An Overview of Recent AI Applications in Combined Heat and Power Systems

Abstract

Combined heat and power (CHP) systems are among the important components for enhancing energy efficiency and sustainability by simultaneously generating electricity and useful thermal energy, reducing waste and costs. Consequently, the effective control of these systems is considered important. To that end, this paper provides a comprehensive review of the intelligent methodologies applied to CHP systems, emphasizing their prevalence in the USA and Europe through statistical insights. It outlines the mathematical foundations of CHP systems, analyzing the advancements in intelligent control methods for optimal planning, economic dispatch, and cost minimization. Artificial Intelligence (AI) models, such as Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Random Forest, are described and applied to a simulated CHP system. The Key Performance Indicators (KPIs) derived from these models demonstrate their efficacy for optimizing CHP performance. This paper also highlights the impact of AI-driven models for enhancing CHP system efficiency, while identifying the challenges in AI-CHP integration and envisioning CHP systems as important components of future sustainable energy systems.

1. Introduction

1.1. Background

The technology of CHP systems, also referred to as cogeneration, marks a significant advancement in energy technology by integrating the synchronized production of electricity and thermal energy, departing starkly from conventional power generation models and enhancing overall energy efficiency [1]. Unlike traditional systems, which waste a considerable portion of energy as heat, CHP systems efficiently utilize this byproduct for practical applications, such as space heating, industrial processes, or district heating, resulting in a reduction in wastage and an optimization of energy use [2]. Moreover, CHP systems have a substantial impact on environmental sustainability by reducing greenhouse gas emissions through the efficient use of fuel inputs and the generation of both electrical power and useful heat, thus making them a compelling option for industries, commercial buildings, and institutions and aiming to enhance their energy resilience, lower operational costs, and contribute to a greener energy sector. With the global focus shifting towards cleaner and more efficient energy solutions, CHP technology stands out as a versatile and impactful approach for meeting evolving demands.

This global shift towards clean and efficient energy, including CHP technology, can also be seen in the published research in recent decades on CHP systems in the Scopus database, as shown in Figure 1.

Journal papers represent the dominant category, peaking at 679 publications in 2023, before slightly declining to 658 in 2024 and to 27 in early 2025. Conference papers reached a high of 323 in 2017, with a consistent decline thereafter, reaching only 92 in 2024 and just 1 in early 2025. Book chapters had a peak of 29 in 2017, and have maintained around 20 publications annually, except in 2024 and 2025. Books are relatively rare, with five such entries in 2014 and a few other instances across the years, dropping to zero in 2024 and 2025. Editorials have an occasional presence, peaking at four in 2016 and appearing intermittently in subsequent years.

Regarding CHP plant utilization around the world, within the United States (U.S.), CHP technology has emerged as a solution for advancing energy efficiency and resilience across diverse sectors, including industries, healthcare facilities, universities, and commercial complexes. Widely embraced, CHP systems in the U.S. present with adaptability for addressing the simultaneous need for electricity and thermal energy. This technology has garnered traction due to supportive incentives and grants, encouraging its widespread adoption. Beyond its economic benefits, the concentration on energy resilience has driven the integration of CHP into infrastructure, facilitating on-site power generation and bolstering grid reliability. Continuous technological innovations and the integration of smart grid features have further augmented the performance of CHP systems, thereby leading to a more sustainable and secure energy sector. These advancements, and an overview of CHP generation and its technological evolution in the USA, are depicted in Figure 2.

In Figure 2, the statistics present the annual generation of CHP technology and systems in the United States, measured in terawatt-hours ([TWh]), over a span of twelve years. The numbers indicate a fluctuating trend, with an overall increase from 254 [TWh] in the initial year to 360 [TWh] in the final year. Noteworthy is the general upward trajectory, suggesting the growing contribution of CHP systems to the overall energy landscape in the USA. The peak year, with a generation of 374 [TWh], indicates a significant period of heightened CHP output, possibly driven by its increased adoption and technological advancements.

The wide adoption of CHP systems across diverse sectors, including industrial processes, district heating, and commercial buildings, exemplifies Europe’s commitment to simultaneously generating electricity and capturing waste heat. This integrated approach not only enhances overall efficiency but also aligns with ambitious climate goals by reducing greenhouse gas emissions. The prevalence of district heating systems, powered by CHP plants, underscores a holistic strategy towards sustainable urban development. The CHP generation in European countries is presented in Figure 3.

1.2. Review Phases and Search Algorithm

The literature reviewed in this paper was selected and reviewed based on the process shown in Figure 4.

This paper conducted three overall phases of literature search/selection, its review, and post-review. In the first phase (literature selection), firstly, the keywords were defined to start the searching process. Then, a digital library of the collected literature was created. In the next step, the proposed collected sources were filtered based on if they are peer-reviewed and high-quality. If the sources met both proposed terms, they were included in the analytics; otherwise, they were excluded. In the next phase, the filtered literature was analyzed in the fields of CHP systems and their recent advances, as well as their integration with AI/ML algorithms. Finally, three known and regular AI models are fully described, and a simulation is conducted comparing their performance in a CHP system. Lastly, the final notes and future perspectives are investigated.

1.3. Contributions

The contributions of this paper are summarized as follows:

(1)

This is the first updated paper that reviews recent AI/ML models in CHP systems;

(2)

It provides modelling of and recent advances in CHP systems;

(3)

We conduct a simulation integrating three AI models in CHP systems and compare their performance.

1.4. Paper Structure

The paper is organized as follows: Section 2 and Section 3 present the general modelling of CHP systems and the recent literature in this field. Then, the AI/ML models, their review, and simulation are investigated in Section 4 and Section 5. Finally, Section 6 draws the conclusions.

2. CHP General Formulation

Relying upon the structural details, a CHP system includes two units regarding electricity and heat generation, as conceptualized in Figure 5.

As illustrated in Figure 5, the main unit typically serves as the primary power generator, concurrently producing both electricity and heat. Supplementary components such as heat units, heat exchangers, or gas boilers are manipulated for heat generation to water and space heating purposes. Given its decentralized nature, cogeneration often encounters surplus and deficit energy situations to meet demand since it does not rely on a centrally generated grid. To enhance reliability and conserve energy, the mechanisms of electricity storage and heat storage can be combined with cogeneration systems. The framework of a cogeneration setup comprises a main energy source/fuel, a power plant, and a heat recovery system. Additionally, multiple cogeneration systems can incorporate electricity storage and heat storage functionalities. Among RESs, water, geothermal energy, and solar power are predominant energy forms, while biomass, GH_2, and biofuels are also classified as other types of RESs [5].

Technologies for CHP systems can be considered in two categories based on project efficiency: technologies that use thermodynamic changes to produce energy and technologies that do not [6]. The first technique uses internal or external heat engines to produce electricity from fuel. Steam engines, Stirling engines, gas engines, diesel engines, and organic Rankine cycles are a few examples. The second type of technology is energy-generating technology, which includes fuel cells and solar power generation. The heat recovery equipment found in the majority of generating systems can supply power, high-temperature water, space heating, or cooling to a combined cycle. To improve system efficiency without requiring more fuel, heat recovery unit (HRU) devices are in charge of collecting thermal energy from the primary carrier [7]. Depending on the technology, the drive can be utilized in conjunction with the main engine or independently, and it can be turned on or off. A heat exchanger (TEG), which operates on the principle of heat transfer, is one of the technologies used in HRUs. Heat exchangers are available in a wide variety of materials and forms, including micro, tube, plate, shell, and track types. The heat recovery steam generator (HRSG), which is operated with steam/heat generation exhaust gases, is another device that resembles a TEG. HRSGs are generally low in sound and operate at temperatures between [370 and 540 °C], whereas they require routine maintenance to keep their high blast level.

Utilizing a TEG to transform the heat into electricity is an additional method of recovering heat from the generating system. In order to create a voltage, the TEG exploits the temperature differential between the anode and cathode to permit electrons to move from the n-type (negative) semiconductor to the p-type (positive) semiconductor. The operating temperature range and semiconductor materials used in p-pole and n-pole TEG technologies vary. In many instances, cogeneration systems substitute the need for natural gas by producing heat instead of heating [8,9,10,11,12]. Cooling systems can employ a wide range of technologies, such as refrigerant cooling and induction, and their associated techniques. The two operating fluids used in suction cooling are absorption and cooling. An induction heater produces cold air or water by condensing refrigerant vapor through the use of solid material. The four primary components of a basic induction heating cycle are the absorber, the condenser, the evaporator, and the evaporator. Water and ammonia are commonly used as working pairs of absorbent and refrigerant in absorption cooling (H₂O/NH₃), as well as lithium bromide–water (LiBr/H₂O).

