Mitigation of transient beam loading in a compact high duty cycle injector via hardware upgrade and multi-objective optimization Authors: Hao Hu, Yu’ang Xia, Yan Wang, Yifeng Zeng, Kaifeng Liu, Tongning Hu Published: 2026 ABSTRACT AbstractCompact normal-conducting photoinjectors, such as the Independently Tunable Cells (ITC) injector, are critical for high-power terahertz free electron laser (FEL) sources. However, under high duty cycle operation, the transient beam loading effect significantly degrades the uniformity of electron beam energy and phase within the macropulse, thereby compromising the FEL radiation performance. Here, we propose a strategy to mitigate these transient effects and enhance the macropulse beam quality of the ITC injector for the HUST THz-FEL facility. The proposed hardware upgrade decouples the RF power feeds for the standing wave (SW) cavity and the traveling wave (TW) linac to enable independent timing control, and introduces an adjustable coupling coefficient for the SW cavity. Furthermore, the Non-dominated Sorting Genetic Algorithm III (NSGA-III) is employed to perform multi-objective optimization on four key variables: beam injection time, RF phase, coupling coefficient, and the inter-structure RF delay. Simulation results indicate that single-variable optimization is insufficient to balance energy and phase stability. In contrast, global multi-objective optimization reveals that increasing the coupling coefficient effectively reduces the cavity filling time, while independent RF timing synchronizes the accelerating fields with the beam transient. The optimized configuration achieves a stable macropulse energy of approximately 14 MeV with minimized energy spread and phase variation. Consequently, the combination of independent RF feeding, adjustable coupling, and global optimization effectively compensates for transient beam loading effects. This approach markedly improves the macropulse uniformity of the ITC injector in simulation, offering a potential pathway for similar compact injectors.Subject terms: Engineering, Physics Introduction Free Electron Lasers (FELs) are distinguished as the only high-power light sources capable of providing tunable coverage across the entire electromagnetic spectrum1. Particularly in frequency bands that are inaccessible to conventional sources—ranging from Terahertz (THz) waves and soft X-rays to hard X-rays—FELs can generate radiation with both high peak and high average power. Consequently, they have been widely adopted in fundamental research and various industrial applications2,3. In recent years, short-wavelength FELs have provided new pathways for overcoming critical technological bottlenecks, such as those found in advanced lithography4,5. Simultaneously, long-wavelength FELs have gradually expanded from military applications to civil sectors, including large-scale security screening and long-distance communications6,7. To meet the significant demands of civil industry and national defense, the generation of high photon flux FEL radiation imposes increasingly stringent requirements on the power parameters of the driving electron beams. Among the nearly 100 FEL facilities currently in operation or under construction worldwide, the driving accelerators for high-power radiation primarily follow two technological routes: superconducting and normal-conducting (room-temperature) structures. While superconducting structures offer superior beam quality and high radiation power8–11, they are associated with system complexity and high construction and operational costs. Consequently, normal-conducting structures continue to constitute a significant proportion of existing facilities. These include short-wavelength FELs such as SwissFEL (Switzerland), PAL-XFEL (Korea), SACLA (Japan), and SDUV-FEL (China)5,12–15, as well as long-wavelength facilities like FELIX (Netherlands), NovoFEL (Russia), and FELiChEM and HUST THz-FEL (China)16–19. Notably, the HUST THz-FEL facility employs an Independently Tunable Cells (ITC) injector. This design circumvents the need for complex, expensive photocathode guns and bulky thermionic gun-plus-bunching systems, striking a balance between beam quality and structural compactness. It operates with a macropulse repetition rate of up to 50 Hz, where the micropulse repetition rate is synchronized with the RF frequency19. From an economic perspective, upgrading the power of existing normal-conducting FEL injectors is a highly suitable approach for civil applications. With advancements in high-power microwave sources and ultra-pure copper processing technologies20,21, enhancing injector beam power by increasing the RF duty cycle has become feasible. However, operating normal-conducting RF injectors at high duty cycles presents new challenges22. High-repetition-rate operation introduces significant thermal loads and other engineering issues that can compromise stability and, in severe cases, threaten equipment