The Importance of AI-Enabled Internet of everything Services for Smart Home Management

Conventional homes will be transformed into smart homes due to significant technological breakthroughs and revolutions in human activities and lifestyles. Smart home technologies offer numerous benefits over traditional homes, including increased convenience and efficiency. They allow for remote access and control, optimize energy usage, automate routine tasks, enhance security, provide comfort and customization, promote health and well-being, and offer valuable data insights for informed decision-making. [1]. As smart home technologies are widely used, consumers will benefit from increased ease and security. Authentication and access control mechanisms are essential in ensuring the security of smart home devices and data. These mechanisms require users to verify their identity through passwords, biometrics, or multi-factor authentication, thus preventing unauthorized users from accessing sensitive information or gaining control of devices. The use of these mechanisms enhances overall security and privacy in smart homes. [2]. Since most individuals spend a significant portion of their time there, residential energy consumption has a global impact. Artificial intelligence (AI)-powered home systems can reduce energy costs while maintaining comfort. They analyze user behavior, optimize energy usage, coordinate appliances, and adapt to external factors like weather conditions to minimize energy waste. [3]. Energy storage systems (ESS) and thermostatically regulated loads might benefit from dynamic electricity pricing, allowing for a reduction in energy expenses. ESS in smart homes stores excess energy from renewables/off-peak hours and releases it during peak periods. This reduces reliance on the grid, minimizes bills, and ensures uninterrupted power supply. ESS enhances energy efficiency (EE) and sustainability in smart homes. About 40% of a home’s overall energy use goes toward heating, ventilating, and air conditioning (HVAC), one type of thermostatically adjustable load that causes sensible homeowners to worry about energy costs. [4]. Optimizing the energy cost of smart homes without compromising thermal comfort is vital since the major function of HVAC systems is to preserve the inhabitants’ thermal comfort. [5]. The Internet of things (IoT) is the infrastructure of connected, Internet-connected appliances that can be individually identified and may exchange data. [6]. In a smart home, one’s electronic gadgets may be tracked and managed from another location. The need for solid IoT security solutions has so increased. [7]. Internet of Everything (IoE) can be used to create smart homes where various household appliances and devices are connected to the Internet. Home automation using the IoT allows household appliances to be controlled automatically through the Internet. The appliances are connected to a central hub or network, which enables users to remotely monitor and control them using specialized apps or platforms on their smartphones or other Internet-connected devices. This offers greater convenience and flexibility for managing household appliances. Using smart home technology raises concerns about security, such as hacking, data breaches, privacy, and unauthorized access or control by malicious individuals [8]. This allows for remote control and automation of devices, making homes more convenient and energy-efficient. The IoE allows for the remote control and automation of devices in the home by connecting them to a central network. This enables users to manage appliances and systems remotely using smartphones or other devices. IoE facilitates convenience and EE through scheduling, monitoring, and adaptive control based on user preferences and real-time data [9]. The IoE is an intelligent networked environment that impasses people, data, things, and processes. Smart homes are customizable, energy-efficient, and convenient; nevertheless, they have downsides such as internet dependence and high cost [10]. IoE-based energy management systems optimize electricity use, switch to more cost- and resource-effective regimes, and determine effective and sustainable energy consumption strategies based on use patterns by analyzing real-time power consumption data. The advantages of IoE-based energy management systems for optimizing electricity use in smart homes include real-time monitoring and control, personalized energy profiles, integration with renewable energy sources, predictive analytics for EE, and remote access and automation capabilities. [11]. One disadvantage to a smart home is the probability of an Internet outage. Even if it only happens occasionally, being without electricity or the Internet is inconvenient. [12]. AI-powered home automation systems can track appliances, estimate their future energy needs, and modify use accordingly, all while keeping consumption within a set price range. AI-powered home automation optimizes comfort and energy usage through personalized settings, occupancy sensing, predictive analytics, demand response (DE) integration, and energy monitoring. [13]. AI-enabled Smarter system operations are possible because of the ability of IoT devices to evaluate data for patterns and insights. AI algorithms can analyze data from home IoT devices to identify usage patterns, detect anomalies, predict maintenance needs, optimize energy consumption, analyze user behavior, and assess environmental impact. These insights can help homeowners improve their home’s efficiency, sustainability, and comfort. This has the potential to lower energy costs without compromising comfort in homes. [14]. Eventually, AI will be included in practically every appliance in the house. Connected smart devices may carry out pre-programmed AI orders or react to a user’s voice commands from afar. Examples include Alexa, Siri, and Google Now. [15]. The AI is used to predict the smart home energy management system’s (SHEMS) future power consumption and turns off one or more devices to reduce the overall consumption at the end of the month [16]. Smart home energy management relies heavily on AI-based technologies to realize the five key energy optimization viewpoints of users, comfort, safety, design, and maintenance. [17]. Emerging technologies like deep neural networks (DNNs) may analyze these data for insights that help improve energy management in the here and now. [18]. DNNs may be trained on enormous datasets of past energy consumption and usage data and environmental and weather information to anticipate future energy consumption and make educated energy use decisions. [19]. DNN’s many hidden layers and neurons enable automated feature extraction, allowing for precise model regression or classification. When a neural network (NN) is trained using historical data, it learns underlying patterns and relationships. The model learns patterns within the data, allowing it to make accurate predictions for new instances. Regularization and cross-validation techniques prevent overfitting. After proper training, a DNN will exhibit great generalization and may be applied immediately to new instances without the need for time-consuming and resource-intensive numerical calculation. DNNs are a more computationally efficient alternative to traditional model-based approaches. This is due to their automatic learning of hierarchical representations from data, which reduces the need for manual feature engineering. Additionally, DNNs utilize the parallel processing capabilities of modern hardware, such as GPUs, to accelerate computation. This efficiency enables DNNs to handle large datasets and complex tasks accurately, making them superior for many applications. The DNN is far more computationally efficient than traditional model-based approaches without sacrificing accuracy. [20].

