Average Probability Research Articles

Numerical comparison of exhaled particle dispersion and infection probability in hospital wards heated by mixing ventilation and impinging jet ventilation

Impinging jet ventilation has sufficient supply momentum to overcome thermal buoyancy, with the potential to reduce cross-infection risk in winter. Using an Eulerian-Lagrangian approach validated by experiments, this study compared the exhaled particle dispersion in a two-bed hospital ward heated by impinging jet ventilation and mixing ventilation. Exhaled particles with ten different diameters were tracked under four typical air changes per hour. Particle concentration distribution and residence time were compared. A revised Wells-Riley model was used to calculate the airborne infection probability of coronavirus disease 2019, with the help of a custom field function in Fluent software. The results indicated that impinging jet ventilation increased the air velocity in the occupied zone compared with mixing ventilation. The higher air velocity facilitated the control and removal of fine and medium particles, resulting in lower indoor particle concentration and shorter residence time for 0.5 – 10 μm particles under impinging jet ventilation. With the same air changes per hour, the average infection probability of coronavirus disease 2019 at the breathing zone of susceptible individuals under mixing ventilation could be 1.3 to 12.8 times higher than that under impinging jet ventilation. Additionally, the local infection probability at the face of susceptible individuals was lower under impinging jet ventilation. This study showed that impinging jet ventilation can be implemented in hospital wards to reduce the cross-infection during winter.

Unveiling radii and neutron skins of unstable atomic nuclei via nuclear collisions

Total reaction, interaction, and charge-changing cross sections, which are kinds of cross sections standing for total nuclear collision probability in medium-to high-energy region from a few to several hundred MeV, have been extensively utilized to probe nuclear sizes especially for unstable nuclei. In this mini review, experimental techniques and recent findings from these cross sections are briefly overviewed. Additionally, two new methods to extract neutron skin thickness solely from the above cross sections are explained: One is utilizing the energy and isospin dependence of the total reaction cross sections, and the other is the combination of the total reaction and charge-changing cross section measurements.

Energy Optimization and Secure Aware Spectrum Allocation Using Hybrid Meta-Heuristic Algorithms in Cognitive Radio Network

In cognitive radio networks (CRN), spectrum allocation or channel allocation is a major concern for the efficient use of the available spectrum and is explored in detail. To improve spectrum efficiency, great effort must be put into channel allocation. Because ineffective channel assignment often results in interference between SUs and PUs in CRN, channel allocation requires great care. In addition, both SU network data and power transmission powers cannot exceed the predetermined maximum interference threshold for the PU quality-of-service (QoS). With the expected spread of CRN, security concerns are increasing, energy efficiency is a primary issue of major focus in cognitive networks. Therefore, energy-aware channel allocation against jamming attacks is considered the objective of this work. Initially, the optimal channel or spectrum is assigned using the Chaotic Water Strider Algorithm (CWSA). To solve the problem of power optimization, it is necessary to use the Osprey Optimization Algorithm (OOA). MATLAB is used as the platform for implementing the proposed scheme. Several factors are taken into consideration when evaluating the performance of the proposed scheme, including network lifetime, energy efficiency, delay, throughput, packet loss, delivery rate, spectrum allocation, energy consumption, total failure probability, and signal-to-noise ratio (SNR).

Response shift results of quantitative research using patient-reported outcome measures: a meta-regression analysis.

Our objectives were to identify characteristics of response shift studies using patient-reported outcomes (PROMs) that explain variability in (1) the detection and (2) the magnitude of response shift effects. We conducted a systematic review of quantitative studies published before June 2023. First, two-level multivariable logistic regression models (effect- and sample-levels) were used to explain variability in the probability of finding a response shift effect. Second, variability in effect sizes (standardized mean differences) was investigated with 3-level meta-regression models (participant-, effect- and sample-levels). Explanatory variables identified via the purposeful selection methodology included response shift method and type, and population-, study design-, PROM- and study-quality characteristics. First, logistic regression analysis of 5597 effects from 206 samples in 171 studies identified variables explaining 41.5% of the effect-level variance, while no variables explained sample-level variance. The average probability of response shift detection is 0.20 (95% CI: 0.17-0.28). Variation in detection was predominantly explained by response shift methods and type (recalibration vs. reprioritization/reconceptualization). Second, effect sizes were analyzed for 769 effects from 114 samples and 96 studies based on the then-test and structural equation modeling methods. Meta-regression analysis identified variables explaining 11.6% of the effect-level variance and 26.4% of the sample-level variance, with an average effect size of 0.30 (95% CI: 0.26-0.34). Response shift detection is influenced by study design and methods. Insights into the variables explaining response shift effects can be used to interpret results of other comparable studies using PROMs and inform the design of future response shift studies.