safety. Furthermore, the prolonged transient beam loading effect in long-pulse normal-conducting accelerating structures results in performance non-uniformity among bunches within a macropulse, thereby reducing the effective pulse duration. In the specific case of the HUST THz-FEL, given the compact nature of the ITC injector beamline and the scarcity of adjustable parameters, it is necessary to introduce new system-level control variables to guarantee the overall performance of the driving beam. Based on the HUST THz-FEL experimental platform, this paper elucidates a strategy for optimizing the macropulse performance of the ITC injector. We explore and propose a cost-effective improvement scheme, aiming to provide a robust technical foundation for power upgrades and high-power applications in similar normal-conducting injectors. Macropulse beam optimization of the current ITC injector Overview of the injector system The HUST THz-FEL is a long-wavelength free-electron laser (FEL) experimental platform jointly developed by Huazhong University of Science and Technology (HUST) and the University of Science and Technology of China (USTC). Aimed at civil and commercial applications, the facility focuses on researching key technologies for compact, high-power THz sources. Its ultimate objective is to generate radiation in the 1–10 THz frequency range, achieving a peak power of up to 1 MW and an average power of up to 10 W. Figure 1 illustrates the layout of the HUST THz-FEL facility. A 2.6 MeV, picosecond-scale electron beam generated by the microwave electron gun is accelerated by a normal conducting constant-gradient traveling wave (TW) linac. The beam is subsequently transported via a quadrupole triplet to a deflection section composed of two dipole magnets and achromatic quadrupoles which can also be used to measure the emittance23. Finally, the beam enters the undulator and is diverted to the beam dump by a downstream dipole magnet. Preliminary experimental results demonstrate that the injector outputs an electron beam with an energy of 13.7 MeV, an energy spread of 0.41%, and a normalized emittance of 8.8 mm·mrad, satisfying the design specifications24. Schematic layout of the HUST THz-FEL facility19. As depicted in Fig. 1, the ITC injector comprises two standing wave (SW) cavities (Cell 1 and Cell 2) and a TW LINAC. In the current configuration, the ITC and the LINAC share a common RF power source. Correspondingly, to achieve independent regulation of the accelerating field amplitude and phase, two sets of phase shifters and attenuators are installed at the RF feed waveguides of Cell 2 and the LINAC, supporting maximum input powers of 3.56 MW and 20 MW, respectively. For potential future upgrades, the installation of a new, independent power source for the ITC is being considered to enable variable power injection timing. To monitor the beam status effectively, three Fast Current Transformers (FCTs) are installed along the beamline for real-time detection of the beam waveform and intensity; these are positioned upstream and downstream of the TW linac, and at the entrance of the beam dump. The entire facility features a highly compact design, with the overall layout confined within a 3 m × 3 m area. Impact of transient response on macropulse beam performance The ITC injector comprises two standing wave (SW) cavities, designated as Cell1 and Cell2, along with a traveling wave (TW) linac. Cell1 primarily functions to capture and longitudinally compress the electron beam, where the input beam can be approximated as a direct current (DC) source. Conversely, Cell2 is employed for energy spread compensation and further acceleration. Consequently, the energy distribution of the bunches within the final macropulse is predominantly governed by the transient responses and cavity voltage characteristics of Cell2 and the TW linac. Equation (1) describes the temporal evolution of the cavity voltage within Cell225: Here, represents the coupling coefficient between the waveguide and the SW cavity, denotes the equivalent shunt impedance of the cavity, and τ is the decay time constant. and correspond to the injected RF power and phase, respectively, while and represent the macropulse beam current and the beam injection time. As indicated by Eq. (1), for a fixed set of RF parameters, the beam injection time significantly influences the temporal profile of the cavity voltage in Cell2. Consequently, plays a critical role in determining the energy distribution of the bunches within the output macropulse. For the TW linac, the voltage distribution within the macropulse can be derived from the power balance equation. Similar to the SW cavity, this distribution is also dependent on the beam injection time: In Eq. (2), and denote the amplitude of the accelerating electric field and the filling time, respectively. l represents the structural length of the linac. r and Q are the shunt impedance and quality factor of the linac, respectively, and