Accurate parameter identification in nonlinear and chaotic dynamic systems requires optimization algorithms that can reliably balance global exploration and local refinement in complex, multimodal search landscapes. To address this challenge, a modified artificial protozoa optimizer (mAPO) is developed in this study by embedding two complementary mechanisms into the original artificial protozoa optimizer: a probabilistic random learning strategy to enhance population diversity and global search capability, and a Nelder–Mead simplex-based local refinement stage to improve exploitation and fine-scale solution adjustment. The general optimization performance and scalability of the proposed framework are first evaluated using the CEC2017 benchmark suite. Statistical analyses conducted over shifted and rotated, hybrid, and composition functions demonstrate that mAPO achieves improved mean performance and reduced variability compared with the original APO, indicating enhanced robustness in high-dimensional and complex optimization problems. The effectiveness of mAPO is then examined in nonlinear system identification applications involving chaotic dynamics. Offline and online parameter identification experiments are performed on the Rössler chaotic system and a permanent magnet synchronous motor, including scenarios with abrupt parameter variations. Comparative simulations against APO and several state-of-the-art optimizers show that mAPO consistently yields smaller objective function values, more accurate parameter estimates, and superior statistical stability. In the PMSM case, exact parameter reconstruction with zero error is achieved across all independent runs, while rapid and smooth convergence is observed under both static and time-varying conditions.

Many signal processing applications such as acoustic echo cancellation and wireless channel estimation require identifying systems where only a small fraction of coefficients are actually active, i.e. sparse systems. Zero-attracting adaptive filters tackle this by adding a penalty that pulls inactive coefficients toward zero, speeding up convergence. However, these algorithms determine which coefficients to penalize based solely on their current size. This creates a problem during early adaptation since active coefficients that should eventually grow large start out small, making them look identical to truly inactive coefficients. The algorithm ends up applying strong penalties to the very coefficients it needs to develop, slowing down the initial convergence. This paper provides a solution to this problem by introducing a dual-domain approach that looks at coefficients from two perspectives simultaneously. Beyond just tracking coefficient magnitude, we introduce an error-memory vector that monitors how persistently each coefficient contributes to the adaptation error over time. If a coefficient keeps showing up in the error signal, it is probably active even if it is still small. By combining both views, the proposed dual-domain sparse adaptive filter (DD-SAF) can identify active coefficients early and eliminate penalties accordingly. Moreover, a complete theoretical analysis is derived. The analysis shows that DD-SAF maintains the same stability properties as standard least-mean-square (LMS) while achieving provably better steady-state performance than existing methods. Simulations demonstrate that the DD-SAF converges to the steady-state faster and/or converges to a lower mean-square-deviation (MSD) than the standard LMS and the reweighted zero-attracting LMS (RZA-LMS) algorithms for sparse system identification settings.

Accurate parameter identification is a critical prerequisite for reliable modeling, analysis, and control of nonlinear dynamical systems. This study introduces the stellar oscillation optimizer (SOO), a recently proposed metaheuristic inspired by the oscillatory behavior of stars, and investigates its effectiveness in estimating system parameters through a unified optimization framework. The identification problem is formulated as the minimization of a trajectory–mismatch cost function, where candidate solutions are iteratively refined by the oscillatory dynamics of SOO. To comprehensively evaluate its performance, four benchmark systems were considered: three canonical chaotic models (Lorenz, Chen, and Rössler) and a practical engineering case represented by a permanent-magnet synchronous motor (PMSM). The outcomes were benchmarked against several state-of-the-art algorithms, including Kirchhoff’s law algorithm (KLA), Tianji’s horse racing optimization (THRO), puma optimizer (PO), and hiking optimization algorithm (HOA), under a standardized protocol. The results show that SOO consistently achieves numerically convergent solutions with machine-precision-level residuals under deterministic and noise-free simulation settings, while maintaining strong robustness across independent runs. In chaotic benchmarks, the reported residuals approach floating-point limits, which indicates stable numerical convergence rather than guaranteed physical identifiability under real measurement conditions. On the PMSM model, SOO demonstrates accurate and repeatable parameter estimation within the adopted simulation framework.

