Modeling and Optimization of Isolated Combined Heat and Power Microgrid for Managing Universiti Teknologi PETRONAS Energy

With the rapid growth of isolated microgrids, combined heating and power (CHP) can be integrated with photovoltaic (PV) system. The integration of CHP-PV systems has a tremendous potential to increase system reliability and efficiency as well as reduce energy consumption and CO2 emission. However, the operating models require considerable analysis due to the uncertainty load demand, so it is not easy to apply the models in order to simulate the current trend of baseline systems. In this paper, we developed models based on optimum clustering data to perform the behavior of CHP-PV using the integrating of accelerated particle swarm optimization (APSO) and fuzzy subtractive clustering (FSC). The objective of the APSO algorithm is to tune the parameters of data clustering-based FSC using proportional integral (PI) controller. The paper’s main goal is the minimal total energy and fuel consumption without compromising load demand of cooling. The proposed model interacts to the energy by gas turbine generators (GTGs) and PVs system. Also, it subdivides cooling load with the use of the partial load condition according to the outdoor weather. A case study of Universiti Teknologi Petronas (UTP), Malaysia was used to investigate its CHP plant. The model is validated using actual data obtained from University CHP plant. The results demonstrate that the proposed optimum system including CHP, PV, and storage systems is outperformed on the baseline system and basic CHP models. The proposed optimum models save 7 % and reduce 4.72 % of total daily electrical and steam production, respectively. Also, the optimal system is kept cooling demand satisfied.