Bayesan Model to Predict R Status After Neoadjuvant Therapy in Pancreatic Cancer.

To build a Bayesian approach-based model to predict the success of surgical exploration post-neoadjuvant treatment. Pancreatic cancer (PDAC) is best treated with radical surgery and chemotherapy, offering the greatest chance of survival. Surgery after neoadjuvant treatment (NAT) is indicated in the absence of progression, knowing the limits in accurately predicting resectability with traditional radiology. R Status being a pathological parameter, it can be assessed only after surgery. Patients successfully resected for histologically confirmed PDAC after NAT for BR and LA disease were included, with attention to the predictors of R status from the existing literature. The Bayesian logistic regression model was estimated for predicting the R1 status. The area under curve (AUC) of the average posterior probability of R1 was calculated and results were reported considering the 95% posterior credible intervals for the odds ratios, along with the probability of direction. The final model demonstrated a commendable AUC value of 0.72, indicating good performance. The likelihood of positive margins was associated with older age, higher ASA score, the presence of venous and/or arterial involvement at preoperative radiology, tumor location within the pancreatic body, a lack of tumor size reduction post-NAT, and the persistence of an elevated Ca19.9 value. A Bayesian approach using only preoperative items is firstly used with good performance to predict R Status in pancreatic cancer patients who underwent resection after neoadjuvant therapy.

Estimation of a~New Asymmetry Based Process Capability Index 𝐶𝑐 for Gamma Distribution

Abstract In this work, we provide five different classical estimation methods for the gamma distribution and unknown parameters, which are used to estimate the newly constructed process capability index (PCI) C c C_{c} . Next, we use the aforementioned estimation techniques to consider five bootstrap confidence intervals: standard bootstrap (SB), percentile bootstrap (PB), student’s 𝑡 bootstrap (STB), bias-corrected percentile bootstrap (BCPB), and bias-corrected accelerated bootstrap (BCAB). These allow us to construct confidence intervals of the PCI C c C_{c} . Using the suggested techniques of estimate for small, moderate, and large sample sizes, Monte Carlo simulations are run to compare the performances of the bootstrap confidence intervals’ average widths and coverage probabilities. Simulation results indicate that MLE gives the least AW for all BCIs (SB, PB, BCPB, TB, and BCAB) when compared to all other estimation approaches evaluated in this paper. Further, simulation results demonstrate that, for almost all sample sizes and all of the examined estimation approaches, BCAB CI generates smaller AW and larger CP than their counterparts. Finally, for illustrative purposes, three real data sets are analysed.

Scheme selection with probabilistic multiple objectives optimization

This article presents the scheme/alternative selections with probabilistic multiple objectives optimization (PMOO). In the PMOO assessment, all response objectives (attributes) are divided into beneficial or unbeneficial types according to their function and preference to equivalently contribute their partial preferable probabilities simultaneously, the total preferable probability of an alternative is the multiplication of its all partial preferable probabilities, which determines the optimal evaluation uniquely and comparatively. The application examples contain a personnel selection and a production quantity optimal control. The former is for an engineer position in a software development department from five alternative candidates withstanding seven optimal criteria (response objectives) comparatively, and the latter aims to get an optimal production quantity with a higher profit rate and lower final cost. In the personnel selection, the seven optimal response objectives (criteria) include relevant education, work experience in the field, relevant certificates, level of presentation and communication, ability of personnel management, capabilities of planning and organization, and expressiveness of foreign language. All these seven response objectives are attributed to the beneficial type of attribute to join the assessment. As to the production quantity control, the profit rate belongs to the beneficial type of objective, while the final cost is attributed to the unbeneficial type of attribute. The evaluated results reveal that the optimal alternate for the personnel selection is candidate No. 4, and the optimum production quantity is at x* = 54 items. The achievement of the present article indicates the validity of the corresponding approach and algorithm with rationality. The novelty of this work is to reflect the simultaneity of the response objectives (criteria) in the optimal system by using probabilistic multiple objectives optimization, and all response objectives either beneficial type or unbeneficial type are evaluated separately in an equivalent manner.

Disentangling the effects of intermittent faecal shedding and imperfect test sensitivity on the microscopy-based detection of gut parasites in stool samples.