ABSTRACT In this paper, the localisation of an unmanned aerial vehicle (UAV) in a cooperative manner in global positioning systems (GPS)‐denied environments using multiple base stations and multiple integrated reflecting surfaces (IRSs) is performed. A two‐stage scheme is proposed to localise the UAV. In the first stage, a phase gain matrix is used for IRSs, and a symbol is transmitted by the base transceiver station (BTS), and the first received symbol is observed by the UAV. In the second stage, the inverse phase gains are set for IRSs, and again the same symbol is transmitted by BTSs, and a second received symbol is observed by the UAV. Using only these two received symbols, the distances between the UAV and the BTSs and the distances between the UAV and the IRSs are estimated in the two different versions of the proposed algorithm, and then the location of the UAV is extracted via trilateration. Some distance estimation error analysis is also provided in the paper. Simulation results show the efficiency of the proposed scheme in terms of one order of magnitude faster run time of the BTS version of the proposed algorithm, whilst it is slightly less accurate than a state‐of‐the‐art algorithm in the literature.

In this article, we address the problem of masked graph signal recovery (GSR) in the presence of unknown false data injection (FDI) attacks. We introduce, for the first time in the graph signal processing literature, a mask detection mechanism that operates without any prior knowledge of the masking operator and using only a single snapshot of the observed signal. A subset of nodes is assumed to be compromised by adversaries injecting arbitrary perturbations, while both the number and locations of these nodes are unknown. To detect the masked (adversarial) nodes, we develop a statistically grounded and computationally efficient test based on differential smoothness. The detector evaluates the change in graph signal smoothness when substituting the value of each node with a nominal estimate, yielding a closed-form decision statistic and enabling low-complexity mask identification. Following adversary detection, signal recovery is performed through a smoothness-promoting reconstruction framework formulated as a fractional optimization problem and solved efficiently via Dinkelbach's method. We further derive analytical expressions for the detector's operating characteristics, enabling principled threshold selection. Simulation results demonstrate that the proposed secure GSR algorithm significantly outperforms graph median filtering and other baseline methods, achieving notable gains in reconstruction accuracy under diverse adversarial conditions.

Accurate identification of parameters in chaotic and nonlinear systems is essential for ensuring precise modeling, control, and prediction of complex dynamical behaviors. However, conventional metaheuristic algorithms often struggle to maintain an effective balance between exploration and exploitation, leading to premature convergence and estimation inaccuracies. To address these challenges, this study proposes an enhanced golden jackal optimizer (en-GJO) that integrates three complementary mechanisms (Laplacian crossover learning, elite group learning, and opposition repair learning). These hybrid strategies collectively strengthen population diversity, accelerate convergence, and prevent stagnation, thereby improving both the global search capability and local refinement accuracy of the original GJO. The effectiveness of the en-GJO is first validated through extensive benchmarking on twenty-three standard test functions, including unimodal, multimodal, and fixed-dimensional multimodal problems. Comparative results against nine well-established metaheuristics (such as SSA, SCA, HHO, AEO, EO, GBO, RUN, and ARO) demonstrate that en-GJO achieves superior convergence precision and robustness, consistently yielding the lowest mean and standard-deviation values across all categories. To further verify its real-world applicability, the en-GJO is applied to the parameter identification of a memristive chaotic system, formulated as a nonlinear optimization problem using a least-squares-based objective function. Simulation results reveal that the proposed method attains the most accurate estimates of the system parameters ( a,b,c,d) , with negligible deviation from their true values. Statistical analyses and convergence profiles confirm that en-GJO not only converges faster but also delivers more stable and repeatable performance than competing algorithms. In comparative evaluations with reported techniques such as PSO, ABC, SPSSA, GWO, POA, and FPPOA, the en-GJO achieves the smallest cost value (1.3850 × 10−13) and with a mean fitness of 1.0507 × 10−9 and a standard deviation of 2.5392 × 10−9, outperforming all compared algorithms by several orders of magnitude. The estimated system parameters converge to their true values with error rates below 0.001

