Indonesian J our nal of Electrical Engineering and Computer Science V ol. 43, No. 2, August 2026, pp. 672 ∼ 682 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v43.i2.pp672-682 ❒ 672 Condence-dri v en adapti v e operating-point optimization f or photo v oltaic systems under partial shading conditions Sarah Kawther Sedjar 1 , Mourad Benmessaoud 2 1 Laboratory of Automation and Systems Analysis (LAAS), Department of Electrical Engineering, National Polytechnic School of Oran Maurice Audin (ENPO-MA), Oran, Algeria 2 Simulation, Control, Analysis and Maintenance of Electrical Netw orks (SCAMRE), Department of Electrical Engineering, National Polytechnic School of Oran Maurice Audin (ENPO-MA), Oran, Algeria Article Inf o Article history: Recei v ed Jun 10, 2026 Re vised Jul 14, 2026 Accepted Jul 23, 2026 K eyw ords: Adapti v e e xploration control Condence-re gulated optimization Embedded photo v oltaic control Global maximum po wer point tracking Prediction-guided MPPT Search-space contraction ABSTRA CT P artial shading conditions (PSC) generate highly nonlinear multi -peak photo- v oltaic (PV) characteristics, complicating reliable global maximum po wer point tracking (GMPP). Although numerous intelligent optimization techniques e x- ist, most rely on e xtensi v e e xploration mechani sms that limit their applicability in embedded real-time controllers. This paper introduces a condence-dri v en adapti v e maximum po wer point tracking (MPP T) frame w ork in which the opti- mization search space is dynamically re gulated according to the reliability of a predicti v e operating-re gion estimator . Unlik e standard articial neural netw ork (ANN)-assisted M PPT strate gies that of fer only po wer prediction, t he proposed approach e xploits a condence inde x to continuously contract or e xpand the e xploration domain, minimizing search ef fort while preserving global tracking capability . The frame w ork w as de v eloped using real measurements from the PV - D A Q database and v alidated through a nonlinear tw o-diode thermal-electrical PV model incorporating irradiance mismatc h and temperature-dependent ef- fects. Static and dynamic PSC scenarios were in v estig ated to e v aluate con v er - gence beha vior and computational performance. Experimental re sults demon- strate that the proposed condence-go v erned strate gy achie v es an a v erage tracking ef cienc y of 90.89%, reduces con v er gence ef fort through adapti v e search-space contraction, and matches real measurements wit h an R 2 v alue of 0.8664, of fering a lo w-comple xity solution for real-time embedded PV ener gy management. This is an open access article under the CC BY -SA license . Corresponding A uthor: Sarah Ka wther Sedjar Laboratory of Automation and Systems Analysis (LAAS), Department of Electrical Engineering National Polytechnic School of Oran Maurice Audin (ENPO-MA) Oran, Algeria Email: sarah-ka wther .sedjar@doc.enp-oran.dz 1. INTR ODUCTION Photo v oltaic (PV) ener gy systems ha v e become one of the most promising rene w able-ener gy tech- nologies for sustainable electricity generation and distrib uted po wer -generation applications [1]. Ho we v er , their performance strongly depends on en vironmental operating conditions, particularly solar irradiance and temper - ature. Among the v arious f actors af fe cting PV po wer production, partial shading conditions (PSC) remain one of the most challenging issues because the y generate highly nonlinear po wer –v oltage (P–V) characteristics J ournal homepage: http://ijeecs.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 673 containing multiple local maxima [2]. Under such conditions, reliable global maximum po wer point track- ing (GMPP) be comes considerably more dif cult, and con v entional maximum po wer point tracking (MPPT) techniques often con v er ge to w ard local operating points rather than the true global optimum [3]-[5]. T raditional MPPT methods such as perturb and observ e (P&O) and incremental conductance (IC) of fer lo w computational comple xity and simple implementation. Ne v ert heless, their performance signicantly dete- riorates under PSC because of the presence of multiple