The main contribution of the paper is as follows:

Designing the AI-enabled Internet of Everything Services (AI-IoES) for efficient smart home energy management. AI-based systems for smart homes improve energy management efficiency. They offer predictive analytics for usage forecasting, personalized recommendations, automated scheduling, adaptive learning, and real-time monitoring for energy conservation.

The remainder of the paper is arranged as follows: Section 2 deliberates the related study, Section 3 proposes the AI-IoES model, Section 4 discusses the simulation outcomes, and Section 5 concludes the research article.

Xu et al. [21] suggested the multi-agent reinforcement learning based data-driven methods (MARL-DDM) for smart home management (SHM). Scheduling energy use an hour in advance was properly expressed as finite Markov decision processes (FMDP) with discrete time phases. A data-driven method was created to address this issue using an NN and Q-learning algorithm to provide optimal schedules for HEM systems at low cost. In particular, an extreme learning machine (ELM) was used to anticipate uncertainty based on the real-time processing of energy cost and solar photovoltaic (PV) generation data in rolling time frames. The outcomes of the tests show that the suggested data-driven home energy management system works as intended.

Zhang et al. [22] proposed the fuzzy expert systems efficient energy smart home management system (FES-EESHM) for the renewable energy resource. The suggested fuzzy expert framework was used to streamline the construction process of smart microgrids, including energy storage, renewable energy, and flexible load management. In addition, the FES framework enhances storage and energy to make the most of renewable energy and the smart grid’s economic benefits. The FES framework maximizes integrating renewable energy sources into the grid through energy storage technologies. It balances supply and demand, smooths intermittent generation and time shifts energy, provides grid support services, and optimizes renewable energy dispatch. The holistic approach promotes a more sustainable, resilient, and efficient energy system. Furthermore, electricity price, insolation, load energy, and wind speed may all be used as input variables in a controlled and uncontrolled manner by the fuzzy expert system to facilitate energy management. Inputs such as solar radiation, wind speed, energy quality, and the output of fixed and variable loads facilitate energy management. The outcome demonstrates that the product for consumer use and the expert technique can be tested through the hourly parameter value input.

Akbari-Dibavar et al. [23] recommended hybrid robust-stochastic optimization (HRSO) for SHM. Assuming the worst-case scenario for PV generation, flexible robust optimization approaches (ROA) were used to develop a controllable equal of the issue and handle the ambiguity of day-ahead (DA) market costs. When calculating ROA, you have two options – more conservative or less conservative. A more conservative approach prioritizes downside protection and risk mitigation over maximizing returns, resulting in lower expected returns but reduced volatility. A less conservative approach involves taking on more risk to pursue higher potential returns, resulting in higher expected returns but greater volatility and a higher risk of significant losses. Ultimately, the choice between the two depends on an investor’s risk tolerance, investment objectives, and time horizon. By varying the control variable’s value, the author finds solutions with varying degrees of conservatism in the ROA. Conservatism in the context of ROA is essential in financial analysis. It plays a crucial role in accounting for uncertainty, risk management, avoiding over-optimization, ensuring long-term sustainability, and instilling stakeholder confidence. Financial analysts and investors can better navigate volatile markets while preserving capital and achieving their objectives by prioritizing resilience to uncertainty and prudent risk management. Stochastic Programming (SP) accounts for the unpredictability of the RT energy market in the suggested optimization framework. Uncertain PV generation and energy pricing aspects were currently modeled using likely scenarios. Stipulating likely scenarios in uncertain PV generation and energy cost is crucial for risk management, decision-making, resource planning, financial analysis, scenario-based planning, and stakeholder engagement. It helps stakeholders assess potential risks, make informed decisions, plan resources, evaluate financial viability, develop adaptive strategies, and foster collaboration. The residents’ comfort was considered, and it was assumed that loads could be managed. The benefits of the suggested hybrid approach were shown via the analysis of the results, providing assurance to decision-makers on the financial viability of energy management.

