Oil and fat traceability methods and systems based on chemical composition evidence chains

Abstract

This invention discloses a method and system for tracing the origin of oils and fats based on a chemical composition evidence chain. It collects spectral and chemical indicators from oil samples at each node of the supply chain, extracts changes in degradation offset direction and sterol oxidation ratio using the thermodynamic sequence of oil degradation as a reference, identifies directional reversals and gradual drifts across nodes, and forms directional anomaly markers and oil source mixing markers. Combining degradation timestamp deviation quantification and trace contaminant facility fingerprint matching, a chemical evidence chain rating with dual-channel signal cross-confirmation is established. The rating conclusion and the hash value of the preceding node are jointly encoded to form a differential hash packet, which is then locked as a traceability chain block record through multi-party consensus voting. Topological clustering is performed on abnormal nodes in the chain block to extract fraud cluster offset types and organizational forgery degrees, integrating and outputting a chemical traceability report. This report distinguishes between two different forms: acute adulteration and gradual organizational fraud, providing verifiable chemical evidence for full-chain accountability of oil and fat products.

Description

Technical Field

[0001] This invention relates to the field of oil and fat product testing technology, and in particular to a method and system for tracing the origin of oils and fats based on a chain of evidence of chemical composition. Background Technology

[0002] The illegal inflow of catering oil into the edible oil supply chain is a persistent food safety risk. During the distribution process, oils are mixed into legitimate channels through multiple intermediaries. This adulteration occurs at multiple points, with the responsible parties scattered. Relying solely on end-point sample testing can only determine the state of the oil at that specific point, but cannot pinpoint which distribution stage the adulteration occurred in or by which party. Regulatory enforcement faces difficulties in gathering evidence across the entire supply chain.

[0003] Current approaches primarily proceed in two directions: at the physicochemical testing level, oil quality is identified by comparing indicators such as sterol composition and polar substance content; however, the test results are only for individual samples and lack continuous tracking of chemical change trends across different nodes. At the record-keeping level, electronic ledgers or blockchain are introduced to solidify circulation information; however, the stored evidence is independent of the chemical test results, lacking technical binding and verification, and the possibility of subsequent data addition or tampering cannot be ruled out. These two approaches have not achieved technical synergy and integration, making it difficult to establish complete chemical evidence in multi-party joint traceability scenarios. Summary of the Invention

[0004] This invention discloses a method and system for tracing the origin of oils and fats based on a chemical composition evidence chain. It performs dual-path spectral and chemical detection on oil samples at each node of the supply chain, extracting fingerprint offset direction and oxidized sterol ratios to identify directional reversals and cross-node progressive drift, forming directional anomaly markers and oil source mixing markers. Anomaly nodes are cross-verified using degradation timestamp deviation and trace contaminant facility fingerprints to establish a chemical evidence chain rating. The rating conclusion is jointly encoded with the hash value of the preceding node and subjected to multi-party consensus voting to form a traceability chain block record. Topological clustering is performed on the abnormal nodes in the chain block, and a chemical traceability report is output.

[0005] The first aspect of this invention proposes a method for tracing the origin of oils based on a chain of evidence of chemical composition, comprising the following steps: Collect optical fingerprint data and chemical detection data, collect the offset direction of each node from the optical fingerprint data to generate the fingerprint offset direction, and determine the oxidized sterol ratio based on the change of sterol oxidation rate of each node collected from the chemical detection data. Based on the fingerprint offset direction, the direction reversal between nodes is identified to form a direction anomaly mark, and the oxidized sterol ratio is used to identify the multi-node progressive drift feature of the ratio to establish an oil source mixing mark. The degradation timestamp deviation is obtained by comparing the offset amplitude of the directional anomaly marker with the degradation rate over time. The processing imprint is generated by analyzing the combination characteristics of trace pollutants at each node for the oil source mixed marker. Based on the degradation timestamp deviation and the processing imprint, the chemical evidence chain rating is established by identifying the overlap rate of dual signal triggering at the same node. Based on the chemical evidence chain rating, the fingerprint offset direction is collected and combined with the preceding hash encoding to obtain a forward differential hash packet. The forward differential hash packet is used to verify the consensus voting of multiple chemical conclusions to form a chemical consensus record. The traceability chain block record is determined from the chemical consensus record to verify the drift dynamic consistency of the node. Based on the traceability chain block record, a fraud cluster set is obtained by identifying the topological association density between abnormal nodes. An anomaly type report is obtained by extracting the offset type and the degree of organization forgery from the fraud cluster set. The anomaly type report is then integrated with the chemical evidence chain rating to output a chemical traceability report.