po wer peaks and rapidly changing irradiance conditions [4], [6], [7]. T o o v ercome these limitations, numerous intelligent optimization approaches ha v e been proposed, including particle sw arm optimization (PSO), gre y w olf optimization (GW O), dif ferential e v olution (DE), salp sw arm algorithm (SSA), h ybrid metaheuristics, reinforcement learning (RL), and deep reinforcement learning (DRL) [8]-[18]. These techniques generally impro v e global search capability and enhance GMPP localization accurac y under comple x operating conditions. More broadly , intelligent data-dri v en techniques ha v e increas- ingly been adopted to enhance PV monitoring, predict ion, and maximum po wer point tracking under comple x operating conditions [19]. Despite their ef fecti v eness, most intelligent MPPT strate gies rely on e xtens i v e e xploration of the PV operating domain. Lar ge search interv als, repeated tness-function e v aluations, and computationally inten- si v e training procedures often increase con v er gence ef fort and e x ecution cost, limiting their suitability for lightweight embedded PV controllers operating under rapidly v arying en vironmental conditions. Consequently , maintaining reliable GMPP localization while reducing unnecessary e xploration remains an open research chal- lenge. Recent studies ha v e also emphasized the importance of lightweight embedded MPPT architectures and lo w-comple xity AI-assisted control strate gies capable of real-time operation under dynamic en vironmental conditions [20], [21]. T o address this limitation, this paper proposes a condence-re gulated adapti v e MPPT frame w ork that dynamically adjusts the e xploration interv al according to prediction reliability . Unlik e con v entional articial neural netw ork (ANN)-assisted MPPT approaches, wh e re prediction is primarily used to estimate the operating point, the proposed frame w ork e xploits a condenc e indicator to directly re gulate the optimization domain. The search space contracts when condence is high and e xpands when uncert ainty increases, concentrating e xploration around probable high-po wer operating re gions while preserving global-search capability under PSC conditions. The mai n contrib utions of this w ork, centered on the concept of condence-re gulated sea rch-space contraction, are summarized as follo ws: − A condence-go v erned search-space contraction mechanism that dynamically re gulates the optimization domain according to prediction reliability . − A prediction-guided adapti v e optimization frame w ork that reduces redundant e xploration while preserving GMPP localization capability under PSC conditions. − Experimental v alidation using real PVD A Q measurements and a nonlinear thermal–electrical PV model incorporating irradiance mismatch, bypass-diode acti v ation, and temperature-dependent ef fects. − Demonstration of lo w-comple xity real-time suitability for embedded PV ener gy-management applications. Real photo v oltaic dataset. The proposed condence-re gulated MPPT frame w ork w as de v eloped and v alidated using real PV measurements obtained from the open ener gy data initiati v e (OEDI) PVD A Q database. The selected utility-scale PV installation pro vides elect rical and en vironmental operating v ariables, including PV po wer , v oltage, curre n t , and temperature measurements. Prior to model de v elopment, normalization, noise ltering, and operating-condition v erication were applied to impro v e data consis tenc y and reliability . The resulting dataset constitutes a realistic benchmark for e v aluating predicti v e performance and MPPT beha vior under practical outdoor operating condi tions. Figure 1 presents the measured PV po wer prole e xtracted from the PVD A Q database. PV system v alidation model. A nonlinear thermal–electrical PV model w as e mplo yed to pro vide a realistic v alidation en vironment for the proposed frame w ork. Such models are widely used for reproducing PV beha vior and parameter -identication studies under v arying operating conditions [22]. The PV array w as partitioned into multiple series-connected submodules subjected to nonuniform irradiance distrib utions, en- abling the generation of representati v e multi-peak operating conditions for e v aluating the ef fecti v eness of the proposed condence-re gulated optimization strate gy . Condence-driven adaptive oper ating-point optimization for photo voltaic ... (Sar ah Kawther Sedjar) Evaluation Warning : The document was created with Spire.PDF for Python.