Chen et al. [24] discussed the two-stage non-intrusive appliance load monitoring (TS-NIALM) for SHEMS and residential DSM. This research implements an artificial neural networks (ANN)-based NIALM method over fog-cloud computing to enable the SHEMS prototype to covertly monitor critical electrical appliances without the invasive connection of a plug-load power meter (smart plug). The SHEMS prototypes were built on small, intelligent, embedded IoT controllers that link the sensor and meter and act as smart homes or smart construction gateways to provide residential DSM. This research results show that a two-phase NIALM technique used in the existing SHEMS prototype was feasible and practical.

Chen et al. [25] deliberated the multi-objective reinforcement learning (MORL) for user preference-based DE in smart homes. The suggested technique improved over traditional algorithms by addressing the uncertainty of future cost and renewable energy production and minimizing the influence of the change in user preferences. The suggested method considers energy cost and consumer dissatisfaction using two Q-tables, and it adapts to user preferences by learning from past scheduling decisions to better suit individual needs. This means the suggested approach may be used for various problems. Furthermore, the simulation analysis of real-world data showed that the suggested method can provide exceptional performance following a shift in user preferences, with savings of 8.44% compared with mixed-integer nonlinear programming-based DE and an increase in dissatisfaction of only 1.37% on average.

Nakıp et al. [26] presented the recurrent trend predictive neural networks forecast embedded scheduling (RTPNN-FES) for renewable energy management in a smart home environment. The RTPNN-FES was an innovative form of NN architecture that can predict renewable energy production and plan when appliances will be used. RTPNN-FES is a predictive NN with a scheduling module that optimizes energy usage and resource allocation in smart grids. It uses a recurrent neural network (RNN) architecture, trend extraction, prediction, and forecast embedded scheduling module. It also has a hierarchical optimization framework and feedback mechanisms, and it undergoes training and optimization processes to improve prediction accuracy. Furthermore, RTPNN-FES provides a schedule resilient against forecasting mistakes because of its embedded structure, eliminating the need for separate algorithms for predicting and scheduling. RTPNN-FES adjusts to energy demand forecasting inaccuracies by continuously monitoring energy demand and consumption patterns, utilizing adaptive learning to update its predictive model, implementing dynamic re-optimization techniques, conducting scenario analysis to evaluate potential outcomes and uncertainties, incorporating risk mitigation strategies, and involving human intervention in cases where automated adjustments are insufficient. This approach enhances energy management systems’ resilience, reliability, and efficiency in smart grid environments. In addition to proposing an algorithm for an IoT-empowered smart home, this research assesses its performance. According to the study, RTPNN-FES was 37.5 times quicker than optimization, outstripping predicting approaches while providing near-optimal scheduling.

Lu et al. [27] suggested reinforcement learning and artificial neural networks (RL-ANN) for DE for home energy management. A stable price forecasting model based on an ANN was offered to deal with the ambiguity in future costs. In addition, multi-agent reinforcement learning was used with pricing projections for various household appliances to arrive at decentralized, optimum choices. Simulations are performed using movable and fixed loads and those that may be controlled to test how well the suggested energy management plan works. To compare the performance of a DE algorithm to a benchmark, researchers typically follow a systematic experimental design and evaluation procedure. This involves selecting an appropriate benchmark, designing an experimental setup, collecting data, defining performance metrics, executing predefined scenarios, conducting statistical and sensitivity analysis, and summarizing the findings. Researchers can assess the DE algorithm’s effectiveness, efficiency, and practical utility in real-world energy management scenarios. Compared with a benchmark without DE, experimental findings show that the suggested DE algorithm may greatly lower the user’s power cost by managing the energy consumption of numerous appliances.

Li et al. [28] proposed the novel smart energy theft system (SETS) for IoT-based intelligent homes. The prediction model, which employs a multi-model predicting system, is the first of three phases of decision-making modules. This system combines machine-learning models into a unified framework to estimate future energy usage. Organizations can implement a multi-model prediction system that integrates different machine learning models to accurately and reliably predict future energy consumption. The system involves collecting and preprocessing historical data, conducting feature engineering, choosing and training machine learning models, implementing an ensemble learning approach, generating predictions and aggregating them, evaluating and optimizing the system, and deploying it in operational environments with monitoring and feedback mechanisms. In the second, the main decision-making model employs simple moving averages (SMAs) to filter out out-of-the-ordinary data. Finally, the subordinate decision-making models create the third and last phases of the decision on energy robbery. The numerical outcomes show that the suggested system has a 97.96% success rate in detecting, which improves the security of the IoT-based intelligent home.