A CHP system can be generally formulated as in [13], in an objective function of f₁, and the parameters and variables are defined in

Table 1. The parameters and variables of the CHP system [13].

Parameters	Insights	Unit
$[c_i^p,c_i^c,c_i^h]$	CHP operational cost, heat/power-only units	[$]
$[e_i^p,e_i^c,e_i^h]$	CHP emissions, heat/power-only units	[Kg]
$[f_1,f_2]$	Objective functions of the system	[$, Kg]
$[A_i^{hc},B_i^{hc},C_i^{hc}]$	Heat-only cost coefficients	[${\displaystyle \frac{\mathrm{\$}}{\mathrm{M}{\mathrm{W}}^2}},{\displaystyle \frac{\mathrm{\$}}{\mathrm{M}\mathrm{W}}},\mathrm{\$}$]
$[A_i^{pc},B_i^{pc},C_i^{pc},D_i^{pc},E_i^{pc}]$	Power-only cost coefficients	[${\displaystyle \frac{\mathrm{\$}}{\mathrm{M}{\mathrm{W}}^2}},{\displaystyle \frac{\mathrm{\$}}{\mathrm{M}\mathrm{W}}},\mathrm{\$},\mathrm{\$},{\displaystyle \frac{\mathrm{R}\mathrm{a}\mathrm{d}}{\mathrm{M}\mathrm{W}}}$]
$[A_i^{pe},B_i^{pe},C_i^{pe},D_i^{pe},E_i^{pe}]$	Power-only emission coefficients	[${\displaystyle \frac{\mathrm{K}\mathrm{g}}{\mathrm{M}{\mathrm{W}}^2}},{\displaystyle \frac{\mathrm{K}\mathrm{g}}{\mathrm{M}\mathrm{W}}},\mathrm{K}\mathrm{g},\mathrm{K}\mathrm{g},{\displaystyle \frac{1}{\mathrm{M}\mathrm{W}}}$]
$\left[A_i^{cc},B_i^{cc},C_i^{cc},D_i^{cc},E_i^{cc},F_i^{cc}\right]$	CHP cost coefficients	[${\displaystyle \frac{\mathrm{\$}}{\mathrm{M}{\mathrm{W}}^2}},{\displaystyle \frac{\mathrm{\$}}{\mathrm{M}\mathrm{W}}},\mathrm{\$},{\displaystyle \frac{\mathrm{\$}}{\mathrm{M}\mathrm{W}}},{\displaystyle \frac{\mathrm{\$}}{\mathrm{M}{\mathrm{W}}^2}}$]
$[A_i^{ce},A_i^{he}]$	Emission coefficients for CHP and heat only	[${\displaystyle \frac{\mathrm{K}\mathrm{g}}{\mathrm{M}\mathrm{W}}},{\displaystyle \frac{\mathrm{K}\mathrm{g}}{\mathrm{M}\mathrm{W}}}$]
$A_{ij}^l$	Power loss coefficient between unit i and j	[${\displaystyle \frac{1}{\mathrm{M}\mathrm{W}}}$]
$B_i^l$	Power loss coefficient of unit i	–
$C^l$	Constant power loss	[MW]
pl	System total power loss	[MW]
$[H^d,P^d]$	HP demand	[MW]
$[p_i^{\mathrm{m}\mathrm{i}\mathrm{n}},p_i^{\mathrm{m}\mathrm{a}\mathrm{x}}]$	Power-only minimum/maximum power output	[MW]
$\left[H_i^{\mathrm{m}\mathrm{i}\mathrm{n}},H_i^{\mathrm{m}\mathrm{a}\mathrm{x}}\right]$	Heat-only minimum/maximum power output	[MW]
$[h_i^c,h_i^h]$	CHP heat output and heat-only unit	[MW]
$[p_i^c,p_i^p]$	CHP power output and power-only unit	[MW]
$[h_i^{\mathrm{m}\mathrm{i}\mathrm{n}}(p_i^c),h_i^{\mathrm{m}\mathrm{a}\mathrm{x}}(p_i^c)]$	Minimum/maximum CHP unit heat output	[MW]

[13].

$$min{f}_1=\sum c_i^p\left(p_i^p\right)+\sum c_i^c\left(h_i^c,p_i^c\right)+\sum c_i^h\left(h_i^h\right)$$

(1)

where the generating units of $c_i^p,c_i^c,{\mathrm{a}\mathrm{n}\mathrm{d} c}_i^h$ are considered quadratic functions related to heat/power [14,15,16,17,18,19,20]:

$$\left\{\begin{array}{c}{c}_i^p=A_i^{pc}(p_i^p{)}^2+B_i^{pc}{p}_i^p+c_i^{pc} \hfill \\ c_i^c=A_i^{cc}(p_i^c{)}^2+B_i^{cc}{p}_i^c+D_i^{cc}(h_i^c{)}^2+E_i^{cc}{h}_i^c+F_i^{cc}{h}_i^c{p}_i^c\hfill \\ c_i^h=A_i^{hc}(p_i^{hc}{)}^2+B_i^{hc}{h}_i^h+c_i^{hc} \hfill \end{array}\right.$$

(2)

Accordingly, CHP can be considered another form of the cost function related to units that have only power, which includes a rectified sinusoidal term, as follows [21,22,23,24,25,26]:

$$c_i^p=A_i^{sc}{\left(P_i^p\right)}^2+B_i^{sc}{P}_i^p+c_i^{sc}+\left|D_i^{sc}\sin \left(E_i^{sc}\left(P_i^{\min }-P_i^p\right)\right)\right|$$

(3)

Considering the influence of the valve profile, CHP provided nonconvexity in the model objective/cost function. If the effect of the valve signal is ignored, the cost functions will be in types of convex quadratic functions. For the total emissions, it is modelled as [27,28,29,30,31,32,33]

$$\mathit{min}\{f_2\}=\sum e_i^p\left(p_i^p\right)+\sum e_i^c(h_i^c,p_i^c)+\sum e_i^h\left(h_i^h\right)$$

(4)

where $e_i^p$ is defined as the emissions of each unit ($e_i^p,e_i^c,e_i^h$):

$$\left\{\begin{array}{c}{e}_i^p=A_i^{pe}(p_i^p{)}^2+B_i^{pe}{p}_i^p+C_i^{pe}+D_i^{pec}{e}^{E_i^{pc}{p}_i^p}\hfill \\ e_i^c=A_i^{ce}{p}_i^c \hfill \\ e_i^h=A_i^{he}{h}_i^h \hfill \end{array}\right.$$

(5)

Subjected to

$$\left\{\begin{array}{c}\sum p_i^p+\sum p_i^c=P^d+p^l\hfill \\ \sum h_i^c+\sum h_i^h=H^d \hfill \end{array}\right.$$

(6)

In which $p^l$ is considered as power loss [24,28]:

$$p^l=\sum _i\sum _j{p}_i^p{A}_{ij}^l{p}_j^p+\sum _i\sum _j{p}_i^p{A}_{ij}^l{p}_j^c+\sum _i\sum _j{p}_i^c{A}_{ij}^l{p}_j^c+\sum _i{B}_i^l{p}_j^p+\sum _i{B}_i^l{p}_i^c+C^l$$

(7)

This is constrained to the physical limitations considered to the generating units as

$$\left\{\begin{array}{c}\begin{array}{c}{p}_i^{\min }\le p_i^p\le p_i^{\max }\hfill \\ p_i^{\min }\left(h_i^c\right)\le p_i^c\le p_i^{\max }\left(h_i^c\right)\hfill \end{array}\hfill \\ h_i^{\min }\left(p_i^c\right)\le h_i^c\le h_i^{\max }\left(p_i^c\right)\hfill \\ H_i^{\min }\le h_i^c\le H_i^{\max }\hfill \end{array}\right.$$

(8)

Integrating these limits leads to a two-dimensional region with cogeneration unit feasibility as

$$\{\left(h_i^c,P_i^c\right):p_i^{\min }\left(h_i^c\right)\le p_i^c\le p_i^{\max }\left(h_i^c\right),\hspace{0.17em}{p}_i^{\min }\left(h_i^c\right)\le p_i^c\le p_i^{\max }\left(h_i^c\right)\}$$

(9)

This leads to a source and origination of model nonconvexity.