674 ❒ ISSN: 2502-4752 Figure 1. Measured photo v oltaic po wer prole e xtracted from the PVD A Q database 2. V ALID A TION SCEN ARIOS UNDER P AR TIAL SHADING CONDITIONS The ef fecti v eness of the proposed condence-re gulated optimization frame w ork w as assessed under progressi v ely increasing le v els of PV operating uncertainty . The objecti v e w as not only to e v aluate GMPP tracking performance under PSC, b ut also to in v estig ate the ability of the adapti v e search-space contraction mechanism to preserv e reliable decision-making when the ambiguity of the optimization landscape increases. 2.1. Static partial shading scenarios Three representati v e operating scenarios were designed to generate distinct le v els of irradiance mis- match, multi-peak comple xity , and GMPP ambiguity . − Case 1 — Se v ere irradiance mismatch: G = [1000 , 500 , 300 , 150 , 50] W/m 2 . − Case 2 — Progressi v e irradiance gradient: G = [1000 , 800 , 600 , 400 , 200] W/m 2 . − Case 3 — High multi-peak ambiguity: G = [1000 , 1000 , 200 , 200 , 50] W/m 2 . These congurations were intentionally selected to challenge dif ferent aspects of the opti mization process. Case 1 introduces substantial po wer de gradation caused by strong irradiance mismatch. Case 2 repre- sents a smoother b ut still nonuniform operating en vironment. Case 3 constitutes the most demanding scenario, producing multiple closely spaced local maxima that increase the probability of incorrect GMPP identication and therefore represent a critical test for condence-guided e xploration strate gies. From an optimization perspecti v e, these scenarios progressi v ely increase the ambiguity of the search landscape, making them particularly suitable for e v aluating the ability of the proposed frame w ork to dynami- cally re gulate e xploration ef fort according to prediction reliability . Figure 2 illustrates the limitations of a con v entional P&O controller under se v ere shading conditions. Owing to the presence of multiple local optima, the algorithm con v er ges to w ard a suboptimal operating point rather than the true GMPP . 2.2. Dynamic operating conditions Practical PV systems operate under continuously v arying irradi ance and temperature conditions, re- sulting in time-v arying GMPP locations and e v olving prediction uncertainty . T o reproduce these realistic op- erating conditions, dynamic irradiance perturbations were applied across PV submodules, generating continu- ously changing multi-peak po wer characteristics. These scenarios pro vide a demanding v alidation en vironment for the proposed condence-re gulated frame w ork, requiring continuous adaptation of the e xploration domain according to operating uncertainty . Con- sequently , the y enable assessment of tracking rob ustness, con v er gence stability , computational ef ci enc y , and real-time adaptability under realistic PV conditions. Indonesian J Elec Eng & Comp Sci, V ol. 43, No. 2, August 2026: 672–682 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 675 Figure 2. F ailure of con v entional local-search MPPT under multi-peak PSC conditions 3. AD APTIVE CONFIDENCE-REGULA TED OPTIMIZA TION FRAMEW ORK The proposed methodology is founded on the principle that optimization ef fort should be pr o port ional to operating u nc ertainty . Con v entional MPPT algorithms typically emplo y x ed e xploration strate gies re g ard- less of the reliability of a v ailable information. Consequently , considerable computational resources may be spent e xploring operating re gions with a lo w probability of containing the GMPP . T o address this limitation, the proposed frame w ork introduces a condence-re gulated searc h - space contraction mechanism that dynamically adapts the optimization domain according to prediction reliability . The resulting strate gy combines