Alilou et al. [29] discussed the multi-objective scheduling method (MOSM) for home energy management in an inhabited smart microgrid. A grouping of multi-objective dragonfly algorithms and the analytical hierarchy processes technique is employed to maximize the techno-economic objective functions and locate the optimal scheduling of devices. The price-based DE scheme also includes real-time pricing tariffs. A smart microgrid consisting of 20 smart homes is used to test the effectiveness of the suggested technique. The numerical result verifies that the suggested home energy management approach successfully lowers the peak demand of the residential smart microgrid and the power cost of smart houses.

Rocha et al. [30] deliberated the AI-based scheduling algorithms (AI-SA) for demand-side energy management in intelligent households. DSM uses a genetic algorithm that considers power price variations over time, equipment priority, running cycles, and battery banks. Furthermore, the DSM deliberates a DA projection of dispersed generation using the support vector regression method. In addition, the K-means clustering technique was used to analyze actual data from a smart house to validate numerical simulations of the users’ experiences. When comparing intelligent homes with and without battery banks and dispersed generation, the suggested AI combination resulted in a 51.4% decrease in costs, proving its efficacy.

Franco et al. [31] recommended IoT-based load monitoring and activity recognition in smart homes. For load monitoring, this study suggests a new intrusive technique using the IoT architecture to build an activity detection system. The layers of the proposed IoT are the appliances, perception, communication networks, middleware, and applications. Using the appliance in question, the created activities of daily living algorithm assigns each ADL to a set of criteria. Data like on/off schedules and power consumption rates determine the most salient characteristics. The accuracy is >0.9 in both the feed-forward NN and the long short-term memory (LSTM) network, and it is close to 0.8 in the support vector machine network. Additional experiments are conducted using a different test set to assess the classifier model. Sensitivity analysis investigates how group size affects the classifier’s performance.

Kumar et al. [32] propose combining wind and solar energy to generate electricity and solve these sources’ variability and cost issues. The study uses advanced techniques like ANN-based expert systems and crop production systems to predict plant response and fertilizer allocation. The results show promise, with the suggested model explaining 96.9% of plant growth and development variation based on input variables.

Sathyaprakash et al. [33] present an overview of efficient e-healthcare risk prediction systems. Challenges include data quality issues, privacy concerns, and algorithm bias, but opportunities for innovation exist.

Based on the survey, there are numerous issues in existing models in achieving EE and user experience and reducing energy consumption, such as MARL-DDM [21], FES-EESHM [22], HRSO [23], TS-NIALM [24], and MORL [25]. Hence, this paper recommends AI-IoES for efficient smart home energy management.

The term “Smart Home Management” refers to the unified and comprehensive way a homeowner can monitor and adjust the state of all connected devices, from appliances and security sensors to cameras and power management. IoT home automation is a means of automatically controlling home appliances utilizing different control system methodologies. Controls for the home’s refrigerators, windows, lights, fans, kitchen timers, fire alarms, and other electronic machines may come in a wide variety. To prevent accidents in the home when home automation controls all electronic devices, implement safety protocols such as fail-safes, motion sensors to detect occupancy, regular maintenance checks, and manual overrides for critical systems. Connectivity options for IoT sensors include Bluetooth, ZigBee, Wi-Fi, and Z-Wave. An essential system component is a controller, sometimes a hub or gateway. It has an Ethernet cable hooked up to the modem/router at home. This central gateway is used for communication between the sensors and the control system, various security measures have been implemented to ensure that data transmitted through the central gateway remain private and secure. These measures include encryption protocols such as SSL/TLS, authentication mechanisms like biometrics or passwords, firewall protection, regular software updates, and strict adherence to industry standards and regulations, such as GDPR or HIPAA. The central gateway serves as the primary communication hub in a smart home, facilitating interaction between sensors, devices, and control systems. It manages data transmission, coordinates device commands, and ensures seamless connectivity throughout the home automation network. The IoE is a networked system of mechanical and digital machines, autonomous computing devices, and other items that may exchange data with one another and with other computers or IoE nodes without human operators’ intervention. By empowering networks and devices with the ability to learn from their own choices and actions in the past, forecast future behavior, and enhance their own performance and decision-making skills in real-time, AI releases the full potential of the IoE harnessed to analyze massive amounts of data generated by interconnected devices. This analysis enables predictive insights, autonomous decision-making, and process optimization, facilitating seamless communication and coordination within the IoE framework. Hence, this paper recommends AI-IoES for efficient smart home energy management. AI-powered home automation systems can track appliances, forecast their future energy needs, and adjust use accordingly. Using historical data, machine learning algorithms can help smart homes predict future energy usage. Popular algorithms include time series forecasting, regression analysis, NNs, support vector machines, decision trees, random forests, gradient boosting machines, and clustering algorithms. The algorithm depends on data complexity, model interpretability, computational resources, and system requirements. This has the potential to lower energy costs without compromising comfort. Furthermore, intelligent supporters like Google, Apple’s Siri, and Amazon’s Alexa allow one to operate a smart home by voice command.