photovoltaic (PV) are considered a compromise for isolated microgrids [4], [5]. It offers better solutions and imparts an advantage of reducing greenhouse gas emissions and fuel consumption. It also increases system reliability and energy efficiency [5], [6], [7], [8]. In recent years, several studies with different operational strategies have been proposed to manage and optimize the integrated CHPs sizing [3].
Franco et al. [9] proposed the use of composite indicators for planning the operation of optimum design of small-scale CHP units in district heating systems. The study has operational strategy to manage system energy with the objective of increasing the system global efficiency and reducing heat losses, this constraint determines sizing of the CHP plants at quite reduced values. Reference [10] mixed-integer linear programming (MILP) model proposed of a multi-energy system. The study considered controllable loads oriented demand response for heat pumps, CHP, energy storage, air conditioning systems, PV and WT sources. The work aims to meet the electrical, heating, and cooling demands of end-users at a minimal cost based on the examined PV and WT models. Also, Reference [11], introduced hybrid renewable energy system (RES) using techno-economic optimization for the concurrent supply of electricity and heat. The system consists of CHP, PV and WT sources. After hybrid RES optimization, CHPs plant investigated and it was in best performance. Furthermore, the authors of [12] optimized CHP unit sizing using Artificial Bee Colony (ABC) algorithm to minimize the energy costs of residential areas. The proposed ABC algorithm outperformed onto the Genetic algorithm (GA) when it compared with to determine optimum CHP sizing. Liu et al. [13] proposed a hybrid system of smart energy building cluster with CHP and PV using the Alternating Direction Method of Multipliers (ADMM) to solve the coalitional energy management. Both CHP and PV are acted as role-exchangeable energy sellers/buyers with the design of Two-level reward allocation scheme to measure the contributions from the CHP operator and PV prosumers.
Different operation schemes of CHP units have been performed and analyzed to assess energy and pollutant emissions [14]. Typical optimization techniques used in combined cooling, heating and power (CCHP) system such as dispatch and genetic algorithm (GA) to reduce energy, CO2 emissions, and operation cost [15], [16]. Reference [17] formulated a model to solve CHP using a multi-objective optimal dispatch problem for heat and power. A CHP economic dispatch problem by whale optimization aims to minimize the fuel cost [18]. In study by [19], developed a model and optimized it based on identifying a dynamic CHP system to simplify and minimize the operating cost of cooling load, electricity and gas consumption. Reference [3] developed an optimization problem by MILP and GA for CHP plant. The model integrated with the TES to benefit for higher operational flexibility and betterment overall performance. Sundaram et al. [20] has proposed non-dominated sorting genetic algorithm II and multi-objective PSO technique, which it was effectively reducing fuel costs and the emission levels. Reference [21] suggested a hybrid CHP economic dispatch's (CHPED) algorithm to solve the complex problem of CHP system. A study in [22], CCHP modeling and energy management have been analyzed the technical, economic, and environmental aspects. A model has used to investigate the performance of CHP-PV in term of economic aspects [23], [24]. Huang et al. [25] has built an integrated energy micro-grid model to manage the CCHP plants to enhance the reliability, flexibility, economic aspects of energy supply in multiple districts.
In [26], introduced the investigation the performance of a CCHP solar GT system to assess 4E (energy, exergy, exergoeconomic, and environment). Lin et al. [27] has adopted an analytical method to solve multi-energy problems between the CCHP integrated with micro-energy; PV and wind turbine (WT), and the consumers. The PV and WT main objectives in CHP plant are to obtain economic profits and balance energy for consumers. A fuzzy has used as a tool to assess the performance of CHP behavior [28]. Wang et al. [29] has developed a multi-objective particle swarm optimization (PSO) to solve the problem of CHP economic dispatch. Xu et al. [6] proposed CCHP multi-microgrids to solve the problem of economic dispatch according to load types. Reference [30], applied an optimal fuzzy P-graph to optimize cogeneration plant. Also, a modified fuzzy has applied to assess a group of CHPs [28]. Reference [31] proposed optimal demand side management to interconnect between TES and solar PV to provide cold demand while reducing electricity usage. Kim et al. [1] developed an optimal neural network (NN) based GA of economic dispatch for CHP plant. NN toolbox is used to train the dataset CHP models while GA is optimized output parameters.
In study by [32] presented a multi-PSO technique for a campus energy plant to reduce thermal consumption and operation cost. An optimal model has developed for CCHP combining with PV system. It is used to minimize energy usage and operation cost based following electrical and thermal loads [33]. In [34], a model has formulated for optimizing CHP and scheduling its operation strategy. The proposed model considered CHP sizing which it includes PV where it uses as a source for managing CHP plant, while the rest of the models did not include TES and BES systems. An optimal dispatch model is proposed for CCHP for power interaction based on the use of renewable energy resources [6]. Lin et al. [27] introduced an analytical method to solve multi-energy trading problems between CCHP microgrid and consumer based on Game Theory method. Chu et al. [35], a non-linear programming (NLP) model is developed for the CCHP system to be solved by PSO. The objective of PSO was to solve the problem of the CCHP energy dispatch. Reference [36] suggested an optimal strategy to manage cooling, thermal, and electrical loads for building microgrids based on external energy network. The work used data sets to assess and validate the model performance for summer and winter seasons. Urbanucci et al. [3] developed a new optimization formula to simulate dynamically the operation of the TES for VOLUME 11, 2023 74389 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.
CHP plant. The TES is used two levels GA and MILP. These algorithms are used to optimize TES operation conditions. Reference [37], with the use of MATLAB, a model has been developed to simulate the integrated cooling and heating energy system, and it applied on different load (heating and cooling) conditions.
To sum up: (1) if the management strategy does not follow optimum operation for CHP sizing, it will not take full advantage to reduce energy consumption and polluting emissions. (2) system management also lies in the oversized complicates because large CHPs fluctuate in the demand [9]. Therefore, small CHP sizing plants operating with low primary energy and emission savings while large CHPs sizing cause large heat wasted into the atmosphere [9]. In the same time, micro-CHPs sizing offer high reliability and robustness. Nevertheless, small CHP sizing units cannot respond to increased demands even when operating at full load rated power [7], [12]. Despite CHPs oversizing (≥ 100 kW & ≤ 5MW [38]) drawbacks, but they are widely used the most among other sizes especially in commercial units [7], [39]. Therefore, CHPs sizing and operating strategy are responsible for economical performance, if they optimize for instance by GA, MILP and ABC algorithms [3], [12]. From an economic point of view, integrated optimal sizing and management along with hybrid RES can help reach a more environmental-friendly. Then it can offer saving-energy with high reliability and better sustainability which are a good choice [7], [11].
Some of literatures focused on the economic dispatch for CHP. Some studies proposed integrating renewable (PV or WT) along with optimum CHPs sizing. Other studies suggested combining CHP with TES systems. Moreover, some literatures considered BES system along with CHP and PVs system. In this paper; optimum CHP with RES including GTG, PV, BES, TES and boiler systems are interacted. Model system suggested and integrated with weather data and then a trend optimization algorithm was used to keep electricity and cooling demand balanced while minimizing steam production.
The contribution of this paper includes on: 1) Propose data modeling based on actual data that interacts with plant equipments to investigate the performance of the CHP-PV plants. 2) Propose data model which it performs to assess the load demand for electricity and cooling energy without/with PVs plus BES system. 3) Propose a hybrid fuzzy subtractive clustering and APSO algorithm. The hybrid algorithm tuned with proportional integral (PI) controller. 4) Accuracy and MSE perform to evaluate performance of the hybrid algorithm and comparative study is made compromising load demand.
The rest paper organizes as follows; system description introduces in section II. Next, the developed model and hybrid algorithm are presented in methodology with section III.
Section IV discusses the results which obtained by the methodology implementation. Finally, this manuscript is concluded in Section V

II. SYSTEM DESCRIPTION
The first in this section introduces system description of actual CHP over-sizing plant which it takes place in UTP campus. While the other part is presented the proposed method used to simulate CHP's overall performance based-on plant historical data.
A. BASELINE SYSTEM STRATEGY 1) CASE STUDY CHP plant or cogeneration in UTP is comprised of two solar gas turbine generators (GTG) with 4.2 MW each, two steam absorption chillers (SACs) with 1250 RT of chilled-water each, 4 electric chillers (ECs) with 325 RT of chilled-water each), 2 heat recovery steam generators (HRSG) with 12 Ton/h each), thermal energy system (TES) with 10,000 RT of chilled-water capacity and 1 boiler with 6 Ton/h). The 4ECs were designed to charge TES when demand is low using electricity supplied from Tenaga Nasional Berhad (TNB). This imparts an advantage of lowering night tariff offered by TNB. A few studies have been proposed for UTP plant; In [40],a control system was designed to manage the chilling operation at UTP using fuzzy logic toolbox. Reference [41], developed a control model to integrate gas district cooling (GDC) and Data Center. A scheduling algorithm based primary-secondary clustered architecture based on three slave clusters was implemented to reduce the problem of chilled water supply-demand gap in the university campus [42]. An investigation study was done to evaluate ECs cooling performance in UTP [43]. Reference [44] has modelled energy and used a numerical algorithm to reduce electricity usage while kept demand balanced [44]. This study is extended for the investigation of UTP plant which it is integration for keeping load demand for electricity and cooling satisfied. This work investigates whole system equipment integrated with photovoltaic (PV) and storage systems including BES and TES. We used baseline system and proposed models in the following section, and Fig. 1 shows CHP equipment and its capacity.
The CHP plant delivers two types of energies to serve UTP campus; electrical energy and cooling. The first energy is generated from 2GTGs and other energy is used to produce cooling load capacity. The cooling load is produced by electrical supply for ECs and steam supply for SACs.