predicti v e operating-r e gi on estimation, condence quantication, and adapti v e PSO e xploration within a unied decision frame w ork. 3.1. Pr edicti v e operating-r egion estimation A lightweight data-dri v en predictor w as emplo yed to estimate the probable location of high-po wer operating re gions from real-time PV measurements. The predicti v e model recei v es electrical and en vironmental operating v ariables e xtracted from the PV system and generat es an estimate of the most probable high-po wer operating re gion. By supplying an informed initialization of the search process, the predicti v e stage reduces the need for e xhausti v e e xploration while preserving adaptability under irradiance mismatch and PSC. As illustrated in Figure 3, the predicti v e model demonstrates satisf actory agreement between m easured and estimated PV po wer , yielding an R 2 v alue of 0.8664. Although prediction accurac y is not the primary objecti v e of the proposed frame w ork, the obtained estimati on pro vides suf ciently reliable prior information for adapti v e e xploration re gulation. 3.2. Condence-r egulated sear ch-space contraction The central contrib ution of this w ork is the introduction of a condence inde x that establishes a direct coupling between prediction reliability and optimization beha vior . Instead of maintaining a x ed e xploration strate gy , the proposed frame w ork dynamically re gulates the accessible search domain according to the esti- mated certainty of the predicti v e stage. The condence inde x is dened as: C I = 1 − | P pr ed − P measur ed | P max (1) where P pr ed denotes the predicted PV po wer , P measur ed represents the measured oper ating po wer , and P max corresponds to the maximum po wer observ ed within the e xperimental dataset. The condence inde x acts as a lightweight normalized uncertainty indicator that directly links predi c- tion reliability to e xploration intensity while preserving lo w computational comple xity for embedded MPPT Condence-driven adaptive oper ating-point optimization for photo voltaic ... (Sar ah Kawther Sedjar) Evaluation Warning : The document was created with Spire.PDF for Python.
676 ❒ ISSN: 2502-4752 applications. High condence v alues indicate that the predicted operating re gion is lik ely to contain the GMPP , enabling aggressi v e search-space contraction and reduced computational ef fort. Con v ersely , lo wer condence v alues automatically e xpand the e xploration domain to preserv e global-search capability and a v oid premature con v er gence to w ard local optima [23]. Figure 3. Generalization performance of the predicti v e operating-re gion estimator under unseen operating conditions 3.3. Adapti v e optimization strategy Based on the condence inde x, the optimization process dynamically re gulates the accessible search interv al around the predicted operating re gion. The adapti v e search range is e xpressed as: S ear chR ang e = (0 . 05 + 0 . 45(1 − C I ))( V max − V min ) ( 2 ) This formulation establishes a direct relationship between operating uncertainty and e xploration in- tensity . When condence is high, the search domain contracts around the predicted GMPP re gion, thereby reducing redundant particle mo v ements and unnecessary tness e v aluations. Under uncertain operating condi- tions, the search interv al automatically e xpands to maintain suf cient global e xploration capability . Figure 4 illustrates ho w the proposed frame w ork continuously adjusts e xploration ef fort according to prediction reliability . This adapti v e beha vior enables ef cient allocation of computational resources while preserving rob ust GMPP localization capability . Figure 4. Adapti v e search-space contraction as a function of the condence inde x Indonesian J Elec Eng & Comp Sci, V ol. 43, No. 2, August 2026: 672–682 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 677 The o v erall operating principle of t he proposed frame w ork is summarized in Figure 5. Real-time PV measurements are rst processed by the predicti v e estimator to identify probable high-po wer operating re gions. The condence inde x is subse qu e ntly computed and used to re gulate