This paper proposes a bi-level optimization problem that models the interconnection between a virtual power plant (VPP) and the private owners (POs) in distribution systems. The upper level of the presented model focuses on the sustainable operation of the distribution system. The VPP presents a multi-objective optimization to, simultaneously, maximize its profit and minimize excessive water extraction. The VPP acts as an intermediary, facilitating interaction between retail and wholesale electricity markets. It participates in the wholesale market to buy energy from and sell it to POs in the retail market to increase profit. At the lower level, the POs participate in the retail market to purchase electricity from the VPP. The aim of the POs is to minimize their cost. They can generate electricity by their own resources or import it from the VPP. The coordination between the VPP and POs is formulated as a bi-level problem that converts to a single level model using the Karush-Kuhn-Tucker conditions and big-M approach. The obtained results present that the bi-objective model reduces the unsupplied energy and water extraction by 95% and 7.53%, respectively.

This letter generalizes noise modulation by introducing two voltage biases and employing non-Gaussian noise distributions, such as a mixture of Gaussians and Laplacian, in addition to traditional Gaussian noise. The proposed framework doubles the data rate by enabling discrimination in both the mean and variance of transmitted noise symbols. This novel modulation scheme is referred to as generalized quadratic noise modulation (GQNM). Closed-form expressions for the bit error probability are derived for the generalized Gaussian (GG) and generalized Gaussian mixture of Gaussians (GMoG) cases. Simulation results demonstrate the advantages of the generalized modulation scheme, particularly under non-Gaussian noise assumptions, highlighting its potential for enhanced performance in low-power and secure communication systems.

This article presents a novel three-dimensional (3-D) 8-ary noise modulation scheme that introduces a new dimension: the mixture probability of a Mixture of Gaussian (MoG) distribution. This proposed approach utilizes the dimensions of mean and variance, in addition to the new probability dimension. Within this framework, each transmitted symbol carries three bits, each corresponding to a distinct sub-channel. For detection, a combination of specialized detectors is employed: a simple threshold-based detector for the first sub-channel bit (modulated by the mean), a Maximum-Likelihood (ML) detector for the second sub-channel bit (modulated by the variance), a Kurtosis-based, Jarque-Bera (JB) test, and Bayesian Hypothesis (BHT)-based detectors for the third bit (modulated by the MoG probability). The Kurtosis- and JB-based detectors specifically distinguish between Gaussian (or near-Gaussian) and non-Gaussian MoG distributions by leveraging higher-order statistical measures. The Bit Error Probabilities (BEPs) are derived for the threshold-, Kurtosis-, and BHT-based detectors. The optimum threshold for the Kurtosis-based detector is also derived in a tractable manner. Simulation results demonstrate that a comparably low BEP is achieved for the third sub-channel bit relative to existing two-dimensional (2-D) schemes. Simultaneously, the proposed scheme increases the data rate by a factor of 1.5 and 3 compared to the Generalized Quadratic noise modulator and the classical binary KLJN noise modulator, respectively. Furthermore, the Kurtosis-based detector offers a low-complexity solution, achieving an acceptable BEP of approximately 0.06.