ω is the angular frequency of the RF field. Macropulse optimization results for the existing facility The energy and phase distribution of the macropulse beam output from the LINAC are governed by the accelerating phase. The longitudinal dynamics can be described by the following differential equations: where and represent the normalized phase velocity in the LINAC and the relative velocity of the particles, respectively. During the drift section between the SW cavity and the TW linac, the phase evolution satisfies: Among the adjustable system parameters are the RF power amplitude, RF phase, and macropulse injection timing. Given that the phase of Cell2 is critical for energy spread compensation, it is maintained at a fixed value. Similarly, to ensure the injector meets the output energy requirements, the RF power amplitude is not arbitrarily varied. Consequently, the optimization strategy focuses on fine-tuning the RF phase near the crest and independently adjusting the macropulse injection time. Based on Eqs. (1)–(4) and the specific structural parameters of the HUST THz-FEL, the macropulse parameters at the LINAC exit were numerically calculated. In this model, each bunch within the macropulse is represented by a single macroparticle, assuming relative phase stability among bunches. Figure 2 illustrates the calculated energy and phase distributions of the macropulse under varying RF phases and injection times. Due to the transient beam loading effect, energy and phase variations are observed within the initial 1 of the macropulse, after which the beam tends toward a steady state. Macropulse energy (a) and phase distribution (b) under different injection times and RF phases. Figure 3 presents the experimental comparison of the macropulse waveforms before and after optimization. In the pre-optimization case shown in Fig. 3a, the injection parameters were set to and , whereas in the optimized case shown in Fig. 3b, the parameters were adjusted to and . The experimental results demonstrate that the optimized injection conditions significantly enhance the stability of the macropulse beam. As observed in Fig. 3a, the waveforms detected by the first two FCTs are relatively consistent. However, after passing through the dipole magnet and the energy slit, the macropulse captured by FCT3 exhibits a sharp peak followed by a drop-off, indicating that the subsequent beam failed to propagate effectively to the end of the injector beamline due to energy mismatch. Conversely, the optimized waveform at FCT3 in Fig. 3b is much more stable and exhibits a significantly higher amplitude compared to the pre-optimization case, indicating improved transmission efficiency and beam quality. Macropulse beam waveforms: (a) pre-optimization and (b) post-optimization. Analysis of injector improvements and their impact on macropulse beam performance Rationale for injector upgrades Designed to provide high-quality electron beams for civil and commercial FEL facilities, the ITC injector must strictly adhere to the core requirements of compactness and cost-effectiveness. Consequently, any future facility upgrades should aim to minimize both economic costs and the extent of modification workload. Previous studies26 indicate that the electron beam emitted from the pre-injector exhibits a transverse focusing trend, where the focal point coincides with the position of minimum emittance. Crucially, longitudinal characteristics such as energy spread and bunch length do not deteriorate significantly at this location. Therefore, this specific position is selected as the interface point for the TW linac and focusing solenoids. Furthermore, given that the waveguide transmission line is already equipped with adjustable attenuators and phase shifters, it is advisable to maintain the existing assembly positions of the beamline components to preserve the integrity of the original design layout. Further analysis of Eq. (1) suggests that adjusting the coupling coefficient significantly influences macropulse stability, while employing a split RF power feeding scheme enhances tuning flexibility. It is noteworthy that neither of these proposed modifications requires altering the existing spatial layout of the beamline or waveguide components. In summary, this study proposes a strategy that preserves the original spatial configuration. The upgrade involves converting the fixed-coupling SW structure to an adjustable-coupling mechanism and implementing independent RF power injection for the SW and TW structures. This approach aims to achieve highly flexible macropulse regulation and performance optimization at minimum cost. Impact analysis of adjustable standing wave cavity coupling The standing wave (SW) cavity is designed to boost the electron beam energy from hundreds of keV to approximately 2.6 MeV. To achieve this energy gain, a peak input power of 3.56 MW is required. This high-power RF is magnetically coupled into the cavity through a coupling