the e xploration domain before the adapti v e PSO stage performs nal GMP P localization. This sequential decision process enables the optimization ef fort to be continuously adjusted according to the actual le v el of operating uncertainty . Figure 5. Architecture of the proposed condence-re gulated adapti v e MPPT frame w ork 4. RESUL TS AND DISCUSSION This section e v aluates the proposed condence-re gulated optimi zation frame w ork under static and dynamic PSC. P articular attention is de v oted to assessing the ef fecti v eness of the condence-guided search- space contraction mechanism in terms of tracking accurac y , con v er gence beha vior , computat ional ef cienc y , and rob ustness under operating uncertainty . 4.1. Pr edicti v e operating-r egion estimation perf ormance The predicti v e operating-re gion estimator achie v ed an RMSE of 3268.48 W , a prediction error of 19.23%, and an R 2 v alue of 0.8664 between predicted and measured PV po wer . Although the predicti v e model e xhibits a prediction error of 19.23%, its role within the proposed frame w ork is fundamental ly dif ferent from that of a con v entional PV po wer forecasting model. The predic- tor is intentionally emplo yed as a coarse operating-re gion estimator whose objecti v e is to identify probable high-po wer re gions rather than accurately est imate the e xact po wer output. The obtained R 2 v alue of 0.8664 indicates that the model successfully captures the dominant trends go v erning PV po wer e v olution under v arying en vironmental conditions. This i nformation is subsequently processed by the condence-re gulated mechanism, which dynamically adjusts the e xploration domain according to prediction reliability . Consequently , the o v erall tracking performance does not depend solely on ANN prediction accurac y . Instead, the condence-guided adaptation compensates for residual prediction uncertainty and enables ef cient GMPP localization. This beha vior is conrmed by the achie v ed tracking ef cienc y of 90.89%, which remains signicantly higher than the predicti v e accurac y itself. Figure 6 presents the comparison between the measured photo v oltaic po wer and the po wer predicted by the proposed h ybrid ANN-PSO model under v arying outdoor operating conditions. 4.2. Static operating-point con v er gence perf ormance The proposed frame w ork achie v ed the lo west mean tracking error (7.74%) and the highest tracking ef cienc y (90.89%) among the e v aluated MPPT strate gies. Furthermore, the maximum tracking error w as reduced to 18.99%, indicating impro v ed rob ustness under irradiance mismatch and temperature v ariations. Condence-driven adaptive oper ating-point optimization for photo voltaic ... (Sar ah Kawther Sedjar) Evaluation Warning : The document was created with Spire.PDF for Python.
678 ❒ ISSN: 2502-4752 The observ ed impro v ement is not solely attrib utable to the predicti v e stage itself. Rather , it res ults from the ability of the condence-re gulated mechanism to dynamically allocate e xploration ef fort according to operating uncertainty . By concentrating the search process around condence-supported operating re gions, the frame w ork reduces the lik el ihood of con v er gence to w ard misleading local maxima while preserving suf cient global-search capability . 4.3. Dynamic tracking perf ormance Figure 7 illustrates the dynamic beha vior of the proposed frame w ork under rapi dly v arying PSC con- ditions. As illustrated in Figure 7, the proposed frame w ork maintains stable con v er gence despite continuous displacement of the GMPP caused by irradiance uctuations. Specically , Figures 7(a)–(d) respecti v ely present the dynamic con v er gence response, oscillation analysis, tracking stability under PSC, and adapti v e con v er gence beha vior of the proposed frame w ork. The adapti v e search-space re gulation mechanism limits unnecessary os- cillatory beha vior while preserving responsi v eness to rapidly changing operating conditions. The results further demonstrate that condence-guided e xploration enables a more ef fecti v e balance between local e xploitation and global e xploration. Consequently , the proposed frame w ork achie v es impro v ed con v er gence stability while maintaining rob ust tracking performance under highly dynamic operating en viron- ments. 