[0006] A second aspect of this invention proposes a traceability system for oils and fats based on a chain of evidence of chemical composition, comprising: The data acquisition module is used to acquire optical fingerprint data and chemical detection data, to collect the offset direction of each node from the optical fingerprint data to generate the fingerprint offset direction amount, and to determine the oxidized sterol ratio based on the sterol oxidation rate change of each node collected from the chemical detection data. Anomaly detection module is used to identify directional reversal between nodes based on the fingerprint offset direction to form directional anomaly markers, and to establish oil source mixing markers by identifying multi-node progressive drift characteristics of the ratio based on the oxidized sterol ratio. The evidence fusion module is used to compare the offset magnitude of the directional anomaly marker with the degradation rate over time to obtain the degradation timestamp deviation, analyze the trace pollutant combination characteristics of each node for the oil source mixed marker to generate the processing imprint, and identify the double signal trigger overlap rate of the same node based on the degradation timestamp deviation and the processing imprint to establish a chemical evidence chain rating. The hash evidence storage module is used to collect the fingerprint offset direction based on the chemical evidence chain rating and obtain a forward differential hash packet by combining the forward differential hash packet with the forward differential hash packet to verify the consensus voting of multiple chemical conclusions to form a chemical consensus record. The module then uses the chemical consensus record to verify the drift dynamics consistency of the node and determine the traceability chain block record. The report output module is used to obtain a fraud cluster set based on the topological association density between abnormal nodes identified by the traceability chain block records, extract the offset type and organizational forgery degree of the fraud cluster set to obtain an anomaly type report, and integrate the anomaly type report with the chemical evidence chain rating to output a chemical traceability report.

[0007] The beneficial effects of this invention are reflected in the following points: First, this invention uses the thermodynamic sequence constraint of oil degradation as a reference. From the existing single-node detection framework, which can only output spectral deviation and sterol ratio signals to determine quality compliance, it further extracts two types of fraudulent features: the reversal of spectral peak directions between adjacent nodes and the cross-node drift of sterol oxidation rates. These correspond to two distinct fraud modes: acute adulteration and gradual mixing, respectively. This allows the detection conclusion to transcend single-node quality evaluation and extend to the morphological identification and location of supply chain fraud. Second, degradation timestamp deviation and trace contaminant facility fingerprints are two measurement paths with independent physical mechanisms. The probability of simultaneous triggering at the same node is far lower than the false triggering rate of any single path. This invention establishes a chemical evidence chain rating based on the overlap rate of the same node, ensuring that the chemical conclusions cannot be structurally interfered with by manipulating a single detection dimension. This solves the weakness of existing traceability schemes that rely on single-path chemical judgments and whose conclusion credibility is easily affected by sample processing manipulation. Finally, this case embeds the chemical evidence chain rating conclusion into the node hash calculation input, enabling subsequent modifications to chemical data to propagate downstream along the chain and trigger chain verification failures. This solves the structural defects of existing blockchain evidence storage schemes, such as the separation of chemical conclusions from on-chain records and the failure of data tampering to trigger chain anomalies. Furthermore, it performs topological clustering on abnormal nodes in the chain, summarizing the scattered evidence conclusions of each node into a qualitative output of the overall form of cross-batch fraud clusters, filling the methodological gap that node-level chemical evidence cannot be elevated to the identification of supply chain-level collaborative fraud patterns. Attached Figure Description

[0008] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0009] Figure 1 This is a flowchart illustrating the oil traceability method based on the chemical composition evidence chain of this invention.

[0010] Figure 2 This is a structural block diagram of the oil traceability system based on the chemical composition evidence chain of this invention. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0013] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0014] The technical solutions of the embodiments of this application will be described below.

[0015] like Figure 1 As shown, this embodiment of the invention provides a method for tracing the origin of oils based on a chain of evidence of chemical composition, including the following steps S101-S105: Step S101: Collect optical fingerprint data and chemical detection data; collect the offset direction of each node from the optical fingerprint data to generate the fingerprint offset direction amount; and collect the sterol oxidation rate change of each node based on the chemical detection data to determine the oxidized sterol ratio amount.

[0016] Specifically, optical fingerprint data and chemical detection data were collected. The optical fingerprint data included near-infrared hyperspectral and FTIR spectra. Hyperspectral band characteristic values ​​and GC-MS and ICP quantitative values ​​were used to establish a joint calibration model via partial least squares regression to predict multi-instrument chemical characteristic values ​​using spectral analysis. Among these, the carbonyl group (1746 cm⁻¹) of the ester after repeated high-temperature treatment of the oil was analyzed. -1 The inverse double bond (965 cm⁻¹) decays at a rate of approximately -0.12% / h. -1 The enhancement rate is approximately +0.18% / h, with the two peaks in opposite directions and a difference in amplitude exceeding 2 times. Optical fingerprint data must be processed at a rate of 4cm. -1Only wavenumber resolution acquisition can simultaneously distinguish the changing directions of two peaks. Insufficient resolution will lead to overlapping of adjacent peaks and area integration errors exceeding 7%. Before acquiring optical fingerprint data, the sample must be placed in a constant temperature environment of 25±0.5°C for 30 minutes. The average of three parallel scans is taken for each node. If repeatability is not met, resampling is triggered. Only optical fingerprint data with qualified temperature control and repeatability are included in the fingerprint offset direction collection process. Chemical detection data are quantified by GC-MS for six components in the oil at each node: β-sitosterol, campesterol, stigmasterol (three phytosterols), and 7α-hydroxysterol, 7β-hydroxysterol, and 7-ketosterol (three oxidation products). Fe is simultaneously quantified by ICP-OES. 2+ / Fe 3+ And the concentration ratios of transition metal elements such as Cu and Ni. The concentration matrix of the six components at each node of the chemical detection data is the basis for calculating the node sequence of oxidized sterol ratios. For each node, 0.1g of oil is taken, saponified, and then silylated for derivatization. The derivatization conversion rate must be no less than 95%. Insufficient conversion rate will cause the peak areas of the six components in the chemical detection data to remain low, resulting in an underestimation of the oxidized sterol ratio. Fe obtained from ICP-OES 2+ / Fe 3+ Concentration ratios, as independent chemical markers of high-temperature heating history, together with GC-MS sterol oxidation data and hyperspectral combined model predictions, constitute three lines of evidence that comprehensively reflect the degree of lipid oxidation, heating history, and heavy metal enrichment characteristics.