iris on the cavity wall. There are two primary methods for adjusting the coupling coefficient: modifying the dimensions of the iris or introducing a coupling tuning post. Previous research27 indicates that, for the same tuning range, modifying the iris requires higher machining precision and induces significant shifts in the resonant frequency and other RF parameters. Consequently, the HUST-FEL injector adopts the post scheme. Figure 4a illustrates the structural model of the coupler, while Fig. 4b presents the S11 reflection coefficient curves corresponding to different coupling coefficients. The minor resonant frequency shifts observed during coupling adjustment can be compensated by the cavity tuners installed on the cavity wall. This approach offers greater flexibility and operational ease while minimizing perturbations to other cavity parameters. (a) Model of the adjustable coupler; (b) S-parameters corresponding to different coupling coefficients. However, while increasing the coupling coefficient shortens the transient filling process of the SW cavity, it inevitably reduces the maximum accelerating voltage, thereby lowering the output beam energy. Nevertheless, the proposed system improvement—splitting the RF power feeds for the SW and TW structures—allows for effective compensation. By independently increasing the input power to the SW cavity, the energy deficit caused by the increased coupling coefficient can be fully offset. Based on optimization results from the existing facility, a coupling coefficient of (the measured value) yields relatively stable energy and phase consistency within the injection time range of and RF phase range of . To systematically investigate the impact of the coupling coefficient on macropulse uniformity, a specific operating point (, ) was selected for beam dynamics simulations under varying coupling conditions. The resulting energy and phase distributions within the macropulse are depicted in Fig. 5a,b. Macropulse energy (a) and phase distribution (b) under different coupling coefficients. As shown in Fig. 5a, although the steady-state energy of the macropulse decreases as the coupling coefficient increases, the flatness in the – window improves significantly. Given that a THz oscillator FEL typically requires a macropulse duration of approximately to reach saturation under non-ideal conditions28, the energy inconsistency within the – interval is adopted as the optimization metric. The transient beam segments prior to can be readily eliminated using the energy slit positioned downstream of the dipole magnet. Regarding phase consistency, observations at 2 m from the injector exit (Fig. 5b) reveal similar trends. Under the optimized and combination, the phase distribution of the microbunches within the macropulse converges towards greater uniformity as the coupling coefficient increases. Impact analysis of independent injection timing Theoretical deductions from Eqs. (1) and (2) indicate that the timing difference between the RF pulses of the TW/SW structures and the beam injection significantly influences the transient response, thereby determining the intra-macropulse performance of the injector. Without altering the existing beamline and waveguide layout, an independent RF power source can be introduced for the SW cavity to decouple the timing control of the SW and TW structures. For clarity in the following analysis, let denote the time difference between the beam injection () and the SW RF injection, and denote the difference between and the TW RF injection. We further introduce a variable to represent the inter-structure RF delay. Note that Δ t = 0 corresponds to the baseline scenario where the SW and TW share a common RF source. Consistent with the previous section, the operating parameters were fixed at the optimized point (, ) with the measured coupling coefficient . Beam dynamics simulations were conducted under varying Δ t conditions. The resulting macropulse energy and phase distributions for different Δ t settings are presented in Fig. 6. Macropulse energy (a) and phase distribution (b) under different RF injection time differences (Δ t). As illustrated in Fig. 6a, an increase in Δ t significantly alters the energy profile. Regarding the phase characteristics shown in Fig. 6b, the amplitude of phase variation within the macropulse progressively decreases as Δ t increases. Combining observations from Fig. 6a,b, the configuration with exhibits better uniformity in both energy and phase. The observed improvement in energy consistency is fundamentally due to a slope compensation mechanism. When Δ t > 0, the RF power injection into the SW cavity is delayed relative to the TW linac. Consequently, when the electron beam macropulse traverses the SW cavity, the field is still in its initial build-up phase, exhibiting a positive temporal slope (increasing voltage). In contrast, within the structure, the accelerating field experienced by the beam exhibits a