Figure 1 shows the proposed AI-IoES model. The research novelty is incorporating smart and non-smart electrical devices into the IoT spaces by manipulating an IoT device as an intermediate between the smart internet-linked side and all other devices controlled over the Internet. The combination of AI and IoE services in smart home energy management contains identifying and categorizing device on/off conditions by measuring electrical signal only at one position of the inhabited user. Smart home appliance energy data are initially collected using IoT sensors [32]. Six sensors, including a water flow sensor, a motion sensor, an energy control sensors, gas sensors, temperature sensors, and a sound sensor, monitor the house’s data and events and relay them to a synchronized gateway. Connecting to an IoT gateways or edge device enables IoT devices to combine their sensor data and either send it to the cloud for analysis or retain it locally. Status data is the most fundamental data on a system’s or device’s current state. Data for the IoT is generated by automated appliances and infrastructure, such as smart thermostats and lights. Data acquired by sensors may be stored in one of three places: at the sensor node where it was gathered (local storage), at a remote data center (external storage), or at another node within the same sensor network (in-network storage). Then, IoES-based smart home energy management requests home data from each sensor provider to guarantee interior comfort and give energy-saving suggestions based on individual user preferences. Using the information gathered, IoES-based SHEMS can detect and respond to (1) the energy consumption pattern, (2) issues with the normal operation of the home, like gas or water leaks and electrical failure (which may enhance energy depletion), and (3) emergencies that compromise the security and integrity of the home (e.g., burglary attempt, flood, fire). After alerting the user, the system displays the most essential details about each service provider (such as their names, addresses, phone numbers, hours of operation, costs, and user references). Energy networks may be optimized with the aid of AI by controlling the energy flow to and from the house, industry, battery, renewable energy source, microgrids, and the main power grids. The outcome is less energy wasted while more people pay attention to energy use. The term DSM is utilized to describe programs and tools aimed at influencing domestic customers to reduce their energy use. DSM involves implementing strategies and programs to actively manage and adjust electricity consumption patterns on the consumer side. This includes initiatives such as load shifting, EE measures, and DE programs that help optimize energy usage and reduce peak demand on the grid. One possible advantage of DSM utilizing DNN is that customers may save money on their monthly power bills by modifying when and how much energy they use. There are essentially four main types of DSM. There are four efficient approaches to DSM and DE: DE is actively managing electricity consumption during high demand or supply constraints, incentivizing consumers to curtail or shift their usage to reduce peak demand on the grid, lower overall electricity costs, and enhance grid stability, EE was maximizing output from a given energy input while minimizing waste and inefficiencies by adopting technologies, practices, and policies that reduce energy consumption, lower utility bills, mitigate environmental impacts, and enhance energy security, virtual power plants (VPP) also called as distributed energy resource (DER) systems that integrate multiple decentralized energy assets into a single, coordinated VPP, optimizing their collective operation and providing grid services such as peak shaving, frequency regulation, and grid balancing, and spinning reserve (SR) is the portion of generating capacity that is online and synchronized with the grid but not currently engaged in producing electricity, acting as a rapid-response backup reserve that can be immediately dispatched to ensure grid stability and reliability during periods of high demand or system disturbances. Remote energy management solution enables monitoring, managing, analyzing, and control multi-location energy operations. Through the Internet, mobile phones, etc., users may actualize prepaid services and remote monitoring of household equipment. The DNN is an extensively utilized technique for predicting time series. RNNs are commonly used in DNNs to predict time series data. A specific type of RNN, known as LSTM networks, is particularly effective in capturing temporal dependencies and patterns in sequential data. This makes them ideal for performing time series prediction tasks. Depending on the weather or period data, the DNN will recognize patterns and understand during which days the smart home electricity demand is the highest or how much energy can be generated in a particular season.

