the year 2000. Only utilities with annual sales greater than 120 GWh were required to report data on their DSM programs to EIA. Of the 1194 such utilities, 363 reported having a DSM program that year. These 363 electric utilities spent $1.2 billion on their DSM programs in 1990, up from $0.9 billion in 1989. Estimates of energy savings (17,100 GWh in 1990 and 14,800 GWh in 1989) and potential reductions in peak demand (24,400 MW in 1990 and about 19,400 MW in 1989) also showed substantial increases. Overall, utility DSM expenditures accounted for 0.7% of total US electric revenues, while the reductions in energy and demand accounted for 0.6% and 4.9% of their respective 1990 national totals. The investor-owned utilities accounted for 70 to 90% of the totals for DSM costs, energy savings, and demand reductions. The public utilities reported larger percentage reductions in peak demand and energy smaller percentage DSM expenditures. These averages hide tremendous variations across utilities. Utility forecasts of DSM expenditures and effects show substantial growth in both absolute and relative terms.

In April 1992, the Energy Information Administration (EIA) released data on 1989 and 1990 electric-utility demand-site management (DMS) programs. These data represent a census of US utility DSM programs, with reports of utility expenditures, energy savings, and load reductions caused by these programs. In addition, EIA published utility estimates of the costs and effects of these programs from 1991 to 2000. These data provide the first comprehensive picture of what utilities are spending and accomplishing by utility, state, and region. This report presents, summarizes, and interprets the 1990 data and the utility forecasts of their DSM-program expenditures and impacts to the year 2000. Only utilities with annual sales greater than 120 GWh were required to report data on their DSM programs to EIA. Of the 1194 such utilities, 363 reported having a DSM program that year. These 363 electric utilities spent $1.2 billion on their DSM programs in 1990, up from $0.9 billion in 1989. Estimates of energy savings (17,100 GWh in 1990 and 14,800 GWh in 1989) and potential reductions in peak demand (24,400 MW in 1990 and about 19,400 MW in 1989) also showed substantial increases. Overall, utility DSM expenditures accounted for 0.7% of total US electric revenues, while the reductions in energy and demand accounted for 0.6% and 4.9% of their respective 1990 national totals. The investor-owned utilities accounted for 70 to 90% of the totals for DSM costs, energy savings, and demand reductions. The public utilities reported larger percentage reductions in peak demand and energy smaller percentage DSM expenditures. These averages hide tremendous variations across utilities. Utility forecasts of DSM expenditures and effects show substantial growth in both absolute and relative terms.

India being a developing country has the need of both addition in generation and replacing the old ones commensurate with ever-increasing demand in the domestic, agricultural, commercial, and industrial sectors. With Demand Side Management (DSM) resorted to effectively, however, there exists potential of shaving peak, and consequently to some extent energy in this important infrastructure of economy. Having an installed capacity of about 81 GW from Renewable Energy Sources (RES) out of a total of 360 GW by July 2019 it has an ambitious plan for significant addition of renewables reaching the level of 175 GW (100 GW from solar including 40 GW of rooftop, 60 GW from wind, 10 GW from bio-mass, and 5 from small hydro), and 275 GW respectively by the end of country’s 13 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> (March 2022) and 14 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> five-year plan (March 2027). Thus, as estimated, by 2030 it may be in a position to contribute about 48% from renewables only. This is based on detailed studies on load forecasting in different sectors geographically over pan-India in different time-frame followed by estimation of potential from RES. The latter consist of development from solar, wind, biomass, waste, etc. both off-grid and on-grid. Finally planning has been carried out simulating likely scenario at different point of time with system having sizeable penetration of renewables in the overall requirement of generation to meet the electricity demand. Technological developments, standardization, and regulatory measures have paved the way for large-scale integration of renewables to the Extra High Voltage (EHV) grid by pooling the surplus from one region for haulage to other regions for distribution of electricity. In the process gradually conventional fossil-fuel based generating plants are being phased out, though it continues with still addition of the backlog in the system. However, in the process Plant Load Factor (PLF) of such type of generation is coming down enabling lessening pollution too. With economy of scale and more accuracy achieved in predicting intermittent generation from renewables, it has already been possible to achieve much reduction in per unit charges of electricity from RES, notably from both, solar and wind.In the paper based on Long-Term Load Forecasting, results of studies to find out optimal mix of existing conventional fossil-fuel based generation integrated with renewables from various sources have been depicted based on computation. While carrying out studies specifically for the two periods ending on March 2022 and March 2027, development envisaged in respect of renewables has been taken into account, and so the reduction in demand due to DSM. Corresponding peak demands projected to be met appear to be about 226 GW and 299 GW with annual energy requirement to the tune of 1,566 and 2,047 Billion Units (BU) of electricity. Reduction in peak demand of 9 GW and 12 GW and energy requirement of 206 and 273 BU too is expected on account of DSM.As per studies projected renewables would be accounting for 175 GW out of a total of 479 GW by March 2022, while 275 GW out of 619 GW by March 2027. Consequently, in terms of installed capacity overall percentage of non-fossil-fuel based generation would rise to 49.3% and 57.4% respectively. However, considering additional coal-based capacity requirement vis-à-vis under construction as well as retirement of old ones, overall contribution of energy from this segment would remain significant, although with reduced average Plant Load Factor (PLF). Considering the transition period of 10-12 years from now, Long-Term studies have been carried out corresponding to the time-frame 2029-30 to find out the optimal mix of primarily RES and fossil-fuel based Thermal plants. Thereafter for the year 2029-30 corresponding to the various critical days how such planned system meets the peak load as well as fulfil energy requirement has been studied. From the results it is observed that 48% of energy is expected to come from RES with installed capacity touching about 65% of the total installed capacity of 831 GW by 2030, predominantly with Solar. The latter has significant role in meeting the daily load demand vis-à-vis energy out of different types of generation. But in the studies with projection of data, at the time of actual peak of the day, occurring typically in the evening, there is almost no contribution from Solar. So, though installed its presence could not been considered at the time of peak demand. This is indeed a big constraint with further growth in load due to economic development or otherwise and desirability of phasing out fossil-fuel based power plants. Under such circumstances flattening of load curve by the extensive use of Pumped-Storage Hydro (PSH) plants, BESS, etc. are definitely the viable option as alternatives, as considered.