negative slope (decreasing voltage) due to the dominant transient beam loading effect. By optimizing Δ t, the rising edge of the SW field effectively compensates for the drooping field in the TW linac. This mutual compensation flattens the net energy gain profile across the macropulse, thereby enhancing the overall energy stability. Macropulse beam optimization via multi-objective genetic algorithm Introduction of coupling coefficient and injection timing difference as independent variables As demonstrated in the preceding analyses, the accelerating voltage waveforms in both Cell2 and the LINAC are influenced by , , , and , leading to non-uniform energy and phase distributions within the macropulse. To systematically analyze the impact of these parameters on macropulse uniformity, we employ the Non-dominated Sorting Genetic Algorithm III (NSGA-III) to perform multi-objective optimization involving multiple adjustable variables. The consistency of the macropulse, which consists of a train of bunches, is quantified using the root mean square (RMS) values of the statistical distribution. The symbols σ (W) and denote the inconsistency levels of the macropulse beam energy and phase, respectively, where smaller values indicate better uniformity. Here, and represent the steady-state energy and relative phase of the macropulse, respectively. Throughout this work, the summation in Eqs. (5)–(6) is taken over the bunches within s, i.e. the portion of the macropulse that survives the downstream energy slit. The definitions themselves remain valid for any chosen window. First, we investigate the individual effects of and Δ t on macropulse inconsistency while keeping and fixed. The operating point is selected from the previous optimization results29: and . As shown in Fig. 7, the energy inconsistency σ (W) initially decreases and then increases with increasing , reaching a minimum at . Conversely, the phase inconsistency monotonically increases (i.e., phase consistency degrades) as increases. A similar trend is observed when Δ t is treated as the sole independent variable: σ (W) reaches a minimum at , while phase consistency deteriorates monotonically. Variation of energy and phase inconsistency with fixed and : (a) with as the sole independent variable; (b) with Δ t as the sole independent variable. Figure 7 reveals a fundamental trade-off: optimal phase consistency and energy consistency cannot be achieved simultaneously when and are fixed. Furthermore, different combinations of and exert varying influences on macropulse consistency. This represents a classic multi-objective, multi-variable optimization problem, which is well-suited for the NSGA-III algorithm. In practical beam commissioning, maximizing the stable output energy () is prioritized alongside macropulse uniformity. Therefore, we define three optimization objectives: minimizing σ (W), minimizing , and maximizing . Building upon the pre-optimized values of and from the existing facility, we first introduce the coupling coefficient as an adjustable variable (with Δ t=0). Using the NSGA-III algorithm, we mapped the variations in output macropulse energy , σ (W), and across a wide range of (1–6). The resulting three-dimensional Pareto front is depicted in Fig. 8a. Pareto fronts obtained from three variable NSGA-III optimizations. (a) optimization of with fixed; (b) optimization of with fixed. The results indicate that the Pareto front is suboptimal. The regulation capability provided by adding solely as an optimization variable is limited, resulting in a discontinuous Pareto front with local optima in the σ (W)- plane. The stable macropulse energy concentrates around 14–15 MeV, and the energy inconsistency is maintained below 4%. However, this performance comes at the expense of phase consistency, as the values in the solution set are predominantly greater than . Next, we consider introducing Δ t as the optimization variable. As shown in Fig. 8b, the projection of the Pareto front onto the σ (W)- plane exhibits better convergence. In the region with larger , the output stable energy remains relatively constant. However, in the region with smaller (better phase consistency), the stable energy scatters between 12–15 MeV, and the energy consistency deteriorates. This implies that to achieve improved phase consistency, the algorithm must sacrifice both the magnitude of the stable energy and the energy uniformity. Global optimization with four collaborative variables (, , , ) The preceding sections demonstrated that introducing or as individual optimization variables yields limited improvements in macropulse consistency, suggesting that these approaches are constrained by local optima. To overcome these limitations, we propose a global optimization strategy that simultaneously adjusts , , , and . As previously discussed, this four-variable configuration is experimentally feasible through the implementation of the adjustable post and the independent RF power feeding