Relying on the described context and concepts of CHP, the recent advances of the CHP systems and intelligent control methodologies will be reviewed next in Section 2 and Section 3.

3. Advancements and Integrations in CHP Systems

This section analyzes significant works that have been conducted regarding recent developments in cogeneration technology and associated systems. The authors of [34] examined how effective heat removal systems can enhance charger performance by addressing the subsystems’ thermal problems. A novel microchannel heat removal system with microchannel cooling components is suggested by the research. The test results demonstrate the system’s flexibility to various operating environments, which enhances heat exchange efficiency and lowers energy usage. Additionally, the capacity for heat transfer is greatly increased when Al₂O₃ nanoparticles are incorporated into the working environment. This study offers important information for the creation of charger heat removal devices that are energy efficient. In [35], which describes a novel approach combining solid oxide fuel cells (SOFCs), biomass CO₂ gasification, and an ammonia Rankine cycle, biomass is examined as a sustainable fuel source. High net electricity efficiency, low CO₂ emissions, and economic viability are demonstrated by a thorough evaluation. Significant carbon capture rates and low carbon emissions are noteworthy accomplishments that demonstrate this system’s ability to reach carbon neutrality. The economic analysis demonstrates its contribution to sustainable and profitable energy conversion and provides additional evidence of its viability.

In [36], a hybrid CHP model with a fuel cell performing as a mover was introduced. A Honey Badger optimizer (HBO) is manifested to develop system quality and enhance its performance. The results demonstrate that the combination of high relative humidity, high inlet gas pressure, and low operating temperature leads to better electrical efficiency and a reduction in the emissions of greenhouse gases. The study demonstrates how considerable it is to optimize system parameters in CCHP design in order to boost efficiency and lessen environmental effects. The work [37] proposes a stochastic planning approach with energy concentration risk limitations that considers load demand and RES generation variability. The method balances risk avoidance with operational and emission costs by utilizing the conditional value at risk (CVaR) mechanism. The study includes innovative algorithms for uncertainty correction, which demonstrate the flexibility of power center management. The simulation results show significant reductions in operating costs and emissions.

A thorough analysis of the use of several energy harvesters in conjunction with electromagnets was presented by the authors of [38]. It goes over the fundamentals of power inverters and solar systems, the variety of thermoelectric generators, their latest uses, and performance analysis. That paper, which presents current advancements and prospective uses of PV/TE systems in solar and solar collectors, is a valuable resource for scholars investigating hybrid solar systems. As mentioned in [39], weather conditions pose a hazard to power systems and call for robust solutions. A method for multicarrier machines to decrease the area in production-oriented research was mentioned in the study. The work focuses on heat-energy–hydrogen (H₂) MGs using cross-sectional lines. Using mixed-integer linear programming, vulnerability is assessed using expected load not supplied (ELNS). The results show a 52% improvement in ELNS.

Smart buildings’ impact on energy, heating, and cooling (HVAC) systems is investigated in [40], which also presents an optimal planning scheme of distributed energy resources (DERs). Self-healing figures of merit are used in a four-step optimization framework to assess a multicarrier system’s efficiency. In examining the temperature properties of a stack of proton-exchange membrane fuel cells (PEMFCs) in a CHP system, ref. [41] examined the relationship between PEMFC temperature, coolant flow, and current density in terms of performance. For every acceptable refrigerant flow rate, an equation illustrating the current density area was created. The findings point to a technique for utilizing PEMFCs in CCHP systems to reach the proper operating temperatures. Three-phase optimal power flow (OPF) controllers for power-resilient grids are presented with the purpose of providing a solution for voltage disruptions in the inverter-based generation distribution network. This method outperforms other approaches like voltage–var control (VVC) and power–volt–watt control (VWC) by determining the best placement for a smart inverter to control the voltage of impacted nodes.

In the field of carbon neutrality in multi-regional integrated energy systems (MIESs), ref. [42] introduced a method to improve the deep policy pipeline (MADDPG). Using a distributed learning and training framework, this approach includes focused methods for effective communication and coordination between clients. Experimental results on multi-domain datasets demonstrate the algorithm and its ability to reduce system operating costs, reduce carbon emissions, and promote multi-system integration. The proposed work investigated the optimal arrangement of multisite integrated energy systems (MIESs) based on unscheduled supply and user demand. The Enhanced Deep Policy Group (MADDPG) method is improved upon, and the issue is solved as a Markov Decision Process (MDP) with global carbon constraints to reduce system operating costs, reduce carbon emissions, and accelerate carbon emissions. The potential of H₂ integrated energy systems (HIESs) is considered in [43] through multiple microgrids (MGs) and H₂ exchange between them. A multi-level and time-scale energy management solution is proposed that takes into account energy–heat–H₂ balances and uncertainties. With predefined scheduling, MPC-based in-cycle delivery, and timing coordination, HIESs have flexibility and more efficient use of energy. An evaluation using the IEEE 33 bus test network confirms the efficiency and economic benefits of the proposed solution compared to reference solutions. Focusing on sustainability and increasing energy density, ref. [44] represents a multigenerational system that combines biomass and thermal energy resources. An improved thermal drive system was combined with a biomass fuel system using a supercritical CO₂ process to provide an efficient system that combines heating, cooling, and power generation. The study uses the multi-objective gray wolf optimization (GWO) algorithm to optimize performance considering power generation as well as heat generation, performance efficiency, and system efficiency. Extensive observations and optimization show the feasibility and potential benefits of the proposed multigenerational system. Ref. [45] reviews recent trend research and made progress on green hydrogen (GH₂) energy systems, providing an overview. Aspects covered include H₂ production processes, energy systems for carbon-free H₂ production, the use of H₂ in fuel cells, and integration into metal hydride energy storage.

Addressing the battery temperature operating system, the authors of [46] described a battery thermal management system (BTMS) that is enhanced with phase-change materials (PCMs) as well as thermoelectric coolers (TECs). The study investigates BTMS performance in different scenarios using a numerical model. The results suggest that adjusting the TEC inlet flow, fin length, and thickness reduces the peak temperature and PCM liquid fraction. Optimal settings provide insight into effective battery thermal management. Focusing on short infrared (SWIR) detectors operating below 210 K, a cooling design [47] combining SWIR and a multistage TEC is implemented. The design includes microchannel heat sinks (MHSs) for improved detection and responsiveness. A finite element method analysis evaluates different cooling solutions, showing that the integration of SWIR, TEC, and MHS systems provide the overall performance and minimum volume.

In [48], a two-layer scheduling optimization model is carried out. The model aims to reduce system operating costs and carbon dioxide emissions by considering joint demand response to alleviate peak capacity pressure. The traditional NSGA-II algorithm was improved for solution accuracy and efficiency. The simulation results show that the optimization model causes the overall energy cost, CO₂ emissions, and peak pressure to be reduced in the electric network. In [49], a CO₂ recovery system based on a molten carbonate fuel cell (MCFC) was investigated for the production of efficient energy and cooling originating from biomass. Using machine learning (ML) techniques, the study optimized key design variables for efficiency maximization while considering the minimization of emissions and the levelized cost of electricity production (LCOP).

One of the main obstacles to integrating RESs into the electrical grid is ensuring the safe functioning of thermal power plants [50]. Under all operating conditions, the presented acceleration timing design, which is based on Active Disturbance Cancellation Control (ADRC), is made to guarantee safe operation. The stability analysis of coordinated control systems, linear coupling techniques, parameter selection, and control complexity analysis are all areas of research. Potential industrial applications of the suggested design are demonstrated by simulations that demonstrate enhanced power tracking and interference cancellation performance when compared to traditional ADRC and conventional proportional–integral control techniques. Focusing on GH₂ production from biomass and geothermal RESs [51], a cogeneration framework was presented that included a gas turbine with a geothermally supported Rankine unit. The study evaluates technical, environmental, and economic aspects, taking into account electricity and GH₂ production, thermal efficiencies and exergetic efficiencies, environmental damages, emission indices, LCOP, and total system costs. A bi-objective optimization approach is used, resulting in an optimized framework that balances efficiency and production costs and shows improvements over baseline conditions.