2) GAS TURBINE GENERATOR MODEL
In deriving parameters of the GTG model, the compressor's efficiency, combustor, and turbine; the specific heat of the air and exhaust gas; and the lower heating value (LHV) of the fuel (natural gas) are assumed to be constant. The GTG net  power and efficiency are defined as follows [1], where P n,GTG : GTG net power, η con : conversion efficiency, η GTG : GTG efficiency, W F,GTG : gas turbine fuel flow, η mec : mechanical efficiency, P tur : power turbine, P tur : shaft power, P com : compressor power. The process variables at nominal operating conditions are summarized in Table 1.

3) HRSG SYSTEM MODEL
The increase in air density is accomplished by means of evaporating water into the inlet air and this decreases air temperature. By decreasing air temperature, fuel gas consumption will decrease with the same GTG output power and hence increase the overall CHP efficiency. The water vapor will pass through the gas turbine, and this water must control in order to not carry into the compressor blades to cause a serious problem. There is a direct relation between mass flow gas that enters the gas turbine and the mass of ambient air supplied to the combustion chamber through compressors. The exhaust gas from turbine is sent to the heat recovery steam generator (HRSG), the HRSG is used to convert water into superheated steam. Here, only 66.6% of the heat from exhaust GTG is captured by HRSG to produce steam, while remaining 33.4% is emitted to the environment [45], [47]. The captured heat (66.6%) by HRSG is sent to the SACs to supply and provide the cooling to UTP facilities. Then, if the produced steam is not enough for SACs, an auxiliary gas boiler (AGB) can be used to; either to produce additional steam, or to increase the steam temperature when the exhaust GTG is not hot enough. The electric power output of the GTG is 4.2 MW, when the temperature of the exhaust gas from the HRSG about 400 • C. Therefore, the output of steam power (Q STM ) by HRSG can be written [6], where Q STM : GTG's steam energy, η THR_GTG (84.55%) and η Ele_GTG (39.4%) are the thermal efficiency and electrical efficiency of GTG, respectively, η WHR (3.54%) is the waste heat recovery, F GTG : GTG's fuel gas consumption of GTG, Q AGB : AGB's thermal energy, η AGB : boiler's efficiency, and F AGB : AGB's fuel gas consumption. The boiler (thermal) and overall CHP efficiencies of UTP plant are written in Eqs. (7)(8), as shown at the bottom of the next page, and the parameter's value are given in Table 2.

4) PV SYSTEM MODEL
The daily electric energy that is produced by the PV can be calculated by Eq. (9) [48].
room temperature at STC, η INV and η WIR : are inverter and wire efficiencies, respectively. The calculated PVs and their parameters for this work are given in Table 3.
The extra/excess produced energy from GTG and PV is used to be stored in BES to be used during peak hour's demand. This extra energy can be expressed by, From Eq. (10), the extra produced energy that comes out from (PV & GTG) is used to design the capacity of the battery bank (P BATT ) which can be calculated as, The BES considered in this work was a VRFB due to its reliability and efficiency. It consists of a battery with a rated power charger and discharger of 400 kW. The BES chosen has a nominal capacity of 900 kWh and system efficiency of 96% [44].
For l th battery (l = 1, 2, . . . , n), the operation condition mode (γ ) of battery can be written [50], where Q BES and C SIZ are battery charge mode and its capacity, respectively. When the battery is fully charged Q BES = C SIZ and γ l = 1, and therefore the battery is become ready to discharge energy (P BES ). The energy condition can be expressed in Eq. (14).  [43], [51]. UTP demand for electricity must commit to the state of charge (SOC) according to,

5) CHILLER'S SYSTEM MODEL
The SAC is produced the chilled water and sent to the building for cooling comfort. It consists of high stage generator (HSG), low stage generator (LSG), condenser, absorber, and evaporator. The steam that produces from HRSG is fed to HSG and refrigerant vapors produced is led to LSG. The absorber extracts with the fuel of 4875 KG from evaporator to produce a chilled water of 1250 RT at supply temperature of 6 • C to be used in the buildings. Therefore, the production of SAC for the amount of cooling depends on two inputs which exhaust heat and the exhaust heat temperature [2]. While ECs use electrical supply to produce cooling load-based chilled water flow rate and its temperature. To calculate cooling, technical data are given in Table (4)-(5) for chillers at nominal conditions. the SAC and EC cooling load in (RT) of kth chillers of UTP campus is determined by [43], where c p : specific heat of chilled-water (4.197 J.kg −1 .K −1 ), m CHW : flow rate of chilled water, T CHWS and T CHWR : supply and return temperatures of chilled-water. The efficiency in the chillers plant is termed as the COP expressed as, The cooling demand (Q UTP ) which is used to cater to the requirements of UTP buildings can be expressed as, Let Q SAC + Q EC = Q CHI , and Q TES is cooling load capacity by 4 ECs to be charged between a period from 10pm-7am, Let T CHWS_EC -T CHWS_EC = T TES , and n hrs is the number of charging hours which is 10 hours and 38 minutes (10h: 38m). Thus, 4 ECs are supplied from the TNB and operated partially and fully at basis 24h.