Figure 1:

Figure 2 illustrates the DNN model. The time series data used in this research have been collected from the dataset [32]. The DNN designs all include one or more hidden layers learned via backpropagation and gradient descent. This research finds that when additional hidden layers are added to DNN models, the ensuing abstract features lead to over-fitting issues. In DNN-based electrical load forecasting system for effective resource management, the design of the prediction model will substantially affect DNN performance, which is very problem-dependent. DNN-based energy prediction models adapt to changes in user behavior, appliance usage, and external factors such as weather conditions and occupancy levels through continuous learning, feature engineering, temporal context, dynamic input feeding, ensemble learning, and transfer learning techniques. This research suggests DNN-based load forecasting models using an empirical load database from the demand side. DNN load forecasting models capture energy consumption patterns using preprocessed data types, such as historical load data, weather data, calendar and time features, historical load patterns, exogenous factors, feature engineering, and data splitting. Preprocessing involves cleaning, normalizing, scaling, imputing, interpolating, and encoding the data. The data are split into training, validation, and test sets while preserving temporal order. The models predict future load demand accurately and reliably. Predictive analytics is essential for managing energy consumption in AI-enabled home automation systems. It enables homeowners to forecast energy demand accurately, optimize usage in real-time, and engage in energy-saving practices. By analyzing historical data and user interactions, the system suggests actionable insights and strategies for optimizing energy consumption, leading to greater efficiency, cost savings, and sustainability in the home. Several factors, including the neurons, the number of hidden layers, activation functions in every hidden layer, and the learning algorithm, are crucial for the suggested method’s effectiveness. A DNN is a highly nonlinear function, a trainable approximator of forms x(·): R^N → R^N where N and M are the dimensionality of input and output space. Operationally, DNNs contain input, output, and hidden layers. The input layers receive the DNN’s input vectors y. The neuron in any other layer receives, as its input, the weighted output of a neuron in previous layers. The weight of the DNNs makes up its weight parameters, denoted as θ. For ease, this study considers DNN with scalar output so that x(·): R^N → R. The original outputs of the DNNs are denoted as x(x|θ), neurons in output layers.

Figure 2:

In common regression applications for smart home energy management, the DNN training sets W contain pairs (y (m),t(m)) ∈ W where m = 1,…, |W| denotes the sample indices (for brevity, this association is often signified as n ∈ W). The quantity t(m) is desired or are target outputs. Throughout the training, θ has been updated in stages so that for every input y(m), DNN outputs x(m) are as close as probable to t(m). The supervised learning algorithm goal is to reduce DNN loss functions L(θ). The W denotes that losses are an empirical measure over samples in W. A prevalent selection of the concluding is the mean squared norm of the variances between targets and outputs for every sample in W as (1) Lθ=121W∑m∈Wxym|θ−tm2 L\left( \theta \right) = {1 \over 2}{1 \over {\left| W \right|}}\sum\limits_{m \in W} {{{\left( {x\left( {y\left( m \right)|\theta } \right) - t\left( m \right)} \right)}^2}}

Let us consider that X denotes the forecast energy consumption, Y signifies the input data, comprising historical energy usage, IoT sensor data, weather data, and another appropriate factor, and f(.) signifies activation functions of the DNN: (2) S=argmaxsLDfy,s S = \arg \max \left( s \right)L\left( D \right)f\left( {y,s} \right)

As shown in Eq. (2), L denotes the loss function that measures the variance among the forecasted output of DNNs and the real output. D indicates the training data of usage and energy consumption, weather and environmental data, and other appropriate characteristics. f(y,s) represents the DNN models that map the input information y to the forecasted output x. The DNN is utilized in this equation to learn the connection among the input data and the forecasted energy consumptions or outputs. By training DNNs on a huge database of past energy usage and associated data, networks can learn to create precise forecasts about upcoming energy use based on the input information. Preparing datasets for training DNNs in energy consumption prediction models, you must collect, clean, select relevant features, normalize and scale, aggregate, split for training and testing, generate input-–output sequences, and organize into batches. These steps ensure the data are clean, relevant, and properly formatted for effective model training and evaluation.

Training the DNNs includes multiple passes entitled epochs, each comprising one pass via every sample in W. In stochastic gradient descents (SGDs), with η ≪ 1 as the learning rates, the variable θ is increment once for each sample m ∈ W as, (3) θ←θ−ηxym|θ−tm∇θxym|θ \theta \leftarrow \theta - \eta \left( {x\left( {y\left( m \right)|\theta } \right) - t\left( m \right)} \right){\nabla _\theta }x\left( {y\left( m \right)|\theta } \right)

Feed-forward calculations provide accurate output in energy forecasting. Error quantifies how well a certain training set worked for a given network. The gradient descent reduces the total error in the training set by adjusting the weights in proportion to the inverse of the error derivative, as shown in the following equation: (4) Δωij=−ηθEθij \Delta {\omega _{ij}} = - \eta \left[ {{{\theta E} \over {{\theta _{ij}}}}} \right]

As inferred from Eq. (4), η signifies the learning rates, and E signifies the total error. Finally, the output signals of networks, which is now part of the training information, are compared with target outputs. The variance among the original and target output signifies the error of the output layer neurons.