A review that was published in [52] covered carbon reduction tactics for industrial port areas (IPAs) all over the world and emphasized the potential application of GH₂ as an RES. The study examines 74 initiatives in 36 IPAs, mostly in Europe, with an emphasis on carbon reduction measures based on H₂. The review looks at the potential and difficulties of using GH₂ in green energy import/export as well as the challenging reduction in CO₂ emissions. It also emphasizes the significance of onshore power systems (OPSs) in port pollution reduction. For additional study on the creation of IPA carbon reduction plans, the technical and financial data offered is helpful. Addressing melting events caused by inappropriate melt initiation methods in SHP plants [53], the study used an inspection strategy based on image grayscale detection. The control strategy uses image gray recognition equipment (IGRE) and is evaluated by experiments.

In another study, a methodology to estimate the thermal storage capacity of hot water cylinders dependent on the different types of equipment usage behavior is presented, and different temperature control methods are compared with a focus on the use of electric hot water cylinders (HWCs) in DSM [54]. The work forecasts domestic hot water (DHW) demand using stochastic methods and creates an intelligent controller that, when compared to fixed-point and ripple controllers, shows reduced unopposed home water demand and larger free storage. The findings demonstrate the potential use of HWC for heat storage at the electrical system’s peak load. In [55], a parallel air-cooled system was developed for electric vehicle batteries under different operating conditions. The system and its performance are investigated with different types of single current, and a temperature difference control strategy of the battery cells is proposed.

The study [56] used a predictive emission monitoring system dataset to estimate the emissions of nitrogen oxide (NOx) and carbon monoxide (CO) from a gas turbine. A variety of regression models and four feature-generating techniques are assessed. A novel modelling pipeline that incorporates hyperparameter adjustment and feature engineering is suggested. Greater potential for CO and NOx emission prediction is demonstrated by the Deep Forest Regression (DFR) model, which also emphasizes the influence of feature design and hyperparameter tweaking on prediction performance. In coal-fired cogeneration plants, power system flexibility modifications, such as high back pressure (HBP) or low-pressure turbine zero power (LZPO), are anticipated [57]. Flexibility, energy, and efficiency are considered in the comparison between HBP and LZPO renovation. The results show that LZPO renovation has higher potential to have improvement in RES housing and energy efficiency.

An economic operational model of low CO₂ was utilized in the study [58] to address the challenges of high operational expenses of integrated energy systems (IES) and the insufficient use of RESs. The model incorporates an electricity-to-gas (P2G) system, a carbon capture system (CCS), and a DSM flexible load. Supply and demand planning is the focus of this study, which was conducted by MATLAB/YALMIP, GUROBI, and IPOPT. The outcomes demonstrate lower daily operational and transaction costs, underscoring the suggested model’s potential for low-carbon economic activity. The study [59] predicted a CHP production-considered dynamic fluidized bed combustion plants model with the gas and steam sides. The model is validated against operational data and applied to analyze the plant internal dynamics. Control strategies, including boiler monitoring, hybrid control, and turbine bypass, are evaluated for their ability to vary load and decouple electricity and heat generation. The study emphasizes the importance of modelling both the gas and water sides to accurately represent the dynamics of fluidized bed combustion devices.

The marine industry’s increasing need to cut emissions has spurred research into the sustainable fuels’ ability to convert energy efficiently. Long-distance transportation has shown interest in solid oxide fuel cells (SOFCs) due to their high efficiency, low emissions, and fuel flexibility. Heat integration is frequently disregarded, even though SOFCs can suit ships’ thermal needs. Similarly, ref. [60] demonstrated the methane-, methanol-, diesel-, ammonia-, or GH₂-fueled 100 kW SOFC system’s electrical and thermal efficiency. To enhance heat recovery, cathodic exhaust gas recirculation, or COGR, was researched. According to the findings, methane has the highest net electrical efficiency of any fuel, and COGR promises higher thermal efficiency. The limited flexibility of cogeneration systems is mentioned in [61] due to limitations in thermal–electrical coupling. To improve flexibility, four technologies have been integrated into the cogeneration system—electric boiler, heating energy storage, electric energy storage, and bypass compensation. Operational optimization is planned, taking into account RES restrictions and load fines. The study uses a 350 MW cogeneration system to demonstrate that the integration of bypass compensation technology significantly increases the maximum H2E ratio and power control capacity. However, there is a trade-off between the technologies, as the electric boiler and bypass compensation reduce coal consumption, while the storage of electrical and heating energy reduces the load.

The research conducted by [62] explained the addition of molten salt TES to a cogeneration facility to boost flexibility during peak loads, considering the rise of RES in the power market. A 350 MW supercritical cogeneration facility is simulated, and a new cascade heating steam removal mechanism is suggested. The outcomes demonstrate how well TES works to account for variations in load across cycles of charge and discharge. In comparison to conventional cogeneration facilities, the suggested method increases thermal and energy efficiency and guarantees optimum economic viability at a low load. In [63], an integration approach for a solar cogeneration system is provided. Exhaust gas steam is replaced with heated steam that is preheated by solar heat. The proposed integration strategy (IV) is compared with three others using solar energy conversion, annual efficiency, and levelized energy analyses. The IV integration strategy achieves higher efficiency between solar heat and electricity, annual efficiency between solar energy and electricity, and a lower LCOP.

The research conducted in [64] manipulated the critical significance that microgrids (MGs) play in increasing the effectiveness of energy networks. It provides a thorough examination of a microgrid’s combined heat and power (CCHP) system, taking into account various resources such as energy storage, RES, and Power-to-Gas (P2G) technology. MOO technology, which aims to lower CO₂ emissions and running costs, is applied. The study addresses carrier interaction uncertainty and shows that the presented model is capable of lowering operating costs and carbon emissions. Furthermore, the research [65] tackled the concerns of the operational adaptability of clusters of cogeneration facilities in windy areas. It provides a hybrid system combining cogeneration units with an electric boiler subsystem and a combined heat and compressed air storage (CH-CAES) subsystem. A two-level optimization algorithm is used.

Research on the economic viability of cogeneration systems for RES utilizing ammonia (NH₃) and H₂ as energy storage was conducted in [66]. The optimization model accounts for the need for 100% RESs in various places as well as scenarios where historical energy costs change. The findings indicate that even at higher energy prices, 100% renewable cogeneration systems powered by renewable NH₃ are economically viable, indicating their potential to supply heat and electricity to remote areas with high energy costs. In order to increase energy security, cogeneration technology and RESs were combined in the study [67]. The current state of cogeneration technologies based on RESs is discussed, emphasizing the need to increase the efficiency of RESs and reduce environmental impacts. The review shows that biomass is an important part of the RES used in cogeneration power plants. The importance of district heating and cooling through cogeneration technology is highlighted, showing the potential reduction in biomass consumption when used in value-added sectors.

The importance of microgrids and energy hubs is assumed in [68], which focuses on optimal design for sharing CHP sources, GH₂ storage systems, and electric vehicles and controllable loads. The research aims to minimize the LCOP and environmental pollution using probabilistic modelling. The model includes a GH₂ storage system and uses load-side response programs to optimize the microgrid. Ref. [69] analyzed a micro-CCHP in which the energy source of the internal combustion engine is the syngas produced by the gasification of wood biomass. The overall efficiency level of the micro-CCHP system is considered, making it economically feasible.

In [70], the wind resistance of the inelastic operational mode of cogeneration equipment of coal-fired power plants is studied. The study proposes to integrate district heating networks (DHNs) into electricity distribution to improve wind energy housing. A new approach involves integrating DHN flow control into powertrain models. The results show a higher adaptability of wind energy at higher flow rates due to better thermal inertia. Wind energy limitation at a flow velocity of 1.0 m∙s⁻¹ is reduced by 5.5% compared to 0.5 m∙s⁻¹. However, one must be careful about excessive flow rates, because the water pumps’ increased power consumption is the main concern.