7) TES OPERATION CONTROL
TES is a technology that can be used for energy efficiency enhancement [52]. The target and benefit of TES used is to shift the campus cooling load from peak hours to off-peak hours. The major impact of designing a TES at UTP is to conserve energy without compromising load demand. The TES operation mode can be expressed as, When the supplied chilled-water (m CHW ) is greater than the consumed chilled-water to the campus (m CAM ), the difference (m CHW -m CAM ) will be charged to the TES tank by ECs. Let, m TES = m CHW -m CAM . At operation mode, EC is started to chill down the water in the TES to increase the calorie to keep the TES storage water at 6 • C. Hence, the dynamic equations for charging chilled-water to the TES can be written [53], [54], If the supplied chilled-water (m CHW ) is lower than the consumed chilled-water (m CAM ) to the UTP buildings, the TES tank will discharge the chilled-water to the campus to maintain the campus demand. Thus, the dynamic equations for the discharging chilled-water of TES can express as; (26) where T 14 (T CHWS_EC ) and T 1 (T CHWR_EC ) are chilled-water supply and return temperatures from the campus buildings. At UTP campus, during the off-peak period from 22:00-23:00 and from 00:00-07:00, the operation mode λ = 1. The chilled water production by chillers is enough to cater the requirements of the campus load demand. The ECs will be operated to store the chilled water at 5 • C in TES tank. The target is to chill down water inside the TES tank to keep the TES storage water at 6 • C during discharging mode. In TES, there are 14 number of temperature sensors of resistance temperature detectors (RTD) installed inside TES tank (T1 ∼ T14) as shown in Figure 2.
Two of RTD located at the top portion of the tank (default T1 and T2), and they are used to control the ECs start/stop sequence interlock during charging mode.
The TES tank discharging mode is occurred, when the demand is high during daytime (08:00-21:00). Sensors T13 and T14 are set for discharging mode sequence interlock. When T13 < 5.8 • C, ECs shutdown sequence is started and the distributed control system (DCS) is prompt a message to the operator to inform that the TES tank is ready for discharging process. When T14 > 6 • C, ECs ''startup sequence'' will start and discharging process continues until another message is sent to stop charging TES tank. They are operated from 00:00 -07:00 and 22:00 -23:00 to charge the TES tank to be used during peak hours from 08:00 -21:00 to maintain campus demand. The TES is adopted the priority control strategy that is used to correct the mismatch between the supply and cooling demand according to the temperature set points of chilled water. The use of TES provides an effective way to take advantage of low electricity rates in effect during the night and other off-peak periods. Figure 3 (a) shows T 1 ∼T 14 of actual TES based on RTD, where these temperatures have a direct effect on the cooling performance of TES tank. The temperatures for a month 12 of 2016 have collected and classified by simulating into 6 clusters and each cluster/group has a center to represent it in Fig. 3 (b). The data (1, 2, 3, 4, 5, and 6) represent cluster (1, 2, 3, 4, 5, and 6) and data 7 represents center 7 of each cluster.   To optimize campus cooling load demand, an estimation cooling factor can be denoted by, Thus, the optimal cooling consumed in UTP campus by m th chillers plus TES condition can be expressed in Eq. (28) Table 6 gives TES cooling capacity with 10,000 RTh and the chilled-water flow and return temperatures from campus (T 1 ) to chill again with supply temperature (T 14 ). Therefore, overall system models can be summarized in Table 7.

A. OVERVIEW
We have considered two schematic diagrams for CHP over-sizing plant to manage UTP campus energy as shown in Fig. 4. Figure 4 (a) shows the current plant where it comprises of 2 GTGs, 2 SACs, 1 AGB, 4 ECs, and 1 TES system. The CHP plant operates manually using fuel gas to produce electric power and steam 2GTGs. The productive through steam through HRSG is used to produce chilled-water systems. UTP recently has deficit in energy, to keep electrical supply and demand balanced, extra energy export from TNB grid. In the plant, there is no optimal control due to uncertainty of the deficit and excessive situation. The operation strategy is, running 2GTGs during peak hours and 1GTG at off-peak hours, and both GTGs are operated partially at 83.3%. Figure 4 (b) shows the proposed microgrid plant where it consists of the current equipments in addition to 325 PVs/cells and 8 BESs with a capacity of 900 kW each. The main goal of integrating CHP-PV is; to reduce fuel gas consumption and handling the deficit/extra energy problem using BES. Also, the structural propose assists in reducing steam and maintaining electricity and cooling demand.  Table 8.
For optimal energy and cooling consumption by GTGs and chillers, an ambient temperature (T AMB ) is considered. For maximum energy production by PVs, irradiance (G rr ), cloudiness factor (l rr ) are also considered. They have taken from https://www.worldweatheronline.com/ipoh-weather-history /perak/my.aspx. The reason of T AMB , is that it has an impact on system performance [2]. The data is used to quantify inputs in order to evaluate 4 models of UTP plant. The investigation models are; (1) output power of GTGs (P n,GTG ), (2) steam production (Q STM ), (3) energy production by PVs (P PV ), and (4) chilled-water by TES (Q TES ).