Figure 3 displays the sequence diagram of the IoE-based smart home energy monitoring system. Using an intuitive web-based interface, it is possible to automate the management of a household’s energy-consuming equipment from anywhere in the world. This method reduces all the manual work that normally goes into maintaining an appliance in the home. The information gathered by the sensors will be uploaded to the cloud through a Wi-Fi controller. No human intervention is required since all data from the cloud will be shown on the website, and all decisions will be made in the cloud. Data collected by edge devices, such as sensor readings and user profiles, may be warehoused in a highly scalable storage server. It can process the Big Data created by homes and expand to accommodate other neighborhoods. A fast and extensible database is necessary to monitor user details and user-home and home-device connections. An operational database was selected, one that can be operated on a scalable storage server. Web services developed using JavaScript allow the client application to communicate with the operational database. These functions move information to and from the database and deliver it back to the user. The client app uses web services for authentication, device monitoring and control, properties observing, device viewing, monthly bill viewing and payment, and graph viewing at the user’s level. The HTTPS protocol is used to develop online services to secure data transmission. The suggested AI-IoES enhances the user experience, EE, and accuracy ratio and reduces energy consumption compared with other existing methodologies

Figure 3:

This paper recommends AI-IoES for efficient smart home energy management. The data have been taken from the Open Smart Home IoT//Energy Dataset for analyzing the energy consumption of home appliances. This file consists of the readings with a period of 1 min of home devices in kWs from an intelligent meters and weather condition of that specific region. Intelligent meters provide real-time data that can help to analyze energy consumption trends, detect anomalies, identify peak demand periods, and offer timely insights for optimizing energy usage, scheduling maintenance, and implementing efficiency measures for smart homes and buildings. This information can be used to optimize energy use and reduce waste. In addition, a smart home energy database that records miscellaneous energy consumption data is publicly offered. The electricity consumption of each device and each radiator in 255 households is tracked separately. Furthermore, anonymized survey information and metadata, such as tenant demographics, appliance characteristics, self-reported energy awareness and attitudes, and room building, supplement the data gathered by sensors. Finally, the performance of the suggested AI-IoES is analyzed based on metrics such as user experience ratio, energy efficiency ratio (EER), accuracy ratio, and energy consumption ratio compared with existing models.

(i)

Energy efficiency ratio

To realize a sustainable smart society, IoT has embraced energy-efficient processes (hardware and software) via IoE to help lessen the load on the environment by decreasing the amount of power now used by applications and services and the carbon emissions such services and devices produce. This reduces the probability of device malfunction and keeps production continuing robust. The EER affects the likelihood of a device malfunctioning or failing. Operating devices efficiently reduce stress on components and lower the chances of overheating and wear and tear. This promotes longevity and reliability, minimizing downtime for repairs and maintenance and enhancing the sustainability of production processes. Furthermore, monitoring energy usage trends means developing realistic targets for future energy management using AI and IoE services. Services for automating energy saving in the house were offered in this research. Structured smart IoE platform for smart households will deliver homes designed to be simple indoor homes and smart, energy-effective living spaces for healthy lifestyles as a more in-depth study on smart models becomes available. EE is calculated as the percentage of the input energy ending up in the desired output of targeted home appliances and has the efficiency. To offer an energy service while reducing energy use is to practice EE. For example, compared with traditional incandescent lights, energy-saving LED bulbs use just a quarter to half as much power while producing the same amount of light. Generally, energy inputs comprise energy counterparts of machinery depreciation, energy to heat the digestor, and electricity to operate the apparatus. Figure 4 illustrates the EER.

Figure 4:

(ii)

User experience ratio

Evaluating usability is a crucial part of the user interface design process since it directly affects the quality of the end user’s interaction with the interface. IoE insights gained via sensory intervention may enrich the design process and ultimately lead to greater quality and user experience. User satisfaction with smart home technologies was measured against three predefined criteria: how helpful, how easy to use, and how accessible they were. By considering how each variable may be used to improve the user experience, this study created a framework for IoE-based smart home services. Users’ perceptions and general comfort with technological tools are highlighted in the technology dimension, while human–computer interaction and user experience are highlighted in the space dimension. The convergence of intelligent computers and architecture results in novel interactive settings. In this always-online setting, residents may engage with others who share their interests and lifestyles, and a wide range of residential services customized to their specific needs are made available. The suggested framework will aid designers, architects, engineers, and academics in searching for a more holistic, integrative understanding of smart homes and their development. Users may set energy use and expenditure goals to create and keep control of their monthly household budgets and prevent unpleasant financial unanticipated events. The smart speaker based on AI can have simple talks with people, satisfy users’ requirements for communication with the appliances, follow the consumer’s instructions to do tasks, shorten the consumer’s operation step, and enhance user experiences. The user experience has been calculated using inputs such as human–machine interaction, adoption rate, system usability scale, satisfaction, and task completion rate. Figure 5 illustrates the user experience ratio.