The work conducted in [71] describes how H₂ might help cut down on the world’s energy usage and presents a multi-time–space scale optimization applied to a regional multi-energy system (RMES) integrated with H₂. The first and second phases are optimized to maintain the system configuration and the region’s day-ahead hourly operation plan with the aim of reducing both the total annual cost and daily operational expenses. In the third phase, an intra-day operation plan for each community is manifested using a moving-horizon optimization approach with a time resolution of 15 min. Finally, in the fourth phase, the operation plan for electricity-related equipment in each community is optimized at 5 min intervals, concentrating on minimizing the required adjustments. The study [72] focused on the use of photovoltaics (PVs) in building energy systems (BESs) and presents a hybrid predictive model in the framework of model predictive control (MPC). The proposed model is validated through real data from a Japanese office building, which shows battery safety and the continuous operation of cogeneration. The hybrid predictive MPC framework shows an 81.6% improvement in battery safety and a 36.4% continuous use of cogeneration compared to baseline control logic.

Coastal regions are represented by a small-scale hybrid electric–freshwater distribution system [73] that combines hydrothermal irrigation–drying (HDH) desalination with combined thermal–isobaric compressed air energy storage (CH-ICAES). With a Gained Output Ratio (GOR) of 2.0678 and an LCOW of CAD 0.57/ton, the system exhibits competitive economic performance even though it uses poor-quality CH-ICAES heat for desalination. Seawater temperature at the humidifier input, the efficiency of the humidifier and dehumidifier, and the CH-ICAES operating conditions are the primary factors influencing system performance.

A CCHP cogeneration system with heat recovery based on geothermal energy was introduced in [74]. The system incorporates the organic Rankine cycle (ORC), ejector cooling cycle (ERC), and Kalina cycle (KC). A multi-objective optimization algorithm performed the optimal parameter calculation utilizing the TOPSIS approach, taking into consideration variables such as energy efficiency, net power, and total investment cost. A geothermal heat source performs better considering thermal efficiency, energy efficiency, and economic viability when it is operating at its optimal temperature of 490 K. Moreover, a mixed-integer nonlinear programming (MINLP) model is considered in [75] for a renewable CCHP (RCCHP) system to evaluate economic and environmental performance across five buildings under two optimization scenarios of (1) economic metrics and (2) the CO₂ emission reduction rate (CER) to guide the optimal design and operation. The results of the proposed work showed that integrating off-peak electricity purchases with daytime CHP and wind turbine generation is cost-efficient and improves energy efficiency.

The research [76] analyzed the energy and cost effectiveness of a solar geothermal combination heating, cooling, and power (SG-CCHP) system. As energy storage modules, the system integrates HP with battery cells, a GH₂ tank, as well as steam turbines, PV/thermal collectors, fuel cell circuits, and absorption chillers. The represented solution reduces life cycle costs and maximizes efficiency by utilizing the best design and control strategies. Control of supercritical CO₂ systems was also performed in [77] during design variation and temporary operation with a focus on a 50 kW CO₂ test plant. The study looks at controlling system inventory to meet management objectives and takes stability into account. The tank’s volume, which is three times the power chain’s volume, increases maneuverability and broadens the scope of control operations. The turbine inlet temperature is managed by a PI controller, demonstrating the possibility of efficient management in systems with reduced CO₂ levels. Also, a metaheuristic-algorithm-based configuration of CHP systems is considered in [78], including the renewables of PV/T, WT, electric heater, and thermal energy storage (TESS). The proposed implemented model was the improved marine predators algorithm (IMPA), which is applied to the total annual cost as the objective function of the CHP system optimal configuration. The research [79] developed a decision tool to lessen the environmental effects of large commercial buildings and HVAC systems. The HVAC system is equipped with a hybrid system that combines traditional HVAC components with building-integrated photovoltaics (BIPVs) and CCHP, using an airport terminal as a case study. Intelligent controllers like the model predictive controller (MPC) and fuzzy sliding mode control (FSMC) enhance the hybrid system by reducing CO₂ emissions and increasing the comfortable temperature.

The authors of [80] concentrated their review on the application of metaheuristic optimization methods in cogeneration systems. It divides optimization issues, such as time, size, and scheduling issues, into single-objective and multi-objective methods. This study investigates the algorithmic optimization in enhancing the efficiency of cogeneration facilities. It also identifies the primary areas of research and evaluates the effectiveness of various optimization techniques. Ref. [81] investigated how large industrial power plants (LIPPs) could improve the ability of power systems to withstand harsh circumstances. Demand response initiatives and mobile energy storage are thought to increase system sustainability. A redesigned IEEE 24-bus power system with LIPPs is the subject of a stochastic mixed-integer linear programming model development. The findings demonstrate the important role that LIPPs play in enhancing power system resilience, lowering the unsupplied load, and highlighting the extra advantages of demand response and mobile energy storage. As a result,

Table 2. A summary of the conducted literature review.

Ref. No	CCHP/CHP Systems	Biomass	Solar Energy	Geothermal Energy	Fuel Cells	Key Findings
[34]	✗	✗	✗	✗	✗	Improved performance with microchannel heat sinks and energy efficiency.
[35]	✗	✓	✗	✗	✓	High electrical efficiency, low CO2 emissions, and economic viability.
[36]	✓	✗	✗	✗	✓	Improved efficiency, reduced greenhouse gas emissions, and optimized importance.
[37]	✗	✗	✗	✗	✗	Reduction in operational costs and emissions; effective energy management.
[38]	✗	✗	✓	✗	✗	Review of solar hybrid systems and insights into recent developments.
[39]	✗	✗	✗	✗	✗	Strategy for enhancing power system resilience; 52% improvement in ELNS.
[40]	✗	✗	✗	✗	✗	Reduction in aggregated system costs; effective optimal planning.
[41]	✓	✗	✗	✗	✓	Analysis of functional relationships; a reliable method for determining suitable operating temperatures.
[42]	✗	✗	✗	✗	✗	Improved multi-agent coordination, decreased system operation costs, and reduced carbon emissions.
[43]	✗	✗	✗	✗	✗	Flexibility and improved energy utilization; effectiveness confirmed on IEEE 33-bus test network.
[44]	✗	✓	✗	✓	✗	System efficiency and combined cycle heating and cooling optimization using multi-objective GWO.
[45]	✗	✗	✗	✗	✗	Overview of GH2 energy systems with a focus on thermal management and process optimization.
[46]	✗	✗	✗	✗	✗	Adjustable TEC input, length, and thickness affect temperature reduction; optimal battery thermal management.
[47]	✗	✗	✗	✗	✗	Integration of SWIR detectors with multi-stage TECs and MHS; best performance and minimum volume.
[48]	✗	✗	✗	✗	✗	A two-way programming optimization model reduced system operating costs and carbon emissions.
[49]	✗	✓	✗	✗	✓	Optimization of design variables and balance between efficiency, emission reduction, and economic viability.
[50]	✗	✗	✗	✗	✗	Safe operations under full conditions, improved power tracking, and disturbance rejection.
[51]	✗	✓	✗	✓	✗	Technical, environmental, and economic evaluation; bi-objective optimization for efficiency and cost.
[52]	✗	✗	✗	✗	✗	Review of decarbonization strategies; focus on GH2, onshore power supply, challenges, and potential.
[53]	✗	✗	✗	✗	✗	Image gray recognition-based defrosting control; improved accuracy and COP.
[54]	✗	✗	✗	✗	✗	Comparison of temperature control methods, thermal storage capacity assessment, and peak-shaving potential.
[55]	✗	✗	✗	✗	✗	Parallel air-cooled system for battery packs; efficient thermal management for varying conditions.
[56]	✗	✗	✗	✗	✗	Predictive modelling for CO and NOx emissions; DFR model for higher prediction potential.
[57]	✓	✗	✗	✗	✗	Comparison of HBP and LZPO renovations; improved peak-shaving, heating capacity, and efficiency.
[58]	✓	✗	✗	✗	✗	Low-carbon model with flexible load, CHP, CCS, and P2G; reduced transaction and operating costs.
[59]	✓	✗	✗	✗	✗	Dynamic model for fluidized bed combustion, control strategies for load changes, and decoupling.
[60]	✗	✗	✗	✗	✓	High efficiency, low emissions, fuel flexibility for marine applications, and methane as an efficient fuel.
[61]	✓	✗	✗	✗	✗	Integration technologies for improved flexibility; increased heat–electricity ratio.
[62]	✓	✗	✗	✗	✗	Molten salt TES integration for flexibility, increased thermal and exergy efficiency.
[63]	✓	✗	✓	✗	✗	Solar-aided CHP integration, advantages in efficiency and economics.
[64]	✓	✗	✗	✗	✗	Multi-objective optimization for cost and CO2 reduction in microgrid scenarios.
[65]	✓	✗	✗	✗	✗	A hybrid system with wind power, improved efficiencies, and reduced environmental impact.
[66]	✓	✗	✗	✗	✗	Economic competitiveness of renewable NH3-enabled CHP systems, even with high energy prices.
[67]	✓	✓	✗	✗	✗	Exploration of renewable CHP technologies and potential reduction in biomass consumption.
[68]	✗	✗	✗	✗	✗	Optimal planning in microgrids, demand-side management, reduced costs, and pollution.
[69]	✓	✓	✗	✗	✗	Innovative micro-CCHP system with woody biomass gasification, high efficiency, and economic feasibility.
[70]	✗	✗	✗	✗	✗	Integration of DHN flow control, increased wind energy adaptation, and warning against excessive flows.
[71]	✗	✗	✗	✗	✗	Role of H2 in decarbonization, economic viability, and optimal planning for RMESs.
[72]	✗	✗	✓	✗	✗	Improved battery safety, 81.6% enhancement in battery safety, and 36.4% in CHP operation.
[73]	✓	✗	✗	✗	✗	Hybrid CH-ICAES and HDH desalination, competitive economic performance, and GOR of 2.0678.
[74]	✓	✗	✗	✓	✗	Geothermal-based CCHP with Kalina cycle; improved thermal efficiency and economic feasibility.
[75]	✓	✗	✗	✗	✗	Optimal design of CCHP system for five buildings under two optimization scenarios.
[76]	✓	✗	✓	✓	✓	Solar and geothermal CCHP with steam turbines, PV/T, improved efficiency, and reduced costs.
[77]	✗	✗	✗	✗	✗	Control of sCO2 systems with a PI controller; better controllability for larger tank volumes.
[78]	✓	✗	✗	✗	✗	Optimal CHP configuration using the IMPA model, minimizing total annual cost.
[79]	✓	✗	✓	✗	✗	HVAC retrofit with BIPV and CCHP, reduced CO2 emissions, and improved comfort.
[80]	✓	✗	✗	✗	✗	Overview of CHP metaheuristic optimization and comparison of optimization methods.
[81]	✗	✗	✗	✗	✗	Industrial energy hubs, mobile storage, demand response, and reduced residual load.