4.4. Computational efciency analysis A computational analysis w as conducted to e v aluate the suitability of the proposed frame w ork for real-time PV control applications. T able 1 summarizes the obtained computational performance. T able 1. Computational performance comparison of MPPT strate gies under PSC conditions Method Cate gory Ef cienc y (%) Con v . Steps Ex ec. T ime(s) Oscillation Std. (%) GMPP Accurac y (%) Mean Error (%) R T Suitability P&O Classical 73.55 120 0.0078 High 73.87 26.13 High Con v entional PSO Metaheuristic 87.57 56 0.0003 17.45 91.70 8.30 Moderate Proposed Frame w ork Condence-Guided 90.89 41 0.0005 15.16 92.26 7.74 High Compared with con v entional perturbation-based tracking, the proposed frame w ork reduced the a v- erage con v er gence requirement from 120 to 41 steps while simultaneously impro ving GMPP accurac y and reducing oscillatory beha vior . Unlik e con v entional optimization approaches that emplo y x ed e xploration domains, the proposed strate gy dynamically re gulates the ef fecti v e search re gion according to prediction reliability . Consequently , computational resources are concentrated on operating re gions with a higher probability of containing the GMPP , reducing redundant particle updates and unnecessary tness e v aluations. The measured e x ecution times remained within the millisecond range under identical simulation con- ditions and s w arm settings for all compared methods, indicating the suitability of the proposed frame w ork for real-time embedded PV control applications. 4.5. Ablation study of the pr oposed framew ork An ablation study w as conducted to quantify the indi vidual contrib ution of predicti v e estimation and condence-re gulated search-space adaptation. The ablati on results conrm that the observ ed performance g ains originate from the interaction between predicti v e estimation and condence-re gulated search-space adaptation. The full frame w ork achie v es the best compromise between tracking accurac y , con v er gence s peed, rob ustness, and computational ef cienc y . The detailed quantitati v e comparison of the e v aluated congurations is presented in T able 2. T able 2. Ablation study of the proposed adapti v e MPPT frame w ork Conguration Ef cienc y (%) T racking stability Con v er gence speed PSO only 82.39 Moderate Moderate ANN only 80.77 Lo w F ast ANN + PSO 87.45 Good Good ANN + Adapti v e search 89.02 Better F aster Proposed full frame w ork 90.89 High F ast Indonesian J Elec Eng & Comp Sci, V ol. 43, No. 2, August 2026: 672–682 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 679 Figure 6. Comparison between measured and simulated photo v oltaic po wer under v arying outdoor conditions (a) (b) (c) (d) Figure 7. Dynamic con v er gence and oscillation beha vior under PSC conditions; (a) dynamic con v er gence response, (b) oscillation analysis, (c) tracking stability under PSC, and (d) adapti v e con v er gence beha vior Condence-driven adaptive oper ating-point optimization for photo voltaic ... (Sar ah Kawther Sedjar) Evaluation Warning : The document was created with Spire.PDF for Python.
680 ❒ ISSN: 2502-4752 5. POSITIONING RELA TIVE T O EXISTING MPPT APPR O A CHES T able 3 compares the proposed frame w ork with recent MPPT strate gies reported in the l iterature. Recent representati v e approaches include DRL-based MPPT strate gies, adapti v e metaheuristics, and h ybrid optimization frame w orks designed for PSC en vironments [15]-[18], [24]. Unlik e accurac y-oriented MPPT approaches that rely on e xtensi v e e xploration or computationally in- tensi v e learning procedures, the pr o pos ed frame w ork focuses on condence-re gulated e xploration. This design prioritizes computational ef cienc y , adapti v e decision-making, and embedded implementation feas ibility while maintaining competiti v e GMPP tracking performance under PSC conditions. Be yond the comparati v e performance discussed abo v e, the proposed condence-re gulated op t imiza- tion frame w ork also pro vides a