Existing methods of state of charge (SOC) estimation have limitations such as requiring an accurate battery model or frequent calibration, making them unsuitable for energy storage system (ESS) applications. These limitations can be overcome using machine learning (ML) techniques. Among ML- based techniques, long short-term memory (LSTM) has a feedback-loop characteristic, which considers current and historical SOC. At present, LSTM has not been used for batteries in ESS during peak demand reductions. Therefore, this paper studies the use of LSTM for battery SOC estimation in ESS during peak demand reductions. LSTM is found to be effective even though the network begins with an inaccurate SOC, which makes it a suitable technique for batteries in ESS for grid services. The estimation accuracy of the LSTM is also higher than conventional SOC estimation techniques such as coulomb counting and empirical model, and other ML techniques such as the feedforward neural network.

Peak load reduction and load shaping in HVAC and refrigeration systems in commercial buildings by using a novel lightweight dynamic priority-based control strategy

Unlike prior work on demand management, which typically requires industrial loads to be turned off during peak times, this paper studies the potential to carry out demand response by modifying the elastic load components of common household appliances. Such a component can decrease its instantaneous power draw at the expense of increasing its duration of operation with no impact on the appliance's lifetime. We identify the elastic components of ten common household appliances. Assuming separate control of an appliance's elastic component, we quantify the relationship between the potential reduction in aggregate peak and the duration required to complete the operation of appliances in four geographic regions: Ontario, Quebec, France and India. We find that even with a small extension to the operation duration of appliances, peak demand can be significantly reduced in all four regions both during winter and summer. For example, during winter in Quebec, a nearly 125 MW reduction in peak demand can be obtained with just a 10% increase in operation duration. We conclude that exploiting elasticity to reduce peak power demand should be an important consideration for manufacturers. From a policy perspective, our study gives regulators the ability to quantitatively assess the impact of requiring manufacturers to conform to smart appliance standards.

Bearings are the most prevalent and easily damaged mechanical components in mechanical systems. Nowadays, Deep learning emerges as a very efficient artificial intelligence method that offers a new approach to feature extraction from raw data. Long short-term memory (LSTM) neural network is a promising type of deep learning, which is mainly used for the diagnosis of faults. In this work, a new intelligent fault diagnosis and classification method based on Long Short-Term Memory (LSTM) neural network is developed. The developed model correctly classifies the different types of faults at different running conditions in real operation conditions, and the results are compared with existing techniques. The performance of the model is evaluated with various performance metrics such as Precision, Recall, F1 score, Receiver Operating Characteristics, Area under Curve, and Accuracy. The best results are obtained from the developed LSTM model than from the existing work.