2) MODEL FORMULATION
Four models have been formulated; Model 1 is used to calculate the productive power by GTG. Model 2 is used to assess the steam production. Model 3 is used directly to produce extra energy by PVs. Model 4 is used to maintain cooling demand using TES. All models data are collected from UTP plant to investigate the performance of some plant equipments. Figure 5 shows the model inputs data.
• Model#1: when the ambient temperature is hot, fuel consumption of GTG increases which it affects on the overall performance of the GTG [1]. Herein, we are considered three variables data q' 1 , q' 2 , q' 3 (W F,GTG , T Ex,GTG , W AIR ) to simulate the output power of GTG (P n,GTG ). Where W AIR is the actual dry-air mass flow to the gas turbine.
• Model#2: In model#1, there are two outputs P n,GTG and T Ex,GTG . Where the P n,GTG production depends upon how much GTG fuel gas consumes, and consequently the more fuel, the more thermal energy. The 74394 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.    design T Ex,GTG is 510 • C, it enters to the HRSG, then temperature of the exhaust gas from the HRSG denoted by (T ex,HRSG ) which is 400 • C. Herein, we developed steam energy (Q STM ) based on q'' 1 , q'' 2 , q'' 3 (T Ex,GTG , W Fu ,T Ex,HRSG ), where W F : input flue gas to 2 HRSGs and 1 boiler. The production of steam by 2 HRSGs and boiler is used for the chilled water production by 2 SACs. The boiler in CHP at UTP plant is used as an auxiliary system if the steam production by 2 HRSGs is not enough to operate 2 SACs. VOLUME 11, 2023 74395 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply. • Model#3: A study by [56], has developed output PV power considered two input variables; irradiation and and T AMB , when T AMB is hot, the PV output energy and efficiency decrease. Besides the outdoor temperature and irradiation in the PVs model, we add additional coefficient which is ''cloudiness''. When cloudiness ratio increases the amount of irradiation decreases, then it will affect and reduce PV's efficiency and energy, respectively. Herein, we considered three variables data; solar irradiation (G rr ), cloudiness ratio (l rr ,), and ambient temperature (T AMB ). The q''' 1 , q''' 2 , q''' 3 (G rr , l rr , T AMB ) are used to simulate the output energy of (P PV ) and then to take advantage of high TNB tariff during peak hours. PV and BES systems are used to supply four ECs for charging/discharging TES chilled water, in addition to their participation in campus buildings.
• Model#4: The steam production by HRSGs and AGB is used produce chilled water by 2SACs to serve campus cooling facilities. The models of the SACs and ECs are considered from baseline system. The cooling comfort have a direct effect due to chilled water flow rate and its temperature [54]. When the T AMB is hot and very humid, chilled TES temperature (T TES ) increases. The increment in T TES will result in consuming power and lowering COP [43], [53], [54], [54]. The cooling demand (Q TES ) model by TES is influenced with two variables q'''' 1 , q'''' 2 , q'''' 3 (m TES , T TES , T AMB ).

3) PROBLEM FORMULATION
The overall objective is to obtain an optimal strategy for a CHP-PV-TES to minimize the total electricity consumption while satisfying campus cooling demand. In this paper, we formulate FSC problem to search for the set points from the collected data system. FSC-based APSO algorithm is adopted to re-assess the optimal datasets of CHP-PV plant. The objective functions in Eqs. (29)(30)(31), as shown at the bottom of the next page, are used to achieve paper's goal after applying optimal clustering datasets.

C. PROPOSED HYBRID FSC-APSO ALGORITHM
This section describes in detail the proposed algorithm. Figure 6 shows the main steps to classify data to assess and evaluate cooling load, steam and electricity production. Firstly, it starts with the identification of input data and the number of clusters and then converting to fuzzy rules with FSC system. Secondly, the FSC is optimized by APSO to enhance its performance. Finally, the outcomes of the FSC-APSO algorithm applied in developed models to simulate the performance of CHP plant. Then, the obtained result is validated by actual data.

1) DATA IDENTIFICATION
As indicated in ''Data Collection'' section, 12×8670 datasets q = q´, q´´, q´´´, q´´´´are classified and grouped to select optimum points. The selected data are integrated as given in Table 8. Integrating data used to quantify inputs-outputs in order to evaluate CHP model as indicated in proposed model section after applying optimal data. To obtain optimal data, three techniques are considered; (1) clustering by FSC algorithm, (2) optimizing FSC by APSO algorithm, (3) clustering radius with tuner PI controller.