Figure 5:

(iii)

Energy consumption ratio

In most cases, smart outlet sensors can accurately estimate the current energy consumption of electrical devices. However, these devices are costly and often need specialized modes of connection. Therefore, numerous factors are considered while setting up a microgrid, including energy consumption of different household appliances, solar irradiation, and overall load. Thus, AI regulates energy use, reduces it during peak hours, discovers issues in advance, and sends alerts. In addition, AI can compress and analyze enormous datasets, which may be used for energy monitoring and interpretation. AI and IoT device controls energy usage and reduces it during peak hours. Operating a commercial house’s cooling system utilizing AI and IoT data can help to reduce energy consumption using an advanced monitoring system. Input factors like electrical efficiency, sunlight, and energy manage electricity and wind and use regulated and uncontrollable loads. Figure 6 signifies the energy consumption rate.

Figure 6:

(iv)

Accuracy ratio of the DNN model

The accuracy of measuring power was found to be strongly reliant on the variable nature of the load, with all devices demonstrating major errors as the load changed. As a result, a comprehensive set of simulations is carried out to test the proposed method’s accuracy and effectiveness using actual datasets. Consequently, the AI-IoES that was developed is capable of attaining a greater accuracy when it comes to forecasting the energy activities that take place in smart homes using DNN. In addition, the system is evaluated to ensure that it accurately detects objects. Nevertheless, other significant aspects are considered for a fully AI autonomous system, such as classifying energy activities daily and night and the interaction between humans and appliances. Similarly, the smart home user can arrange the operating time of appliances so that they may execute and run the appliances according to a schedule that uses as much energy as possible. Based on Eq. (3), the accuracy ratio of the DNN model has been computed. The prediction accuracy for energy consumption was improved through the proposed AI-IoES. The input for this analysis is energy usage per hour, smart home EE performance, and remote control. Figure 7 exhibits the accuracy ratio.

Figure 7:

The simulation results were obtained from the proposed AI-IoES model. The simulation results show that the suggested AI-IoES model increases EE and user experience while reducing energy consumption compared with existing models. The numerical analysis reveals that the AI-IoES model outperforms existing models in terms of EE, user experience, accuracy, and consumption ratios. The proposed AI-IoES model is designed to manage smart home energy efficiently using AI and IoE services. The data used for the analysis are taken from the Open Smart Home IoT//Energy Dataset, which consists of energy consumption data gathered using IoT sensors. The suggested model is evaluated using metrics such as user experience ratio, EER, accuracy ratio, and energy consumption ratio. The research article highlights the importance of achieving EE and reducing energy consumption in smart homes. The proposed AI-IoES model offers an energy-saving solution that automates energy-saving practices in the house. This model is expected to provide a more energy-efficient and sustainable smart society by reducing the amount of power used by applications and services and the carbon emissions such services and devices produce. In conclusion, the proposed AI-IoES model shows promising results in managing smart home energy efficiently. The model is expected to provide a more energy-efficient and sustainable smart society by automating energy-saving practices in the house. The research article contributes to developing energy-efficient smart homes and offers significant implications for the industry.

This article presents the AI-IoES for effective smart home energy management. The suggested system forecasts the 24-hr load pattern a day ahead using date, weather, and previous power consumption data via a DNN training and forecasting phase. Smart micro grid enables energy management services like prediction, monitoring, scheduling, forecasting, and decision-making reinforced by AI and IoT smart sensors to make consumer electricity use more efficient. The results show that the DNN model’s automated feature extraction capabilities contribute significantly to the model’s high level of accuracy. In addition, its strong generalization allows it to outperform the conventional model-based approach computationally. The experimental outcome demonstrates that the suggested AI-IoES system increases the user experience by 98.9%, EER by 97.8%, and accuracy ratio by 97.2%, and reduces energy consumption by 19.2% compared with other existing methods. However, since collecting demand-side load information is comparatively low, the load forecasting data are limited. If the load size information is limited, specific problems with using predicting model might exist. Therefore, future studies will concentrate on Quality of Experience (QoE) assessment from the consumer’s viewpoint in the IoE services field.