Existing adaptive filtering methods typically assume coefficient independence and apply uniform sparsity penalties, overlooking potential structural relationships among active coefficients for system identification. In this paper, a topology-aware sparse adaptive filter (TA-SAF) is presented. The proposed method learns coefficient relationships during adaptation and uses this learned structure to guide sparsity enforcement. Through adaptive parameter design, the proposed method achieves robust performance across varying adaptation speeds. A mean-square deviation (MSD) analysis is provided, deriving the steady-state MSD floor and establishing a formal connection to the standard LMS algorithm. A refined MSD expression that retains the dependence on the base topology factor beta(0) is also obtained, making the effect of the topology influence on the steady-state floor explicit. The per-iteration computational complexity is characterized and compared against existing methods. Simulations demonstrate improvement over the least-mean-square (LMS), the proportionate normalized LMS (PNLMS), the reweighted zero-attracting LMS (RZA-LMS) and the block-sparsity-induced adaptive filter (BS-LMS) algorithms.

In this study, a robust position control strategy is presented for a two-stage electro-hydraulic actuator system using a fractional-order proportional-integral-derivative (FOPID) controller tuned by Kirchhoff's law algorithm (KLA). Electro-hydraulic systems exhibit strong nonlinearities, parameter uncertainties, and external disturbances, making accurate position control challenging. Classical integer-order controllers such as PI and PID often provide limited flexibility in shaping transient dynamics under such conditions, leading to performance degradation when system parameters vary. Although advanced nonlinear and adaptive control strategies can improve robustness, their implementation complexity and tuning burden may restrict their practical applicability. In addition, many existing metaheuristic-based tuning approaches rely on algorithm-specific parameters and may exhibit inconsistent convergence behavior across different operating conditions. To address these limitations, a control-oriented mathematical model of the electro-hydraulic positioning system is first derived from linearized valve flow relations, actuator continuity equations, and load dynamics, and a reduced-order transfer function preserving the dominant behavior is obtained for controller design. The FOPID controller is adopted to enlarge the tuning space through fractional integral and derivative orders, enabling finer dynamic shaping than classical integer-order structures. The controller parameters are determined automatically by formulating a multi-term time-domain objective function that penalizes overshoot, settling time, and accumulated tracking error. The KLA metaheuristic, inspired by electrical circuit laws, is employed to solve this constrained optimization problem without introducing algorithm-specific control parameters. Under identical optimization settings (population size 25, 40 iterations, 20 independent runs), the proposed KLA approach achieves the best statistical performance among the tested optimizers, with a minimum fitness value of 1.2511, mean value of 1.2881, and standard deviation of 0.0258, outperforming APO, PSO, and DE according to Wilcoxon signed-rank tests (p approximate to 8.86 & times; 10(-5)). For a 20 cm step command, the KLA-tuned FOPID controller yields a rise time of 0.1664 s, settling time of 0.3034 s, overshoot of 0.0124%, and IAE of 1.5244 cm & centerdot;s, improving all metrics relative to both competing optimizers and KLA-tuned PI, PIDF, and 2-DOF PID controllers. Additional simulations demonstrate stable and accurate tracking under multi-level and time-varying references, parametric variations from -10% to + 25% , and combined disturbance and measurement noise. The results indicate that the KLA-based FOPID design provides a fast, accurate, and robust control solution for electro-hydraulic positioning systems.

Hyperparameter selection plays a critical role in the convergence speed and performance of artificial neural networks (ANNs), yet conventional tuning methods such as grid search, random search, and Bayesian optimization often suffer from high computational cost and limited adaptability. To address these limitations, this study proposes an adaptive hyperparameter optimization framework based on the Bobcat Optimization Algorithm (BOA) for tuning the learning rate and momentum of an ANN. BOA employs a biologically inspired exploration-exploitation mechanism that dynamically adjusts hyperparameters according to training performance, enabling efficient search without relying on probabilistic surrogate models. The proposed BOA-ANN framework is evaluated on the Modified National Institute of Standards and Technology (MNIST) handwritten digit dataset using a three-layer feedforward neural network. Experimental results demonstrate that BOA-ANN achieves a test accuracy of 98.52%, a mean squared error of 0.014, and an F1-score of 98.30%, outperforming Bayesian optimization, the Secretary Algorithm, and Automated Model Compression (AMC) pruning under identical settings.