✓: Considered; ✗: Not Considered.

presents a summarized overview of the surveyed works.

Thus, a thorough summary of several research works aimed at improving energy systems was given. The results show that technology such as smart inverters and microchannel heat sinks can improve efficiency and performance (Ref. No [34,41,47,55]). Various optimization strategies, such as meta-heuristic algorithms and multi-objective optimization, are essential for enhancing different elements, including carbon emissions and cost (Ref. No [36,37,59,65,80,81]). A dedication to ecologically friendly solutions is highlighted by a strong emphasis on renewable and sustainable integration, especially in biomass, geothermal, and solar energy (Ref. No [35,45,61,67,75]). The body of knowledge is further enhanced by technological advancements, evaluations of economic viability, and an emphasis on power system resilience (Ref. No [52,72,74,77,81]). These revelations support the continuous search for resilient, economically viable, and sustainable RESs that meet the changing demands of society.

4. Artificial Intelligence Integration

A spectrum of intelligent and AI-driven models has been demonstrated for the management, scheduling, and optimal planning of CHP systems, as well as heat, and electricity incorporated power networks, integrated with different RES.

4.1. Overview of Intelligent Model

The intelligent models’ overall classification is shown in Figure 6, which completes the general categories of intelligent AI/ML technologies. AI/ML models are categorized into supervised/unsupervised/semi-supervised and reinforcement learning technologies. In supervised learning, models are trained on labeled datasets that address both continuous (regression) and categorical (classification) target variable tasks. Common algorithms include linear regression [82], logistic regression [83], support vector machines (SVMs) [84], and neural networks (NNs). Unsupervised learning [85] involves models that learn from unlabeled data and is often used for clustering without a predefined target variable. Semi-supervised learning [86] combines labeled and unlabeled information during training. Reinforcement learning (RL) focuses on decision making, where agents learn by interacting with the environment, receiving rewards or punishments. These models are versatile tools that can be applied to various tasks in regression, classification, clustering, and control scenarios, providing solutions to various AI/ML challenges. Based on this context, an overview of recently developed AI models in CHP systems is presented in

Table 3. The literature of recent intelligent methodologies in CHP-incorporated systems.

Ref.	Algorithm	Bench System	Purpose
[87]	Adaptive Fitness-Distance Balance-Based Artificial Rabbits Optimization (AFDB-ARO)	CHP System	Optimal Solution of CHP Economic Dispatch
[88]	ANN	PEMFC-Based CHP	Modelling/Optimization
[89]	Resource Allocation-Energy Sharing Algorithm	Intelligent Community With CHP Systems	Carbon Management
[90]	Deep RL	CHP System	Intelligent Solution of Economic Dispatch
[91]	Multi-Scenario Analysis	Fuel Cell-Integrated CHP	Primary Energy Consumption Minimization
[92]	Multi-Scenario Analysis And Control	Residential Heating System	Emissions and Costs
[93]	Data-Driven Analysis	TESS-Integrated CHP	District Heating
[94]	Heap-Based Technique With Enhanced Discriminatory Attribute	LSCHP	Optimal Solution of Economic Dispatch
[95]	Isochronous Governor Control Strategy Integrated With OPAL-RT	CHP	Zero-Steady-State-Error Frequency Regulation
[96]	Different AI Models	TESSs	Classifications, Roles, and Optimizing Design of TESSs
[97]	Dynamic-Objective Method (DOM)	Multi-Energy Complementary CHP System	Economic and Environmental Benefits and the Cogeneration System Source Side Load
[98]	Adaptive Control	Residential Micro CHP	Energy Cost Minimization
[99]	ML	Phase-Change Material Integrated Systems	Cooling Performance Enhancement
[100]	AI-Supervisory Control and Data Acquisition (SCADA)	CHP Cogeneration Units Based on Landfill Biogas	Some AI-based Diagnostic Tools by Multithreaded Polymorphic Models, Integrated with SCADA Systems
[101]	Real-Time Intelligent Control Laboratory (RT-ICL)	PoewrLabDK	Smart Grid Technology Development
[102]	Data-Driven Method Achieves Reasonable Fidelity With Monitored Demand	Domestic Hot Water Cylinder	Minimal Solution
[103]	Recurrent Deterministic Policy Gradient (RDPG)	District Heating System	HTR and the Stable Control Performance
[104]	Mobile Gasification Unit (MGU)	Small-Scale CHP	Performance Analysis
[105]	Multi-Energy Complementary Model (MECM)	Coupling System of Distribution Network and HPES	Utilization Rate of RES; Shifting/Filling Peaks/Valleys to Economic Analysis
[106]	Kuhn–Tucker	CHP	Optimal Solution and Control
[107]	Multi-Player Harmony Search Method (MPHS)	CHP	Optimal Solution of Economic Dispatch
[108]	Modified Teaching-Learning-Based Optimization (MTLBO)	CHP	Optimal Solution of Economic Dispatch
[109]	Oppositional Teaching Learning-Based Optimization (OTLBO)	CHP	Optimal Solution of Economic Dispatch
[110]	Civilized Swarm Optimization and Powell’s Pattern Search (CSO-PPS)	CHP	Optimal Solution of Economic Dispatch
[111]	Opposition-Based Group Search Optimization (OGSO)	CHP	Optimal Solution of Economic Dispatch
[112]	Advanced Modified Particle Swarm Optimization (AMPSO)	CHP	Optimal Solution of Economic Dispatch
[113]	Gravitational Search Algorithm	CHP	Optimal Solution of Economic Dispatch
[114]	Bee Colony Optimization (BCO)	CHP	Optimal Solution of Economic Dispatch
[115]	Improved Artificial Bee Colony (IABC)	CHP	Optimal Solution of Economic Dispatch
[116]	Different AI Model	Heat and Power Incorporated Networks	Optimal Solution of Economic Dispatch

.