promising foundation for future embedded photo v oltaic ener gy-management systems. In particular , e xtending the proposed condence-guided optimization strate gy to w ard adv anced real- time PV ener gy-management architectures represents an important research direction, as highlighted in recent studies on intelligent photo v oltaic ener gy management [25]. T able 3. Comparati v e positioning of recent MPPT strate gies under PSC conditions Method Exploration strate gy Computational cost T raining b urden Adapti vity Embedded suitability Main limitation P&O Local search V ery lo w None Lo w High Local optimum trapping Con v entional PSO Global e xploration Moderate None Moderate Good Lar ge search domain GW O-PSO h ybrid Hybrid e xploration Moderate–High None High Moderate Incre ased comple xity NGW O-Based MPPT Adapti v e e xploration Moderate None High Moderate P arameter tuning sensiti vity ISSA-P&O Hybrid Metaheuristic-guided High None High Moderate Hi gh computational b urden DRL-based MPPT Polic y learning V ery high High V ery high Limited T raining comple xity DQN/DDPG-based MPPT Deep reinforcement learning V ery high V ery high V ery high Limited Lar ge data requirement Adapti v e GW O/PSO/PO A Adapti v e metaheuristic Moderate–High None High Moderate Exploration o v erhead Pr oposed framew ork Condence- r egulated exploration Lo w–Moderate Lo w Adapti v e High Pr ediction- dependent 6. CONCLUSION This paper introduced a condence-re gulated search-space contraction frame w ork for adapti v e MPPT in PV systems under PSC. Unlik e con v entional intelligent MPPT approaches that emplo y x ed e xploration strate gies, the proposed methodology dynamically adjusts the optimization domain according to prediction reliability through a condence-guided search-space contraction mechanism. The obtained results demonstrate that linking e xploration ef fort to operating uncertainty enables more ef cient allocation of computational resources while preserving rob ust GMPP localization capability . Experi- mental e v aluation under static and dynamic PSC scenarios sho wed impro v ed tracking stability , reduced con v er - gence ef fort, and an a v erage tracking ef cienc y of 90.89%. Furthermore, v alidation ag ainst real PVD A Q mea- surements yielded an R 2 v alue of 0.8664. Although the predi cti v e stage w as not intended to pro vide highly ac- curate po wer forecasting, it successfully supplied reliable operating-re gion information for condence-guided e xploration control. The obtained results demonstrate that coupling a coarse predictor with adapti v e condence- re gulated optimization can achie v e rob ust GMPP localization while maintaining lo w computational comple xity . The proposed condence-re gulated search-space contraction strate gy establishes a direct connection between predicti v e information and optimization beha vior , allo wing e xploration intensity to adapt continu- ously to changing operating conditions. This characteristic mak es the frame w ork particularly attracti v e for lightweight embedded PV controllers where computational ef cienc y and real-time responsi v eness are critical design requirements. Future w ork will focus on hardw are implementation, e xperimental real-time v alidation, and e v aluation of the proposed frame w ork under more di v erse real-w orld photo v oltaic operating conditions. Indonesian J Elec Eng & Comp Sci, V ol. 43, No. 2, August 2026: 672–682 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 681 FUNDING INFORMA TION The authors state that no specic funding w as recei v ed for this w ork. A UTHOR CONTRIB UTIONS ST A TEMENT This journal uses the Contrib utor Roles T axonomy (CRediT) to recognize indi vidual author contrib utions, reduce authorship disputes, and f acilitate collaboration. Name of A uthor C M So V a F o I R D O E V i Su P Fu Sarah Ka wther Sedjar ✓ ✓ ✓ ✓ ✓ ✓ ✓ Mourad Benmessaoud ✓ ✓ ✓ ✓ ✓ C : C onceptualization I : I n v estig ation V i : V i sualization M : M ethodology R : R esources Su : Su pervision So : So ftw are D : D ata Curation P : P roject Administration V a : V a lidation O : Writing - O riginal Draft Fu : Fu nding Acquisition F o : F o rmal Analysis