Gut-parasite transmission often involves faecal shedding, and detecting parasites in stool samples remains the cornerstone of diagnosis. However, not all samples drawn from infected hosts contain parasites (because of intermittent shedding), and no test can detect the target parasites in 100% of parasite-bearing samples (because of imperfect sensitivity). Disentangling the effects of intermittent shedding and imperfect sensitivity on pathogen detection would help us better understand transmission dynamics, disease epidemiology, and diagnostic-test performance. Using paediatric Giardia infections as a case-study, here we illustrate a hierarchical-modelling approach to separately estimating the probabilities of host-level infection (Ψ); stool-sample-level shedding, given infection (θ); and test-level detection, given infection and shedding (p). We collected 1-3 stool samples, in consecutive weeks, from 276 children. Samples (413 overall) were independently examined, via standard sedimentation/optical microscopy, by a senior parasitologist and a junior, trained student (826 tests overall). Using replicate test results and multilevel hierarchical models, we estimated per-sample Giardia shedding probability at [Formula: see text] and observer-specific test sensitivities at [Formula: see text] and [Formula: see text]. Gender-specific infection-frequency estimates were [Formula: see text] and [Formula: see text]. Had we used a (hypothetical) Perfect Test with 100% narrow-sense sensitivity (pPT = 1.0), the average probability of detecting Giardia in a sample drawn from an infected child (Ψ = 1.0) would have been Pr(d|i,PT) = Ψ×θ×pPT≈1.0×0.44×1.0≈0.44. Because no test can be >100% sensitive, Pr(d|i) (which measures clinical sensitivity) can only be brought above ~ 0.44 by tinkering with the availability of Giardia in stool samples (i.e., θ); for example, drawing-and-pooling 3 replicate samples would yield [Formula: see text]. By allowing separate estimation (and modelling) of pathogen-shedding probabilities, the approach we illustrate provides a means to study pathogen transmission cycles and dynamics in unprecedented detail. Separate estimation (and modelling) of true test sensitivity, moreover, may cast new light on the performance of diagnostic tests and procedures, whether novel or routine-practice.

Turbine performance and wind energy potential using probability distribution in Africa

The energy requirement in the world is increasing day by day. The non-renewable energy production is one of the reasons for climate change. The present study explores the selection of possible locations to establish a wind power plant in Ethiopia. The possible locations from Ethiopia that were considered in this study are Weraelu, Enewari and Mekdela. The wind speed characterization, wind direction and wind power densities were analyzed at Waraelu, Enewari and Mekdela. The results of this study reveals that Enewari is the best suitable place to establish the wind power plants among Waraelu, Enewari and Mekdela. In Enewari site the maximum wind speed at 10m height is 6.11 m/s. 75% of wind direction at Enewari site is from North to North West and North to North East. Weibull distribution analysis was incorporated to find average wind speed and probability density among five years of wind speed data. The average wind speed and probability density over five years of Enewari site is 5 m/s and 18.5%. The wind power densities were determined by Weibull method and the maximum wind power obtained from Enewari site is 495.33 w/m^2.

Probabilistic Analysis of Slope Stability Considering Random Fields

The study of slope stability is of paramount importance for society, as it enables the evaluation of slope security and the establishment of alert parameters to mitigate financial losses and, in severe cases, prevent human casualties. Improving alert criteria necessitates a comprehensive assessment not only of the slope's safety factor but also its probability of failure. Therefore, the use of probabilistic techniques becomes imperative, with the Monte Carlo Method (MCM) emerging as a prevalent choice in geotechnical investigations. Employing MCM for slope stability involves modeling the target profile while varying crucial material parameters—such as friction angle, cohesion, and specific weight—in each iteration. As a result, the outcome yields a distribution of safety factors, aiding in the assessment of both average safety levels and failure probabilities. Failure probability is determined by counting simulations with safety factors below unity. Conventional approaches identified in the literature often employ uniform profiles in successive rounds, overlooking the spatial distribution of parameters. Hence, this study introduces a case analysis where failure probability is quantified through MCM, integrating random fields in each iteration. For assessing variability, the PLAXIS software was utilized, employing classical limit equilibrium methods alongside the LAS technique (Local Average Subdivision). This facilitates a comparative analysis of probabilities derived from distinct equilibrium methodologies. The case study will focus on a slope within Brazilian territory, where a comprehensive geotechnical survey encompassing field and laboratory assessments has been conducted.