As common models, the AI algorithms of Long-Short Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Random Forest (RF) are taken into simulation in this review. Their structure is shown in Figure 7.

The AI model LSTM (Figure 7a) is a type of recurrent neural network (RNN) [117] architecture designed to address the gradient problems often encountered in standard RNNs when modelling sequential data with long-term dependencies. LSTMs achieve this by incorporating a memory cell and three key gates of input (i_t), forget (f_t), and output (O_t) that regulate the flow of information into, through, and out of the cell, where relevant information is retained over extended time steps while irrelevant information is discarded. The i_t determines the extent to which new information is added to the memory cell, the f_t controls how much of the existing information is removed, and the O_t decides how much of the cell’s content is used to compute the current output. These gates are controlled by learnable weights and biases that are optimized during training using backpropagation through time (BPTT). BiLSTM networks [118] (Figure 7b) are an advanced type of LSTM model designed to process sequential data in both forward and backward directions, allowing for the capture of contextual information from both past and future states of a sequence. This architecture includes two parallel LSTM layers: one processes the input sequence from start to end (forward pass), while the other processes it from end to start (backward pass). The outputs of these two layers are typically concatenated or combined at each time step. Each LSTM layer within the BiLSTM contains the same components as a standard LSTM, which regulate the flow of information to retain long-term dependencies. As the final model, RF [119] (Figure 7c) is considered. It is an ensemble learning method used for both classification and regression tasks to improve the predictive accuracy and control overfitting issues commonly faced by individual decision trees (DT) [120]. It operates by constructing a multitude of DTs during training and combines their outputs through averaging (for regression) or majority voting (for classification) to produce a final prediction. Each tree in the forest is built using a random subset of the training data, obtained through bootstrapping, and at each split, the algorithm considers a random subset of features rather than all available features, introducing diversity among the trees and reducing the correlation between them. This randomness ensures that the model captures a wide variety of patterns in the data, leading to improved generalization to unseen data. The depth of the trees, the number of trees, and the subset size of features are hyperparameters that can be tuned to optimize performance for specific tasks. To this end, a summary of these models and their applications in CHP systems is presented in

Table 4. Summary of the presented AI/ML models.

Model	LSTM	BiLSTM	RF
First Introduced in Ref.	[121]	[122]	[123]
Advantages/ Disadvantages	Handles sequential data and captures long-term dependencies. Vanishing gradient issue and high training time.	Bidirectional context and improved performance in time-series data. Higher computational complexity and requires more resources.	Robust to noise; simple and interpretable. May overfit and less effective for time-series data.
Applications in CHP Systems	Predicting load demand and optimizing energy distribution	Fault detection and dynamic state prediction.	Anomaly detection in system operations and feature selection.
Dynamics Dependency	✪ ✪ ✪	✪ ✪ ✪	✪ ✪ ✪
Data Dependency	✪ ✪ ✪	✪ ✪ ✪	✪ ✪ ✪
Computational Complexity	✪ ✪ ✪	✪ ✪ ✪	✪ ✪ ✪
Structural Complexity	✪ ✪ ✪	✪ ✪ ✪	✪ ✪ ✪

✪ ✪ ✪: low, ✪ ✪ ✪: moderate, ✪ ✪ ✪: high.

.

Regarding the dependency on dynamics, all models have less sensitivity, which means these algorithms are model-free, and they do not operate based on the system dynamics, whereas they operate based on the data they receive, which is data dependency. Computational complexity is another important consideration, which includes time and hardware resources for running the model. LSTM is computationally intensive, with BiLSTM demanding even greater resources due to its bi-directional structure. However, RF is less computationally demanding due to its simpler structure. The structural complexity of the models also varies, with RF being the simplest, LSTM moderately complex, and BiLSTM the most structurally intricate due to its bidirectional layers.

4.2. Key Performance Indicators (KPIs)

The KPIs for ML models are regularly used metrics that assess the effectiveness and performance of these models in solving specific tasks. In this review, the KPIs of MAE, MSE, and RMSE are defined as in [124,125,126,127].

MAE is a metric in deductive processes. It measures the absolute mean difference between the model predictions and the observed values. The MAE formula is measured by taking the absolute difference between the predicted and observed data and dividing this by the total number of times.

Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) are crucial metrics in regression analysis that assess forecast accuracy by calculating the square difference between anticipated and actual values. MSE calculates the average of squared errors, penalizing larger deviations more. RMSE, the square root of MSE, offers a more detailed estimate of the variable’s components. Both metrics are crucial for evaluating the adequacy of the regression model, with lower values suggesting superior performance. While Mean Squared Error (MSE) is influenced by outliers because of the squared error, RMSE offers a more precise insight into the average magnitude of the error. These measures are commonly utilized in fields like finance, economics, and engineering to direct assessment and enhancement. Consequently, the summary of the proposed KPIs is presented in

Table 5. Summary of the defined KPIs in intelligent models.

Metric	Formulation	Insights in CHPs	Best Value	Worst Value
MAE	${\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n}\left|{\mathcal{l}}_{i,actual}-{\mathcal{l}}_{i,predicted}\right|$	Reflects the average absolute error between actual and predicted values in energy generation, heat output, or other operational parameters, providing a straightforward measure of prediction accuracy.	0	$+\infty$
MSE	${\displaystyle \frac{1}{n}}\sum ({\mathcal{l}}_{Observed}-{\mathcal{l}}_{Predicted}{)}^2$	Penalizing larger deviations more heavily, which is important in CHP systems for avoiding large operational errors and having reliable system outputs.	0	$+\infty$
RMSE	$\sqrt{{\displaystyle \frac{1}{n}}\sum ({\mathcal{l}}_{Observed}-{\mathcal{l}}_{Predicted}{)}^2}$	Useful for understanding the typical magnitude of prediction errors in key CHP metrics such as power and heat efficiency.	0	$+\infty$

, where n is the number of data and ${\mathcal{l}}_{actual}$ and ${\mathcal{l}}_{\mathit{Predicted}}$ are the actual and predicted values, respectively.

5. Case Study Simulation Review

For the purpose of assessing the impact of intelligent models on CHP systems, a CHP model was developed using MATLAB 2019b. This model incorporates the dynamics and interactions within a CHP system, allowing for a detailed simulation of its performance, as illustrated in Figure 8.

Regarding Figure 8, the simulations manifest dynamic responses within the CHP system across different parameters. The generator’s mechanical power, varying from 0 to 1 P.U, demonstrated a swift attainment of steady-state conditions within 0.6 s. Similarly, the final leg voltage, fluctuating between 2 to 9 P.U, reached a steady state in a fraction of a second. The initial leg voltage exhibited oscillations from 0 to 0.3 P.U. Generator speed, with variations from 0.97 to 1 P.U, achieved a steady state of 1 P.U within 1.5 s. The DC voltage exhibited damped oscillations, settling into a steady state of 450 V by 0.9 s. The inverter voltage, ranging from −500 V to 500 V, showcased dynamic behavior, indicating the control and regulation aspects of the CHP system. These simulation outcomes provide perspectives on the transient and steady-state behavior of the CHP components, aiding in the assessment and refinement of the system’s performance and control strategies.

As the next phase is the intelligent and AI-driven analysis of the generated data, it was applied to the CHP data provided in

Table 6. The details of the generated dataset.

	Mechanical Power	Initial Leg Voltage	Final Leg Voltage	Motor Speed	DC Voltage	Inverter Voltage
Mean [P.U]	0.625453	0.998333	0.172192	0.991734	498.5686	498.5686
Variance	0.016934	0.03189	0.000822	6.72 × 10−5	2223.683	2223.683
Amount of Data	102,124	102,124	102,124	102,124	1,021,232	1,021,232

, encompassing mechanical power, initial leg voltage, final leg voltage, motor speed, DC voltage, and inverter voltage. The data summary indicates the mean and variance for each variable, providing insights into the central tendency and dispersion of the dataset. The amount of data, representing the number of instances for each variable, is also specified. These intelligent models aim to leverage the temporal dependencies and intricate relationships within the CHP data to enhance predictive accuracy and contribute to the optimization of the CHP system’s performance across various parameters. The analysis results are shown in Figure 9.