system. The NSGA-III algorithm was employed again to verify this multi-dimensional optimization strategy. The resulting Pareto front, depicted in Fig. 9, exhibits significantly improved convergence compared to the single-variable cases. The solution set is densely clustered near the origin in the projections on all three planes, indicating simultaneous minimization of σ (W) and . The steady-state macropulse energy is highly concentrated around 14 MeV, demonstrating that the overall consistency of the macropulse has been effectively enhanced. The optimized variable ranges are determined as: , , , and . Pareto front obtained using , , , and as optimization variables. The advantage of the four-variable collaborative optimization can be attributed to a “pulse shaping” mechanism. While determines the filling rate (slope) of the cavity voltage, the timing variables () control the synchronization of this rising edge with the beam transient. By simultaneously optimizing these parameters, the system can tailor the temporal profile of the accelerating field to perfectly mirror the inverse of the beam loading transient. Furthermore, provides the necessary degree of freedom to compensate for phase slippage caused by velocity modulation. This multi-dimensional coupling allows for a robust balance between energy flatness and phase stability that is difficult to achieve through single-parameter tuning alone. As evident in Fig. 9, the Pareto front solutions are densely concentrated. In the context of engineering commissioning, a high density of solutions implies high operational robustness, meaning that slight deviations in initial parameter settings will still likely yield a solution within the optimal range. Conversely, sparse regions in the Pareto set (which were more prevalent in previous low-dimensional optimizations) correspond to “knife-edge” operating points where the system performance is highly sensitive to parameter deviations, making them unsuitable for stable operation. In conclusion, implementing independent control over the power sources for the SW cavity and LINAC, combined with adjustable coupling, effectively resolves the issue of large σ (W) and under high duty cycle operation. This approach achieves a synchronized optimization of both energy and phase consistency, providing a robust theoretical and technical basis for the facility upgrade. Virtual experiment results Based on the optimization results of the existing facility and the proposed improvement scheme, virtual verification tests were conducted under various initial conditions. Table 1 summarizes the macropulse consistency performance and steady-state energy output () for these scenarios. Table 1Virtual experimental results for the baseline configuration (No. 1–5), single-variable optimizations (No. 6–7; see Fig. 8), and four-variable global NSGA-III solutions selected from the Pareto front in Fig. 9 (No. 8–10).No.Initial conditionsVirtural experiment results (MeV)0.8085.04.803.2714.630.8590.04.473.1814.620.9092.04.213.6614.610.9396.03.062.9214.611.30105.02.542.4414.611.30105.01.7851.552.252.3714.611.30105.01.6401.302.482.2213.970.81105.11.4361.691.961.5114.450.80103.91.3281.551.881.8714.550.83104.11.4311.581.931.7914.57 Virtual experimental results for the baseline configuration (No. 1–5), single-variable optimizations (No. 6–7; see Fig. 8), and four-variable global NSGA-III solutions selected from the Pareto front in Fig. 9 (No. 8–10). The scenarios labeled No.–No. represent the initial parameter settings derived from the optimization of the existing facility. In these cases, the system is constrained to share a common RF power source for both the TW linac and the SW cavity (i.e., ), and the coupling coefficient is fixed at . Consequently, only the RF injection timing and phase are adjustable. The results indicate that while the baseline configuration achieves high steady-state energy stability, the intra-macropulse energy and phase distributions require further optimization. Among these, No. represents the optimal baseline result; although its steady-state energy is not the highest, it simultaneously minimizes both energy and phase inconsistency. No.–No. are single-variable optimizations derived from No. by independently varying one additional parameter: in No., where delayed SW injection reduces both energy and phase inconsistency; in No., where a reduced coupling coefficient trades a modest decrease in steady-state energy for substantially improved macropulse flatness. No.–No. are globally optimized solutions selected from the four-variable Pareto front (Fig. 9), achieving 30–50% improvement in phase and energy inconsistency over the baseline (No.