Wireless systems with integrated communication and sensing capabilities have recently gained significant attention. Whereas most existing approaches rely on structured communication signals, radar waveforms, or pilot-based sensing, noise communication systems employ intentionally noise-like signals that provide a low probability of interception and robust resilience to interference. Nevertheless, sensing and localization for such systems remain largely unexplored. This paper presents a joint localization and communication framework for multiple-input multiple-output (MIMO) noise communication systems, in which multiple transmitters with unknown locations communicate with spatially distributed receivers via on-off keying (OOK) noise signaling. By exploiting received signal strength (RSS) measurements, the proposed framework jointly estimates transmitter locations and detects the transmitted bits. The resulting problem is formulated as a nonlinear least-squares (NLS) optimization task that jointly involves transmitter positions and symbol-dependent transmit powers. A full joint Levenberg-Marquardt (NLS-LM) estimator is developed to jointly recover all unknown parameters. To reduce computational complexity, an alternating least-squares and Levenberg-Marquardt (LS-LM) algorithm is also proposed, alternating between power estimation and localization refinement. In addition, convergence analyses are provided for both algorithms, establishing monotonic objective reduction and convergence to stationary solutions under standard assumptions. Simulation results demonstrate the effectiveness of the proposed methods and highlight the tradeoff between estimation accuracy and computational complexity.

In this study, a hybrid Schrödinger optimizer with differential evolution (h-SRADE) is proposed for optimal tuning of proportional–integral–derivative (PID) controllers applied to a three-tanks liquid level control system. The original Schrödinger optimizer (SRA), inspired by wave–particle duality and quantum mechanics principles, provides a balanced exploration–exploitation mechanism; however, its convergence sensitivity in the later optimization stages motivates further enhancement. To address this limitation, differential evolution mutation and crossover operators are embedded into the SRA framework, forming the proposed h-SRADE algorithm with improved refinement capability and convergence stability. The control problem is formulated using a linearized third-order model of the three-tanks liquid level system derived from mass balance principles. PID controller parameters are optimized by minimizing a composite fitness function that simultaneously accounts for percent overshoot, steady-state error, settling time, and rise time. The optimization process is conducted within predefined practical bounds to ensure controller feasibility and robust closed-loop operation. Extensive simulation studies are carried out to evaluate the performance of the proposed method. Statistical analysis, supported by a non-parametric Wilcoxon signed-rank test, confirms the statistical significance and robustness of h-SRADE over the conventional SRA. Time-domain and frequency-domain analyses demonstrate that the h-SRADE-based PID controller achieves faster settling, reduced overshoot, lower steady-state error, and satisfactory robustness margins. Furthermore, comparative studies with several state-of-the-art optimization-based PID tuning approaches reported in the literature reveal that the proposed method provides superior dynamic performance and overall control quality. The obtained results indicate that the hybridization of the Schrödinger optimizer with differential evolution effectively enhances convergence behavior and control performance. Consequently, the proposed h-SRADE framework offers a reliable and efficient simulation-based solution for PID controller tuning in liquid level systems, providing a strong foundation for future extension toward nonlinear plant models, varying operating points, and other complex engineering control applications.