For instance, the generator’s mechanical power swiftly reached steady-state conditions, while the final leg voltage achieved stability in a fraction of a second. The initial leg voltage exhibited oscillations, and the generator speed attained a steady state within 1.5 s. Additionally, the DC voltage displayed damped oscillations, settling into a steady state of 450 V by 0.9 s, while the inverter voltage showcased dynamic behavior, reflecting the control and regulation aspects of the CHP system. These findings offer valuable insights for assessing and refining the cogeneration or the CHP system’s performance and control strategies by the application of intelligent models. Consequently, the KPI results of the models are shown in

Table 7. The KPI results derived by the models.

BiLSTM
Parameter	MAE	RMSE	MSE
Mechanical Power	0.315146	0.373435	0.139454
Initial Leg Voltage	0.10045	0.564845	0.31905
Final Leg Voltage	0.031708	0.041718	0.00174
DC Voltage	0.031708	0.041718	0.00174
Motor Speed	0.261886	0.309942	0.096064
LSTM
Parameter	MAE	RMSE	MSE
Mechanical Power	0.095493	0.559512	0.313054
Initial Leg Voltage	0.032234	0.040584	0.001647
Final Leg Voltage	0.194758	0.247855	0.061432
DC Voltage	0.194656	0.247679	0.061345
Motor Speed	0.009416	0.01119	0.000125
Random Forest
Parameter	MAE	RMSE	MSE
Mechanical Power	0.088961	0.565373	0.319646
Initial Leg Voltage	0.036392	0.045858	0.002103
Final Leg Voltage	0.194664	0.247636	0.061324
DC Voltage	0.194664	0.247636	0.061324
Motor Speed	0.010003	0.011923	0.000142

and Figure 10.

As shown in Figure 10, the mechanical power of BiLSTM had an MAE of 0.315, an RMSE of 0.373, and an MSE of 0.139. The MAE for the initial leg voltage was 0.100, the RMSE was 0.565, and the MSE was 0.319. The MAE, RMSE, and MSE values for the final leg voltage and DC voltage were 0.032, 0.042, and 0.002, respectively. BiLSTM demonstrated the following values for motor speed: MAE 0.262, RMSE 0.310, and MSE 0.096. The mechanical power MAE, RMSE, and MSE for LSTM were 0.095, 0.560, and 0.313, respectively. The MAE and RMSE for the initial leg voltage were 0.032 and 0.041, respectively, with an MSE of 0.002. The MAE, RMSE, and MSE values for the final leg voltage and DC voltage were 0.195, 0.248, and 0.061, respectively. The RMSE for motor speed in LSTM was 0.011, the MAE was 0.009, and the MSE was 0.0001. Random Forest produced mechanical power results with respective MAE, RMSE, and MSE values of 0.089, 0.565, and 0.320. In terms of the initial leg voltage, the MAE was 0.036, the RMSE was 0.046, and the MSE was 0.002. MAE, RMSE, and MSE values for the final leg voltage and DC voltage were 0.195, 0.248, and 0.061, respectively. The RF motor speed exhibited an MAE of 0.010, an RMSE of 0.012, and an MSE of 0.0001. When it came to forecasting mechanical power, the Random Forest model demonstrated the lowest values of MAE, RMSE, and MSE, indicating that it exhibited superior accuracy in this regard. Conversely, in terms of initial leg voltage, the LSTM model achieved the minimal MAE and RMSE.

6. Discussion on Challenges and Future Perspectives

Although AI integration in CHP systems presents several benefits, and it is effective in the transition to a green environment and grid modernization, it faces several challenges (

Table 8. Challenges and their potential perspectives.

Challenge	Impact Area	Stakeholders Involved	Technology Enablers	Future Work/Prospective Solutions
Data Availability	Data Quality	Utilities and Researchers	Open Data Platforms and APIs	Establish open-access datasets and collaborative platforms for high-quality CHP data; incentivize utilities and stakeholders to contribute.
Data Integration	Interoperability	System Integrators and Manufacturers	IoT and Data Standards	Develop standardized data protocols and adopt IoT frameworks to ensure seamless, unified data acquisition and communication.
Real-Time Data Processing	Operational Efficiency	Engineers and IT Teams	Edge Computing and AI Accelerators	Leverage edge computing and advanced AI algorithms for real-time analytics; utilize cloud infrastructure for scalable data processing.
System Complexity	System Design	Researchers and Software Developers	Hybrid Modelling and Simulation Tools	Employ hybrid modelling approaches that combine physics-based and AI-driven techniques; design modular architectures for system simplification.
Uncertainty and Nonlinearity	Model Reliability	AI Developers and Operators	Probabilistic Models, RL, and Ensemble Learning	Use robust AI methods to manage uncertainty and complex dynamics.
Scalability Issues	Deployment	System Designers and Utilities	Distributed Computing and Multi-Agent Systems	Enhance AI algorithms for distributed/cloud-based deployment; implement scalable multi-agent control systems for large-scale CHP networks.
Generalization Across Systems	Transferability	Researchers and AI Modellers	Transfer Learning and Meta-Learning	Train models on diverse datasets to improve adaptability; apply transfer learning for efficient deployment.
Real-Time Control	Decision Making	Control Engineers and Operators	Reinforcement Learning and MPC	Apply RL and predictive control strategies for adaptive real-time decision making.
Cybersecurity Vulnerabilities	System Security	IT Security Teams and Regulators	Blockchain and Intrusion Detection Systems	Integrate AI-powered intrusion detection and blockchain technologies to secure communication and data integrity.
Data Privacy Concerns	Compliance	Data Officers and Legal Teams	Federated Learning and Differential Privacy	Adopt privacy-preserving AI approaches to protect sensitive data during model training.

), especially in the data, modelling, scalability, integration, cybersecurity, economic, and skill barriers. A major issue is a lack of sufficient and high-quality datasets on which to train AI models, difficulties in integrating several data sources, and managing real-time data processing for dynamic and predictive applications. Uncertainties that are known to occur in load demands, RES inputs, and nonlinear behaviors are also required to be captured when formulating optimization models for these systems that are inherently multi-energy domain due to the involvement of thermal and electrical components. Disposals of a more distributed and larger system lead to a loss of efficiency of trained AI models and a displacement to other configurations of the represented system. Integration and control cause challenges like retrofitting AI solutions into architectures and ensuring that algorithms have actionable, real-time data within operational limits. Concerns related to cybersecurity and privacy include the susceptibility to hacking and potential risks with sharing operational data.

Predictive analytics has made progress with advanced maintenance models aimed at minimizing downtimes and extending system lifespans, alongside enhanced load forecasting to improve demand-side management and efficiency. Real-time optimization and control have been improved by the implementation of RL and other AI techniques for dynamic energy production and distribution, as well as intelligent control strategies for the integration of RES with CHP systems. Integration with smart grids has led to AI-driven coordination for improved reliability and resilience while also supporting virtual power plant (VPP) functionalities that position CHP systems as flexible energy nodes. Personalized energy solutions are achievable through AI-based customized energy management strategies tailored to specific customer needs and optimized CCHP solutions for urban and industrial applications. Hybrid AI models, which merge physics-based and data-driven approaches, have enhanced accuracy and reliability, while explainable AI (XAI) increases wider adoption. Integration with the internet of things (IoT) and edge computing enables real-time monitoring and localized data processing, reducing latency and improving control actions. Economic and environmental impacts are being addressed through AI-driven cost–benefit analysis tools to assess feasibility.

7. Conclusions

In this review, the significance of CHP systems is presented, with a focus on their prevalence and statistics in the USA and Europe. The mathematical foundations of these systems were provided, and their recent advances were investigated. This review set up a wide range of works, advancements in CHP systems, and the integration of intelligent control methods for optimal planning, economic dispatch, and cost minimization. Intelligent models, including LSTM, BiLSTM, and RF, were fully presented. Subsequently, a CHP system was simulated, and the mentioned models were applied to analyze, evaluate, and compare their performance. KPIs were derived using the models’ effectiveness in governing CHP systems. The perspectives drawn from these analyses are that LSTM, BiLSTM, and RF, generally intelligent models, have the capacity to be considered in CHP systems under the control and optimization methods. Lastly, the future perspectives of CHP systems were outlined, emphasizing their potential as integral components of modern power and energy systems, with several challenges in AI integration, such as data availability and real-time performance. Future works on AI-CHP integration should consider solving these types of existing issues.