2) FUZZY SUBTRACTIVE CLUSTERING
Fuzzy subtractive clustering (FSC) system is one of the fuzzy applications and recently has become one of the dominant techniques applied to classify data. It can be applied in different applications to explore different tasks. It is used to estimate the data clustering numbers by searching the collected data with specific structures to select the best point among data points [43]. Let, x i = q 1 , q 2 ,., q n represent datasets and each has the potential to be cluster center. Thus, the density of datasets x i can be expressed as, where r a : cluster neighborhood radius and ||x l -x j ||: Euclidean distance. The r a has an impact on potential cluster centers.
Here, the datasets have a high density value. Therefore, the 1 st cluster center x CC1 is chosen as dataset with the highest density value y CC1 . In the next cluster center, the density of datasets of each x i is defined as, where r b : neighborhood radius; a positive constant reducing density measure; r b = 1.5r a . The number of clusters defines the number of rules of the FSC. Let x C to be found in the clustering group of y i , this cluster provides an if-then-fuzzy 74396 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.  expression in Eq. (34).
where x i is the l th input feature and µ ij is the membership function (MF) in the rule associated with the l th input. The MF of the fuzzy set has a degree ranging from 0 to 1 and has grades of inputs with the Gaussian function, which is expressed by, where ρ = 4/r 2 a , and σ 2 = 1/2ρ, x Cj : data mean/center of k th cluster, and σ i is the standard deviation of each MF [43].
Each input has a degree of participation in every fuzzy set, based quantitatively on MF. Figure 7 depicts fuzzy sets and each input data has six fuzzy rules. The fuzzy rules have different value for each input; very low (VL), low (L), medium low (ML), medium (M), medium high (MH), and high (H) value. When FSC is performed, the consequence of the fuzzy rule with the highest degree of fulfilment is selected to be the required output class (centroid of each fuzzy rule).
Here FSC extracts a set of rules that models the data behavior by finding potential centers in the dataset given a cluster radius (r a ). The cluster radius defines the range of the search for clusters in a dataset. For instance, the six of clusters in Fig 3(b) defines the 6 of fuzzy-rules of the FSC as shown Fig. 7. Each cluster has a center and the standard deviations according to the Gaussian MF. The combination of dataset of MF creates a rule where it can use to develop a linear data of each model.

3) INTEGRATING FSC-APSO ALGORITHM
As a result, FSC method having limited in determining the number of clusters and also when implementing large data [43], [57]. The APSO algorithm is suggested to overcome the limitations of FSC system [57]. The fuzzy sets of inputs have six membership functions (MFs) optimized by FSC-APSO-based tuner PI system. The structural of integrating FSC-APSO is shown in Fig. 8.

a: INITIALIZING APSO PARAMETERS
First, initialize each parameter and encoded into particles N p (i = 1, 2, . . . , 50). Second, the inputs are partitioned, and each value is chosen randomly within its respective range after calculating the potential data density in Eq. (33). Then, the vectors are assigned based on particle velocity (V ij ) and its position (λ ij ) [57], [58], where Gbest is the global for best solution so far for all particle Np, r ij is the random number (0, 1), α and β are the acceleration parameters typically ≤ 1. The position of particle (λ) shows the fuzzy relation from set of datasets, x i = (x 1 , x 2 ,. . . , x n ), to set of cluster centers, x j = (x CC1 , x CC2 ,. . . , x CCk ). From fuzzy sets in figure (7)  The positions of MFs are re-locating after optimizing by APSO for 6 cluster centers with Np = 50. The No. of cluster centers can be increased or decreased by minimizing or maximizing the values of cluster radius. Also, cluster radius has a direct impact on cluster centers [57].

b: TUNER PI CONTROLLER
From Eq. (33), the cluster radius is selected from α and β as [57], where the cluster radius defines the range of the search for clusters in a dataset. Studies in [59], and [60] show that the clustering radiuses from 0.3 to 0.8 give better FSC accuracy and optimal cluster centres. Eq. (39) did not achieve the ranged values because of random (0, 1) parameters. Here, a PI control model is suggested to tune the radius of clusters to be ranged between 0.3 and 0.8. Figure 9 shows a PI control model for adjusting cluster's radius.
The PI control model for tuning cluster radius (r a ) and neighborhood radius (r b ), expresses in Eqs (40)(41), as shown at the bottom of page 12. where s is the Laplace transform, k p is the proportional gain, k i is the integral gain. In this case, Ø = 2.1, k p = 1.5, k i = varies from 0.05 to 0.15, and Each cluster center point influenced by r a and each particle i generates membership grades of inputs. The center point of each MF is chosen with the goal of distributing the MFs throughout the whole range. Thus, the range is partitioned and then each center value is chosen randomly within its respective partition. So, the 'λ ij ' of x CCi assigned based on MF partition, and 'r a ' of each input is assigned as position vectors as, The r a value is updated by the change of r a ( r a ) with the accelerated particle and therefore, the 'r a ' of each input is assigned as position vectors as, λ i = {λ i1 , λ i2 , . . . , λ iN P } = r ai1∓ r ai1 ,r ai2∓ r ai2 . . . ,r ain∓ r ain (44) where; r aij is used to tune r a that is resulted in optimal data points. The evaluation for particle's vectors λ i is calculated according to Eq. (45). The swarm (N p = 50) of each particle i at iteration τ =200 for cluster sets express as, (45), shown at the bottom of page 13. Since all FSC cluster centres can be further tuned by APSO based on the influencer 'r a ', the initial 'r a ' is simply set between 0.4 -0.7. The change in cluster radius in Eq. (44) varies in x CCij value accordingly. For each iteration τ , a new population is produced, and the old population is replaced to be stored. The new and old populations are compared for each other and the initial positions are created arbitrary according to the all best (gbest) performance. The values of the particle velocity are generated randomly by v i (0), i = 1, 2, 3, . . . , 50. Then, from Eq. (45), the best 'λ' for the 'r a ' of all particles (gbest) are calculated from Eqs. (36) and (37).