This paper presents a novel framework for analog variance noise modulation (VNM) and its corresponding variance-based demodulation technique. Unlike traditional digital noise communication systems such as Kirchhoff–Law–Johnson–Noise (KLJN) schemes that encode binary data using discrete noise levels, the proposed analog VNM continuously maps a signal’s amplitude onto the variance of a white Gaussian noise (WGN) carrier. In this way, information is embedded within the statistical behavior of noise rather than its waveform, enabling inherently secure and noise-like transmission. To accurately reconstruct the transmitted signal, we develop a new state-space modeling and inference framework inspired by the Discrete Fourier Transform (DFT). This model captures the temporal structure of the analog baseband signal through its harmonic components while treating the noise variance as a time-varying latent process. Building upon this structure, a Rao–Blackwellized Particle Filter (RBPF) is designed to estimate the underlying signal by jointly inferring its harmonic dynamics and the hidden variance states, enabling robust demodulation under stochastic noise conditions. Simulation results demonstrate that the proposed analog VNM achieves accurate signal recovery and exhibits strong concealment properties compared to classical amplitude and frequency modulation schemes. Overall, this work introduces a new class of analog noise-based communication with an inherently simple and energy-efficient transmitter, while employing a higher-complexity receiver-side inference framework for accurate signal recovery. This asymmetric architecture makes the proposed VNM particularly suitable for low-power sensing and Internet of Things (IoT) scenarios where ultra-lightweight transmitters communicate with computationally capable gateways or base stations.

The demand for hardware-efficient interference suppression algorithms is growing with the increasing density in wireless communication networks. In this paper, a robust position-only null steering method for linear antenna arrays is proposed based on Honey Formation Optimization with Single Component (HFOSC), a metaheuristic algorithm founded on the ripening process of honey in beehives. By optimizing only the element locations, the proposed method avoids the use of phase shifters and attenuators, thus reducing implementation complexity while maintaining flexibility in pattern control. A 30-element linear array with Chebyshev excitation is used to test the technique under representative interference scenarios such as single-null, multiple-null, and wide-sector nulling cases, as well as constrained practical designs. The simulation results demonstrate that the proposed approach can realize strong interference suppression across different cases while maintaining the main-beam shape and acceptable sidelobe performance. In idealized discrete-interference cases, nulls below -90 dB are achieved, while in a practical constrained design with a minimum inter-element spacing of 0.5λ and a position resolution of 0.1λ, a null depth of -72.89 dB is still achieved, confirming the practical applicability of the method. Moreover, comparative results with GA, PSO, and DE over 100 independent runs illustrate that HFOSC achieves the lowest optimization cost and the smallest standard deviation, along with a favorable overall trade-off between beam preservation and null suppression, with statistically significant superiority in optimization performance. The proposed method does not require phase shifters and attenuators, providing a simple, hardware-friendly, and robust solution for adaptive interference cancellation in wireless communication systems.

This paper proposes a novel multi-dimensional noise modulator that uses multi-parameter non-Gaussian distributions, such as the Generalized Gaussian Distribution (GGD). In this scheme, information bits are embedded in the random noise waveform using low and high values of parameters of the multi-parameter noise distribution. Three types of detectors are proposed to detect information bits: the Maximum-Likelihood (ML) detector, the Bayesian Hypothesis Testing (BHT) detector, and a low-complexity detector. For the GGD noise case with three parameters, the 8-ary noise modulator is investigated in detail. Simulation results demonstrate satisfactory performance in terms of bit error probability (BEP).

Efficient route planning is a critical component of smart tourism and destination management, particularly in regions where touristic attractions are geographically dispersed. This study proposes a discrete adaptation of the enzyme action optimizer (EAO), named discrete enzyme action optimizer (D-EAO), to address permutation-based combinatorial optimization problems, with a specific focus on the traveling salesman problem (TSP). The proposed approach integrates random-key encoding to transform continuous solutions into valid permutations, multiple neighborhood operators to enhance search diversity, and a selective 2-opt local search mechanism to improve solution quality. Through this hybrid structure, D-EAO achieves a balanced exploration-exploitation behavior in discrete search spaces. The performance of D-EAO is first evaluated on several well-known benchmark TSP instances with varying problem sizes and is comparatively analyzed against recent discrete metaheuristic algorithms, including discrete Kirchhoff's law algorithm (D-KLA), discrete Tianji's horse racing optimization (D-THRO), and discrete hiking optimization algorithm (D-HOA). Experimental results demonstrate that D-EAO consistently achieves superior or competitive performance in terms of best, average, and standard deviation metrics. Statistical significance analyses using the Wilcoxon signed-rank test confirm that the observed performance improvements are statistically significant across most medium- and large-scale instances. In addition to benchmark evaluations, the proposed algorithm is applied to a real-world tourism-oriented TSP scenario involving 46 tourist points in Bitlis, Türkiye. The results show that D-EAO generates shorter, more stable, and geographically coherent tourism routes compared to competing methods. These findings highlight the effectiveness and scalability of the proposed approach and demonstrate its potential as a decision-support tool for smart tourism planning and sustainable destination management.