Inter-pixel cross-talk as background to two-photon interference effects in SPAD arrays

Cross-talk is a well-known feature of single-photon avalanche detectors. It is especially important to account for this effect in applications involving temporal coincidences of two or more photons registered by the sensor since in this case the cross-talk may mimic the useful signal. In this work, we characterize the cross-talk of the LinoSPAD2 detector, as well as perform joint measurements of the cross-talk and Hanbury Brown-Twiss two-photon interference, comparing and cross-calibrating both effects. With a median dark count rate of 125 cps/pixel, we report the average cross-talk probability of 0.22% for the nearest neighbor and also observe a long-range cross-talk of the order 2 · 10-5% for channels separated by up to 20 pixels.

Circuits and mechanisms for TMS-induced corticospinal waves: Connecting sensitivity analysis to the network graph.

Transcranial magnetic stimulation (TMS) is a non-invasive, FDA-cleared treatment for neuropsychiatric disorders with broad potential for new applications, but the neural circuits that are engaged during TMS are still poorly understood. Recordings of neural activity from the corticospinal tract provide a direct readout of the response of motor cortex to TMS, and therefore a new opportunity to model neural circuit dynamics. The study goal was to use epidural recordings from the cervical spine of human subjects to develop a computational model of a motor cortical macrocolumn through which the mechanisms underlying the response to TMS, including direct and indirect waves, could be investigated. An in-depth sensitivity analysis was conducted to identify important pathways, and machine learning was used to identify common circuit features among these pathways. Sensitivity analysis identified neuron types that preferentially contributed to single corticospinal waves. Single wave preference could be predicted using the average connection probability of all possible paths between the activated neuron type and L5 pyramidal tract neurons (PTNs). For these activations, the total conduction delay of the shortest path to L5 PTNs determined the latency of the corticospinal wave. Finally, there were multiple neuron type activations that could preferentially modulate a particular corticospinal wave. The results support the hypothesis that different pathways of circuit activation contribute to different corticospinal waves with participation of both excitatory and inhibitory neurons. Moreover, activation of both afferents to the motor cortex as well as specific neuron types within the motor cortex initiated different I-waves, and the results were interpreted to propose the cortical origins of afferents that may give rise to certain I-waves. The methodology provides a workflow for performing computationally tractable sensitivity analyses on complex models and relating the results to the network structure to both identify and understand mechanisms underlying the response to acute stimulation.

HARD-VOTING DAN SOFT-VOTING CLASSIFIER: MODEL KLASIFIKASI RISIKO KEMATIAN PADA PASIEN GAGAL GINJAL KRONIK

Chronic kidney failure is a serious disease that can lead to death if not detected and treated early. This study aims to predict the risk of death in hospitalized chronic kidney failure patients using ensemble machine learning methods, specifically hard-voting and soft-voting. The voting classifier is used to combine predictions from several classification models, where hard-voting makes decisions based on the majority vote, and soft-voting considers the average prediction probability. However, with imbalanced data, classification models tend to be biased toward the majority class. To address this, the synthetic minority oversampling technique (SMOTE) is applied to balance the class distribution. The model's performance is evaluated using accuracy, precision, and specificity metrics. The results of the study show that the hard-voting classifier provided the best performance with an accuracy of 85,156%, precision of 86,726%, and specificity of 89,908%, outperforming the soft-voting classifier. The use of SMOTE proved to improve prediction for the minority class, which is crucial in detecting high-risk patients who may die during hospitalization. This approach is expected to aid in the early detection and better management of hospitalized chronic kidney failure patients, potentially reducing mortality rates from this disease.

Parametric optimization and determination in machining processes by means of probabilistic multi-objective optimization