–No. ). The slight reduction in steady-state energy observed in these cases can be compensated by increasing the SW cavity RF power, which is made possible by its independent power source. Figure 10 illustrates the macropulse energy and phase distributions corresponding to the optimized scenarios No. –No. listed in Table 1. Macropulse energy (a) and phase distribution (b) under the optimal parameter sets (No. 6–No. 10). The virtual test results align with the preliminary theoretical analysis: macropulse consistency is improved when the SW power injection lags behind the TW injection (Δ t > 0). Furthermore, to achieve minimal energy inconsistency, the optimal coupling coefficient converges within the range of 1–3. In summary, the NSGA-III algorithm does not merely seek a single extremum for each objective function but provides a set of globally balanced solutions, offering a robust operating regime for the facility. Conclusion To address the significant transient beam loading effects observed in the HUST THz-FEL ITC injector under high duty cycle operation, this study proposes a system improvement scheme combining independent RF power feeding with an adjustable coupling coefficient for the standing wave (SW) cavity. Based on this upgraded hardware configuration, a multi-variable collaborative optimization model was established using the Non-dominated Sorting Genetic Algorithm III (NSGA-III). This model systematically investigates the influence mechanisms of the coupling coefficient, RF injection timing, and phase on macropulse beam quality. Our analysis reveals that single-variable regulation is insufficient to balance the trade-off between energy gain and macropulse uniformity. In contrast, global collaborative optimization significantly improves the convergence and density of the Pareto solution set. Physically, this enhancement is attributed to a slope compensation mechanism: by optimizing the delay between the SW and TW RF pulses ( vs. ) and adjusting the SW filling rate via the coupling coefficient (), the rising edge of the SW field effectively counteracts the transient field droop caused by beam loading in the TW linac. Virtual experiments demonstrate that the optimized system maintains a stable energy output of approximately 14.5 MeV while reducing intra-macropulse energy inconsistency to approximately 1.5–2.0% and phase variation to approximately 1.9–2.5, representing a 30–50% improvement over the baseline configuration. Beyond the beam dynamics optimization, successfully transitioning this proposed scheme to physical operation necessitates addressing several key hardware considerations. First, the introduction of a movable coupling tuner increases the risk of RF field leakage near the tuner region; this can be effectively managed by employing choke-mode structures or optimized tuner geometries, as detailed in our prior analysis27. Second, the suppression of multipacting during long RF pulses is a critical operational concern, which is currently being systematically evaluated through 3D particle-in-cell (PIC) simulations. Finally, it is worth noting that all proposed upgrades are designed to strictly preserve the original spatial layout of the beamline and waveguides, which will significantly limit modification complexity and hardware costs during actual implementation. Consequently, this work demonstrates the effectiveness of the global parameter optimization strategy in mitigating transient instabilities in compact normal-conducting injectors. The resulting robust operating parameters provide a valuable reference for the future engineering commissioning of the proposed injector upgrade. Finally, it should be noted that the proposed hardware modifications, including the adjustable coupler and independent RF power feeds, remain at the design stage; experimental bench tests and beam commissioning will be required to verify the simulated performance. Author contributions H.H. wrote the original draft and conducted the simulations. Y.X. contributed to the theoretical derivation and simulations. Y.Z. and Y.W. reviewed and edited the manuscript. K.L. supervised the experiments. T.H. conceived the concept and proposed the methodology. All authors reviewed the manuscript. Funding This work was supported by the National Natural Science Foundation of China (No.12575164 and No.12341501). Data availability The datasets generated during and analysed during the current study are available from the corresponding author on reasonable request. Code availability The MATLAB codes for beam dynamics simulations and the NSGA-III multi-objective optimization framework developed in this study have been deposited in a Zenodo repository (https://doi.org/10.5281/zenodo.20606962). Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Kaifeng Liu, Email: kfliuhust@hust.edu.cn. Tongning Hu, Email: TongningHu@hust.edu.cn. References Associated Data Data Availability Statement The datasets generated during and analysed during the current study are available from the corresponding author on reasonable request. The MATLAB codes for beam dynamics simulations and the NSGA-III multi-objective optimization framework developed in this study have been deposited in a Zenodo repository (https://doi.org/10.5281/zenodo.20606962).