Model predictive control has emerged as a leading supervisory strategy for building heating, ventilation, and air-conditioning systems, yet its deployment is constrained by a solver-dependent decision layer that must solve a mixed-integer nonlinear program at each control update, thereby introducing non-deterministic runtime, convergence failure, and fallback events that undermine reliable operation on resource-limited building automation hardware.This study proposes Solver-Free Model Predictive Fuzzy Control (SF-MPFC), a supervisory architecture that replaces only the solver-dependent decision layer of conventional MPC with a deterministic fuzzy-inference table-lookup policy while preserving the shared upstream predictive structure, including disturbance forecasting, horizon-based thermal-potential computation, and demand shaping. The core design principle is operating-condition normalization, in which FCU-side and TES-side thermal potential loads are expressed relative to the predicted full-load heat pump capacity under current conditions, yielding dimensionless decision inputs that are invariant to seasonal operating conditions.To validate this architecture under rigorous annual closed-loop conditions, the TRNSYS-validated PVT–GHX multi-heat-source heat pump system was reformulated as a standalone Python digital twin. This reformulation eliminates co-simulation fragility and interface-level overhead, enabling continuous 8760 h benchmarking while reproducing annual energy behavior within 2.8% relative error across all major energy terms.Under identical predictive layers, constraints, horizons, and actuation grids, SF-MPFC achieved 24.5% annual net energy savings relative to a rule-based baseline, closely matching MPC at 24.1%, while maintaining stable room-temperature regulation with RMSE of 0.84–0.85°C and MAE of 0.64°C for thermal comfort. SF-MPFC reduced per-update computation time by a factor of 1,000–2,500 and eliminated solver fallback events across all tested configurations, whereas MPC incurred fallback events in every configuration, reaching 1,146 events under a 6 h horizon with a 5 min prediction grid.These results establish that the solver-dependent decision layer, not the predictive structure is the binding constraint on MPC deployability in renewable multi-source heat pump systems. Replacing this layer with a deterministic fuzzy-inference policy delivers MPC-grade annual energy performance while guaranteeing bounded execution at every control update, regardless of horizon length or control-grid resolution. More fundamentally, this architectural separation provides an implementation-level foundation for supervisory extensions toward multi-building coordination, asset aggregation, and community-scale load shifting applications in which computational boundedness is a prerequisite for deployment rather than a secondary design consideration..

The recursive inverse (RI) adaptive filtering technique has demonstrated superior performance characteristics compared to various established adaptive filtering methods. A primary challenge facing the RI technique is its dependence on time-varying step-size parameters, which limits its effectiveness in situations involving substantial eigenvalue spread within the covariance matrix. This work presents an enhanced variant of the RI technique. Our proposed methodology employs a nonlinear step-size mechanism designed to mitigate the influence of covariance matrix eigenvalue spread on RI technique performance. Although this nonlinear step-size approach introduces additional computational burden compared to the standard RI technique, the overall complexity remains on par with that of the RLS method. Through implementation of this nonlinear step-size strategy, our proposed technique achieves superior performance in scenarios characterized by large eigenvalue spread within the covariance matrix, effectively resolving the primary constraint of the traditional RI approach. Experimental validation demonstrates that our proposed technique outperforms the traditional RI method and achieves performance levels equivalent to or exceeding those of the RLS technique.