In the present article, it attempts to present the determination of optimal parameters of machining processes by means of probabilistic multi-objective optimization (PMOO), in which the optimal objectives (attributes) are fundamentally divided into beneficial type and unbeneficial type, moreover all attributes of both beneficial type and unbeneficial type are evaluated separately with equivalent manner to get their partial preferable probability. Finally, the total preferable probability of each alternative is obtained by the product of all partial preferable probabilities, which is the unique and decisive representative of the alternative to join the competitive optimization, the optimum alternative is with the highest total preferable probability. An example of parametric optimization and determination of aerospace component with Electro Chemical Machining (ECM) is taken to illuminate the procedure. In the case of ECM, the current, voltage, and feed rate are as the optimal parameters to be investigated, while Material Removal Rate (MRR), and Surface Roughness (SR) are the optimal objective responses to be measured. The experimental runs were designed using an L27 Taguchi orthogonal array. In the assessment of PMOO for ECM, the objective MRR belongs to the beneficial attribute, and the objective SR is as the unbeneficial attribute. The novelty of this work is to reflect the simultaneity and the irreplacement of optimization of objectives MRR and SR in the optimal system. The evaluated results reveal that the optimized experimental scheme is the alternative 8, which is with the optimal responses of MRR of 280.112 g/min and SR of 0.45 mm, the corresponding optimum experimental parameters are voltage of 12 V, electrolyte flow rate of 12 m/s and tool feed rate of 0.4 mm/min, respectively. The achievement of the present article indicates the validity of the corresponding approach and algorithm.

Overqualification Among Second-Generation Children of Immigrants in the Swedish Labour Market

Research on the children of immigrants born in the host country (G2) consistently reveals disparities between their educational achievements and labour market outcomes compared to the majority population. This study provides new insights into understanding this disparity by examining patterns of overqualification—i.e., a downward educational mismatch—among the G2. Specifically, it explores 1) how overqualification patterns differ between the G2, foreign-born immigrants (G1), and the majority population and 2) how overqualification patterns vary across ten G2 ancestry groups compared to the majority population. Utilizing Swedish total population register data and linear probability models, this study estimates the probability of overqualification across different immigrant generations and ancestry groups, employing the Realised Matches method to measure overqualification. The results indicate that while G2 individuals have a lower probability of experiencing overqualification compared to G1, they face moderately higher probabilities of overqualification than the majority population—up to 19% higher. This disparity is particularly pronounced among G2 individuals with tertiary education and those of Iranian, Middle Eastern and North African, and Other Non-Western origins, with up to 39% higher probabilities. These findings suggest that G2 individuals, particularly those of non-Western origins, encounter significant challenges in translating their educational qualifications into commensurate employment within the Swedish labour market.

Susceptibility and risk to inhalation of pathogen-laden aerosol in large public spaces: Evidence from Fangcang Shelter Hospitals under multiple ventilation rates

To enhance the prevention and control of pandemic respiratory infections, we constructed an infection risk prediction model for pathogen inhalation based on the dose-response relationship with reference to the long-distance transmission chain of the SARS-CoV-2 pathogen, applying it to the Fangcang Shelter Hospital (FSH). The model quantitatively describes key processes of pathogens shedding, airborne transmission, suspension, inhalation by susceptible individuals, and lung deposition, thus improving the resolution and accuracy of the results. This study considered four ventilation rates and quantitatively assessed their impact on inhalation infection risk. Results indicate that the infection risk within the multi-patient shelter unit is unevenly distributed, with the maximum probability (3.66 %) being more than 30 times higher than the minimum probability (0.10 %) at a ventilation rate of 8 ACH. Poor ventilation (6 ACH) significantly increases average infection probability, with a rise of 35.96 % compared to the average probability (1.14 %) at 8 ACH. However, excessive ventilation (12 ACH) led to diminishing returns on ventilation performance. Lastly, we also found that poor ventilation was a sufficient and non-essential condition for higher infection probability. Our findings can be extended to similar large-scale scenarios, offering positive support for the sustainable development of society.

Radar Echo Extrapolation Based on Translator Coding and Decoding Conditional Generation Adversarial Network

In response to the shortcomings of current spatiotemporal prediction models, which frequently encounter difficulties in temporal feature extraction and the forecasting of medium to high echo intensity regions over extended sequences, this study presents a novel model for radar echo extrapolation that combines a translator encoder-decoder architecture with a spatiotemporal dual-discriminator conditional generative adversarial network (STD-TranslatorNet). Initially, an image reconstruction network is established as the generator, employing a combination of a temporal attention unit (TAU) and an encoder–decoder framework. Within this architecture, both intra-frame static attention and inter-frame dynamic attention mechanisms are utilized to derive attention weights across image channels, thereby effectively capturing the temporal evolution of time series images. This approach enhances the network’s capacity to comprehend local spatial features alongside global temporal dynamics. The encoder–decoder configuration further bolsters the network’s proficiency in feature extraction through image reconstruction. Subsequently, the spatiotemporal dual discriminator is crafted to encapsulate both temporal correlations and spatial attributes within the generated image sequences. This design serves to effectively steer the generator’s output, thereby augmenting the realism of the produced images. Lastly, a composite multi-loss function is proposed to enhance the network’s capability to model intricate spatiotemporal evolving radar echo data, facilitating a more comprehensive assessment of the quality of the generated images, which in turn fortifies the network’s robustness. Experimental findings derived from the standard radar echo dataset (SRAD) reveal that the proposed radar echo extrapolation technique exhibits superior performance, with average critical success index (CSI) and probability of detection (POD) metrics per frame increasing by 6.9% and 7.6%, respectively, in comparison to prior methodologies.