c: IMPLEMENTATION OF SIMULATION DETAILS
The simulation of FSC-APSO algorithm-based PI controller shown with the flow chart in Fig. 10.
The implementation of the FSC-APSO algorithm with the general steps procedure as follows: 1) Upload: upload data from the selected datasets, and then assign 70% for training data setup and 30% for testing data setup. Next, partition the input datasets into 74398 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.   clusters and derive fuzzy. Next, initialize the radius of clusters, swarm, alpha, beta; and create swarm matrix for particle's velocity, positions, and Gbest. 2) FSC: identify the number of clusters and set the density of data. Then, calculate the first cluster center to last cluster center and update fuzzy-rule based on the MF. The cluster center point of each MF is chosen with the goal of distributing the MFs throughout the whole range. The range is partitioned and each center value is chosen randomly within its respective partition. 3) APSO: set the value of clusters radius initially from data density. Next, create fitness function-based MSE and calculate Gbest for the swarm and update the velocity and the position of each particle. 4) FSC-APSO: if FSC and APSO conditions are not met go to steps (2) and (3) update MF and re-calculate the radius, and then calculate Gbest for the swarm and then update the velocity and the position of each particle. 5) FSC-APSO-based PI: since all FSC cluster centers can be further tuned by PI through APSO. If the cluster centers shifted in the search space seeking for the VOLUME 11, 2023 74399 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply. best values and therefore the radius varies accordingly. The cluster centers are created and the evaluation for particle's vectors is calculated. For each iteration. The values of the particle velocity are generated randomly and the best positions for the radius of all particles are calculated. If FSC-APSO condition is not met go to step (4) and update the radius, and then calculate Gbest and update particle's velocity and position. Otherwise, go to data density and start the process again until getting best the features of clusters. The proposed algorithms were simulated with clustering data-sets with pre-defined parameters given in Table 9.

D. PERFORMANCE OF FSC-APSO ALGORITHM
For each cluster with operating conditions having similar characteristics, the MSE as fitness calculated in the analysis section. The determination number of clusters and big data are curial and difficult to be determined for analytical clustering methods. Despite these challenges, the FSC-APSO proved that it could be overcome in the case of big data, while the number of clusters is decided by trial and error. The proper number of clusters is dependent on data's size and the value of each single data point. There is no a certain number for optimal clustering (grouping) data.
The 2GTGs simulated under current conditions using basic and proposed models. Under the same operation conditions (83.3%) for UTP plant, where plant production maintained cooling demand, and kept electrical supply balanced with corporation between TNB and UTP (import & export). This produced high steam and consumed more fuel gas. When the operating condition sets at 77.3% instead of 83.3%, the energy, fuel gas consumption, and steam are reduced, while satisfying cooling demand. To achieve this, TES is used during off-peak hours, and in addition, 2 ECs operate during day-time alongside 2 SACs. Moreover, to produce the same capacity of chilled water by 2 SACs, AGB is ON alongside 2 HRSGs to compensate the reduction in steam. Table 10 gives actual and simulated results for the basic and proposed models.

1) MODEL 1: GTG ENERGY RESULTS
The energy model of 2 generators has simulated with an operating condition of 0.773. The obtained results were 6144 kW by the basic model. Under same conditions, the proposed models produced 6399 kW, 6405 kW, and 6439 kWh under optimal data based on scenario 1, scenario 2, and scenario 3, respectively. Neither proposal nor basic models could keep demand satisfied. To keep load demand balanced, a PV model integrates.

2) MODEL 2: STEAM PRODUCTION RESULTS
Steam production reduced from (130 to 100.5) tons, when the both 2GTGs were set at 77.3%. This setting condition consumed less fuel gas and reduced CO2 emission while cannot produce enough cooling load capacity. To maintain that, AGB turns ON to add additional 6 tons and it becomes 106.5 tons. Here 2SACs can produce until 2082.5 RT after this steam addition.

3) MODEL 3: PV ENERGY RESULTS
The PVs system is given priority to offer supplying load to compensate the deficit's energy. Thus, PVs produce direct generation of 717.1 kW and 711.7 kW when it simulated by the proposed and basic models, respectively. The PVs calculation was based on the average cloudiness ratio which (0.32+0.79)/2. The 325 of PVs approximately produce between 145 kWh to 1300 kWh according to the sun radiation/brightness and cloudiness factor. Therefore, after installing 325 PVs/cell and it contributes in UTP demand UTP, the deficit energy problem is over. The excess energy can export or store in 10 BESs with a capacity of charge 900 kWh each [44].

