Integrity Review Report: Research on Dynamic Replenishment Vehicle Routing Optimization for Unmanned Vehicles in Same-Day Delivery Service (DOI: 10.16381/j.cnki.issn1003-207x.2024.2034)

Academic fraud report · Geng Detector

Summary

This report presents a text-based integrity review of the paper 'Research on Dynamic Replenishment Vehicle Routing Optimization for Unmanned Vehicles in Same-Day Delivery Service' by Xiao Jianhua et al., published in Chinese Journal of Management Science (DOI: 10.16381/j.cnki.issn1003-207x.2024.2034). The overall verdict is highly suspicious. Three significant issues were identified through cross-checking the narrative text against tabular data. First, the claimed minimum savings of 21.5%, 25.5%, and 28% for the AVADDR_A algorithm contradict Table 4, where minimum savings reach only 19.5% (case 1.2, λ₁=0.5) and 24.7% (case 2.2, λ₂=0.6). Second, the reported standard deviation mean of 0.9 for the 100-order scenario does not match Table 3, where the actual mean is 1.1775; the figure 0.9 more closely corresponds to the 80-order scenario, suggesting copy-paste error. Third, the ALNS algorithm shows variance of zero across 10 runs in Table 2, which is statistically implausible for a stochastic metaheuristic. Image-based analysis was not possible due to text-only input. Confidence is high for the textual/numerical discrepancies, while image-related concerns remain untested.

Verdict

Highly suspicious. The narrative text contains multiple numerical claims that are directly contradicted by the paper's own tables, including overstated performance gains and statistics that appear to be copied from a different experimental scenario. Additional concerns about data plausibility (zero variance for a stochastic algorithm) warrant formal investigation. Image-based analysis was not possible in this review.

Key findings

Overstated performance gains in narrative vs. Table 4: The text claims minimum savings of 21.5%, 25.5%, and 28% under different order rates, but Table 4 shows minimum savings of only 19.5% (case 1.2 at λ₁=0.5) and 24.7% (case 2.2 at λ₂=0.6) for the Dec.3 column.
Likely copy-paste of standard deviation statistics: The text reports a standard deviation mean of 0.9 for the 100-order scenario, but the values for 100.1–100.4 in Table 3 are 1.05, 0.90, 1.40, 1.36, yielding a true mean of 1.1775. The 80-order scenario (0.64, 1.44, 0.89, 0.85; mean ≈ 0.955) is closer to the reported 0.9, suggesting cross-contamination of text between scenarios.
Implausible zero variance in Table 2: The ALNS heuristic, which contains random destroy/repair operators, is reported to yield zero variance across 10 independent runs, with mean values matching optimal values to two decimal places (e.g., 23.06, 19.22, 33.89). This is statistically anomalous for a stochastic metaheuristic.

Evidence highlights

Table 4 quotations (Dec.3 column, AVADDR_A savings):
λ₁=0.5: case 1.1 = 20.8%, case 1.2 = 19.5%, case 1.3 = 21.8%
λ₂=0.6: case 2.1 = 24.8%, case 2.2 = 24.7%
Narrative claim for λ₂=0.6 is "minimum 25.5%"; the actual minimum is 24.7%.
Table 3 standard deviations (100-order scenario): 1.05, 0.90, 1.40, 1.36; arithmetic mean = 1.1775, not 0.9 as stated in the text. The 80-order standard deviations (0.64, 1.44, 0.89, 0.85) average ≈ 0.955.
Table 2 (small-scale): All reported ALNS mean-best matches identical to two decimal places; reported variance = 0 across 10 runs.

Notes

DOI: 10.16381/j.cnki.issn1003-207x.2024.2034. All numeric evidence is preserved as reported in the original review.
This review was conducted on text only; no pixel-level image analysis (e.g., ELA, Forensically) was performed on Figures 1–8. Image integrity remains untested.
The discrepancies between narrative and tabular data could in principle stem from transcription errors rather than deliberate fabrication; however, the pattern (a systematically optimistic narrative that under-reports the worst-case benchmark in every cited scenario) is more consistent with intentional overstatement than with random typographical mistakes.
The reported zero variance for ALNS is not impossible in principle (e.g., small instances with sufficiently strong operators may converge deterministically), but the matching of mean to optimal value to two decimals across all instances is improbable and would require verification with raw run logs.
Recommend requesting raw 10-run logs from the authors and submitting a formal concern to the editorial office of Chinese Journal of Management Science (《中国管理科学》).