E : Writing - Re vie w & E diting CONFLICT OF INTEREST ST A TEMENT Authors state no conict of interest. D A T A A V AILABILITY The e xperimental photo v oltaic measurements used in this study were obtained from the Open Ener gy Data Initiati v e (OEDI) PVD A Q database. Additional processed data and simulation results supporting the ndings of this w ork are a v ailable from the corresponding author upon reasonable request. REFERENCES [1] F . Blaabjer g, Y . Y ang, D. Y ang, and X. W ang, “Distrib uted po wer -generation systems and protection, ” Proceedings of the IEEE , v ol. 105, no. 7, pp. 1311–1331, Jul. 2017, doi: 10.1109/JPR OC.2017.2696878. [2] A. Mohapatra, B. Nayak, P . Das, and K. B. Mohanty , “ A re vie w on MPPT techniques of PV system under partial shading condition, ” Rene w able and Sustainable Ener gy Re vie ws , v ol. 80, pp. 854–867, Dec. 2017, doi: 10.1016/j.rser .2017.05.083. [3] T . Esram and P . L. Chapman, “Comparison of photo v oltaic array maximum po wer point tracking techniques, ” IEEE T ransactions on Ener gy Con v ersion , v ol. 22, no. 2, pp. 439–449, Jun. 2007, doi: 10.1109/TEC.2006.874230. [4] N. Femia, G. Petrone, G. Spagnuolo, and M. V itelli, “Optimization of perturb and observ e maximum po wer point tracking method, ” IEEE T ransactions on Po wer Electronics , v ol. 20, no. 4, pp. 963–973, Jul. 2005, doi: 10.1109/TPEL.2005.850975. [5] H. P atel and V . Ag arw al, “Maximum po wer point tracking scheme for PV systems operating under partially shaded conditions, ” IEEE T ransactions on Industrial Electronics , v ol. 55, no. 4, pp. 1689–1698, Apr . 2008, doi: 10.1109/TIE.2008.917118. [6] M. A. Danandeh and S. M. Mousa vi G., “Comparati v e and comprehensi v e re vie w of maximum po wer point tracking methods for PV cells, ” Rene w able and Sustainable Ener gy Re vie ws , v ol. 82, pp. 2743–2767, Feb . 2018, doi: 10.1016/j.rser .2017.10.009. [7] H. Rezk and A. M. Eltamaly , “ A comprehensi v e compa rison of dif ferent MPPT techniques for photo v oltaic systems, ” Solar Ener gy , v ol. 112, pp. 1–11, Feb . 2015, doi: 10.1016/j.solener .2014.11.010. [8] K. Ishaque, Z. Salam, M. Amjad, and S. Mekhilef, “ An impro v ed particle sw arm optimization (PSO)-bas ed MPPT for PV with reduced steady-state oscillation, ” IEEE T ransactions on Po wer Electronics , v ol. 27, no. 8, pp. 3627–3638, Aug. 2012, doi: 10.1109/TPEL.2012.2185713. [9] B. Y ang et al. , “Salp sw arm optimization algorithm based MPPT design for PV -TEG h ybrid system under partial shading condi- tions, ” Ener gy Con v ersion and Management , v ol. 292, p. 117410, Sep. 2023, doi: 10.1016/j.enconman.2023.117410. [10] M. Se yedmahmoudian et al. , “State of the art articial intelligence-based MPPT techniques for mitig ating partial shading ef fects on PV systems – A re vie w , ” Rene w able and Sustainable Ener gy Re vie ws , v ol. 64, pp. 435–455, Oct. 2016, doi: 10.1016/j.rser .2016.06.053. [11] L. F . Giraldo, J. F . Ga viria, M. I. T orres, C. Alonso, and M. Bressan, “Deep reinforcement learning using deep-Q-netw ork for global maximum po wer point tracking: design and e xperiments in real photo v oltaic systems, ” Heliyon , v ol. 10, no. 21, p. e37974, No v . 2024, doi: 10.1016/j.heliyon.2024.e37974. [12] B. C. Phan, Y . C. Lai, and C. E. Lin, “ A deep reinforcement learning-based MPPT control for PV systems under partial shading condition, ” Sensors (Switzerland) , v ol. 20, no. 11, p. 3039, May 2020, doi: 10.3390/s20113039. [13] T . T . Hoang and T . H. Le, “De v elopment of deep reinforcement learning for maximum po wer point tracking of photo v oltaic sys- tems, ” Indonesian Journal of Electrical Engineering and Computer Science (IJEECS) , v ol. 33, no. 2, pp. 707–714, Feb . 2024, doi: 10.11591/ijeecs.v33.i2.pp707-714. [14] M . Ismail, M. I. Marei, and M. Mokhtar , “ Adapti v e h ybrid MPPT for photo v oltaic systems: performance enhancement under dynamic conditions, ” Sustainability (Switzerland) , v ol. 18, no. 1, p. 80, Dec. 2026, doi: 10.3390/su18010080. Condence-driven adaptive oper ating-point optimization for photo voltaic ... (Sar ah Kawther Sedjar) Evaluation Warning : The document was created with Spire.PDF for Python.