4) MODEL 4: TES COOLING RESULTS
Using basic models, the TES was charged 10027.5 RTh during off-peak hours from 22:00-23:00 and from 00:00-07:00 to discharge 1002.8 RTh every hour for 10 hours during peak hours. Meanwhile, same optimal clustering data applied in proposed models using 3 algorithms. The simulation was carried out and the performance of TES was charged 10491.5 RTh and TES for 10 hrs using scenario 1. The optimal data obtained with scenario 1 and scenario 2 are simulated and charged with 1049.11 RTh and 1047.84 RTh every hour for 10 hours, respectively. The charging chilled water discharges and it participates with (2SACs = 2 × 1041.25 RT and 2ECs = 2×270.73 RT) to meet the maximum demand of 4000 RT. As a result, there is a little shortage in RT estimated with 373.3 RT, meanwhile energy consumption and steam were decreased. Energy by PVs system was used to supply 4 ECs during peak hours, while BES was used to discharge TES chilled-water at peak hours and participated in UTP load.
B. DISCUSSION Figure 12 shows 24h daily loads profile of the UTP plant production. Electricity load by two GTGs under the current condition is less than the UTP campus demand, so it imports electricity from TNB grid. The simulation was carried out following the current operation strategy.GTG1 and GTG2 are ON during day-time, and either GTG1 or GTG2 74402 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.  The average of UTP maximum demands about 7000 kW, which sometimes, UTP purchases electricity from TNB. To keep load demand balanced, Fig. 12 shows the interaction energy between CHP microgrid and PVs system. The PVs system was simulated and carried out for both basic and proposed models. The PVs system is given priority to offer supplying UTP load with an average of 10h × (717.1 kW and 711.7 kW) for proposed and basic models.
In case of extra energy, BES system is provided and its constraints are set with SOC min = 30 % and with SOC max = 80 %. The capacity of each BES reaches until 720 kW with SOC max (100 % of 1 BES = 900 kW). Thus, it charges simultaneously 7200 kWh to be discharged to participate in UTP loads alongside with PVs and 2GTGs. Then, any further extra/surplus electrical energy is exported to TNB in order to prevent any over-load charging. In simulation study, it charges 7171 kWh. This charging energy needs 10 storages batteries (BES) due to SOC constraints until it is full in the event of availability of sun. Take into accounts the output energy of (GTG+PV+BES) is more than UTP load demand. In Fig. 11, the actual electricity was 126398 kWh, while the simulated basic and proposed models were 121108 kWh and 121575 kWh, respectively. From the simulation results, it's obvious the deficit energy estimated with 5290 kWh and 4823 kWh. Thus, 2GTGs integrated with PVs, where UTP load demand has become 121108 + 7117 = 128225 kWh for the basic models, and 121575 + 7171 = 128746 kWh for the proposed models. Also in Fig. 12, the actual steam produced from steam generator was 129370 KG (129.37 tons). The simulation was carried out under operation mode condition of 77.3 %. The condition consumed fuel gas 2 × 0.2461kgs (1772kGh) instead to 2 × 0.26kgs (1872kGh). In addition the steam production was reduced from 129.37 tons to 106.5 tons when AGB turns ON to add additional 6 tons which to be set at (83.3%). The reduction in steam will result in minimum CO2 emission. Here the production of chilled water by 2 SACs (2 × 1041.25) RT and 2 ECs (2 × 325) RT cannot keep cooling satisfied during day-time. Hence, to keep cooling, 4 ECs must be operated at 84% to charge about 1093.9 × 10h (10939 RT) in TES at off-peak hours. This stored chilled water discharges during peak hours. Then, the total cooling becomes 2 × 1041.25+650+1093.9 = 3826 RT.

1) COMPARATIVE STUDIES
From results & discussion, it is obvious that the simulation of CHP proposed models with integrated 325 PVs + 10 BES system performed with minimum fuel consumption. This hybrid system met the requirements of UTP demand for electricity and cooling energy. Tables (11)(12) are given the findings obtained with the basic and proposed models.

2) COMPARATIVE ANALYSIS
As indicated in methodology overview, the existing system of CHP micro-plant has been studied and investigated. The actual data, basic, and proposed are summarized as:   1) The basic model without PVs assessed with optimal data, electricity production and cooling didn't meet the demand of UTP campus. But when it was integrated with PVs, the model kept electricity balanced and saved 8944 kWh and reduced steam with 5512 kG (5.5 tons), while it failed to meet cooling (see Table 11).
2) The proposed model without integrating PVs assessed with optimal data, steam and cooling production met, while energy production was not able to meet supply-demand criteria. When the proposed model integrated with PVs, electricity, cooling, and steam met the requirements of university campus. The model saved 9547 kWh and reduced steam with 6110 KG (6.1 tons) without compromising cooling demand (see Table 12).

V. CONCLUSION
This paper introduced an integrated model for large CHP plant to evaluate campus energy at UTP. From comparative results of investigating UTP plant, it found out the current operation strategy needs to reset. The study demonstrated the effectiveness of proposed CHP models integrated with PV, BES, TES models and it concludes the following: • Basic model without PVs summarizes; electricity and cooling energy are not kept balanced, while producing high steam (it means high heat and high CO2 emission). But when basic model integrated with PVs; (i) electricity is kept balanced and it can save 6.61 %; (2) steam production reduces 4.26 % of the total steam production; (3) cooling energy is not met load demand.
• Proposed model without PVs summarizes; electricity is not kept balanced, while satisfying cooling energy and reducing steam production (it means low heat and low CO2 emission) due to optimum CHP sizing. But when the proposed CHP models integrated along with GTG, PVs, BES, and TES; (i) electricity is kept balanced and it save 7.0 % of the total daily production; (2) steam can reduce 4.72 % of the total daily production; (3) cooling energy is kept cooling load demand satisfied. • As a future work, this proposed method will be carried out to examine CHP's cost operating strategy to assess its economical performance. Also a new CO2 modelbased optimal neuro-fuzzy will develop to reduce its emission based on paper current models of GTG and HRSG alongside AGB system. Future work will also include comparative studies and, analysis, and verification.

APPENDIX A
This section is described the optimal outcomes of dataset clustering in Table 13 using FSC-APSO denoted as scenario 1, APSO denoted as scenario 2, and FSC algorithm denoted as scenario 3. We have selected the fifth clusters center from the cluster number 6.