Self-Reported Unfair Police Stops and Skin Tone, Ethnic Origin, and Citizenship Status among Latino Adults

While a growing literature has documented vast ethnoracial inequalities tied to the occurrence and outcomes of police contact, less research has examined the patterning of perceived unfair police stops among Latino populations. The present study relies on insights from tri-racial stratification theory to examine associations between perceived unfair police stops and skin tone, ethnic origin, and citizenship status among Latino adults. Data come from the Pew Research Center’s nationally representative 2021 National Survey of Latinos ( n = 3,019). We regress whether a respondent perceived being stopped unfairly by police in the past year on the Yadon-Ostfeld Skin Color Scale (i.e., 1–10, 10 = Darkest), ethnic origin (i.e., Puerto Rican, Cuban, Dominican, Salvadoran, Spaniard, Central American, South American, or other), and citizenship status (i.e., U.S. citizen or not). Results showed that having a darker skin tone and U.S. citizenship were associated with higher odds of experiencing a police stop perceived as unfair. Latinos with the darkest skin tones had nearly a one in five chance of a perceived unfair police stop. Average probabilities of a perceived unfair stop were higher for Puerto Ricans, Cubans, and Salvadorans than others followed by Mexicans/Chicanos. Dominicans’ probability was lower than five of the eight other groups. Findings identify heterogeneity among Latino populations such that variation in skin tone, citizenship status, and ethnic origin independently factor into self-reported unfair police stops. Research examining ethnoracial disparities in criminal legal contact and its consequences would benefit from more fully disaggregating panethnic groups.

Statistical theory of neutron-induced nuclear fission and of heavy-ion fusion

For both reactions I use an approach similar to that of compound-nucleus reaction theory. For neutron-induced fission, the compound system generated by absorption of the neutron and the nuclear system near the scission point are described as two statistically independent systems governed by random-matrix theory. The systems are connected either by a barrier penetration factor or by a set of transition states above the barrier. Each system is coupled to a different set of channels. An analogous model is used for heavy-ion fusion. Under the assumption that (seen from the entrance channel) the system on the other side of the barrier is in the regime of strongly overlapping resonances, closed-form analytical expressions for the total probability for fission and for fusion are obtained for each value of spin and parity. Parts of these expressions can be calculated reliably within existing compound-nucleus reaction theory. The remaining parts are the probabilities for passage through or over the barrier. These may be determined theoretically from the liquid-drop model or experimentally from total fission or fusion cross sections. Published by the American Physical Society 2024

Resource Allocation in UAV-D2D Networks: A Scalable Heterogeneous Multi-Agent Deep Reinforcement Learning Approach

In unmanned aerial vehicle (UAV)-assisted device-to-device (D2D) caching networks, the uncertainty from unpredictable content demands and variable user positions poses a significant challenge for traditional optimization methods, often making them impractical. Multi-agent deep reinforcement learning (MADRL) offers significant advantages in optimizing multi-agent system decisions and serves as an effective and practical alternative. However, its application in large-scale dynamic environments is severely limited by the curse of dimensionality and communication overhead. To resolve this problem, we develop a scalable heterogeneous multi-agent mean-field actor-critic (SH-MAMFAC) framework. The framework treats ground users (GUs) and UAVs as distinct agents and designs cooperative rewards to convert the resource allocation problem into a fully cooperative game, enhancing global network performance. We also implement a mixed-action mapping strategy to handle discrete and continuous action spaces. A mean-field MADRL framework is introduced to minimize individual agent training loads while enhancing total cache hit probability (CHP). The simulation results show that our algorithm improves CHP and reduces transmission delay. A comparative analysis with existing mainstream deep reinforcement learning (DRL) algorithms shows that SH-MAMFAC significantly reduces training time and maintains high CHP as GU count grows. Additionally, by comparing with SH-MAMFAC variants that do not include trajectory optimization or power control, the proposed joint design scheme significantly reduces transmission delay.