[0017] In some embodiments, the step of generating a fingerprint offset direction from the offset directions of each node collected from the optical fingerprint data includes: extracting the intensity values ​​of key absorption bands of each node from the optical fingerprint data to obtain a node absorption intensity set; calculating the standardized deviation of each band from the mean of normal nodes to form a node intensity deviation set; identifying the position of band deviations that violate the thermodynamic order of degradation based on the node intensity deviation set to form a deviation sign vector; and identifying pathway type reversal events between adjacent nodes based on the deviation sign vector to generate a fingerprint offset direction.

[0018] The node absorption intensity set is obtained by extracting the intensity values ​​of key absorption bands of each node from the optical fingerprint data. 1160cm -1 (Glycerin skeleton), 965cm -1 (Reverse double key), 1746cm -1 (Ester carbonyl), 3006cm -1The four spectral bands (cis double bonds) reflect four independent dimensions of change in the oil adulteration scenario: glycerol skeleton integrity, trans bond formation, ester bond degradation, and loss of unsaturation. This covers the main chemical characteristics of frying oil mixed with fresh oil. The nodal absorption intensity set retains only the peak areas of the four key bands, removing invalid dimensions from the remaining bands. The peak areas of each node in the optical fingerprint data are calculated using a baseline-corrected integration method. When noise spikes appear within the integration interval, linear interpolation is performed to eliminate them before integration. The four-dimensional peak area results constitute the corresponding node entries in the nodal absorption intensity set. In the optical fingerprint data, the baseline of the sample emulsified due to transport bumps is generally raised across all bands, reaching 1746 cm⁻¹. -1 Peak area can be overestimated by more than 10%. Multivariate scattering correction must be performed before extracting the node absorption intensity set. Nodes with baseline errors exceeding the acceptable threshold after correction are marked as potentially emulsified. The confidence weight of the four-dimensional intensity vector of potentially emulsified nodes is reduced to 0.7 times before being included in the node absorption intensity set to prevent baseline distortion, contamination, and bias in direction judgment. For example, in a certain warehouse, oil at a certain node experienced premature aging due to repeated heating, with a peak area of ​​965 cm⁻¹. -1 An abnormal enhancement of the trans bond peak was detected. After accurately extracting this peak from the node absorption intensity set, the node intensity deviation set detected deviation features that should have appeared at later nodes in this dimension, thus providing an early warning of abnormal behavior at this node. The optical fingerprint data shows adjacent nodes of the same type at 1746 cm⁻¹. -1 When a significant jump occurs in the peak area, an additional jump label is added within the batch. When there are too many jump nodes, a low consistency warning is issued for the node absorption intensity data. The average deviation benchmark of the warning batch is expanded to the historical data of the previous and next batches and recalculated. The bias effect of local pollution in a single batch on the benchmark value is thus suppressed.

[0019] The standardized deviations of each band from the mean of normal nodal absorption intensity are statistically analyzed to form a nodal intensity deviation set. Soybean oil and palm oil absorb at 1746 cm⁻¹. -1 The mean absorbance of normal nodes differs by approximately 0.04 absorbance units. When calculating the deviation directly without distinguishing between oil types, palm oil nodes will be frequently mislabeled as abnormal. The node absorption intensity set must be calculated by dividing the standard deviation of normal nodes in each band by the Z-score within the same oil type. When the absolute value of the Z-score exceeds 2.0, the deviation has exceeded the 95% confidence interval. The node intensity deviation set stores the four-dimensional deviation in the form of Z-scores. For example, a batch of soybean oil at the 5th circulation node at 1746 cm⁻¹... -1The peak area is 0.82, while the average of similar normal nodes is about 0.95, the standard deviation is about 0.03, and the Z-score is about -4.3, far exceeding the 2.0 threshold. The node intensity deviation set accurately marks the abnormal rate degradation of ester bonds at this node. During the deviation symbol construction stage, this Z-score contributes a negative sign bit and triggers the enhanced deviation label. For the low consistency warning batch of node absorption intensity set, the Z-score is recalculated after expanding the historical average. If the absolute value of the Z-score still exceeds 2.5 under the expanded baseline, the enhanced deviation label is marked in the node intensity deviation set. The symbol weight of the enhanced deviation labeled node is increased by 1.3 times during the deviation symbol vector generation stage, reducing the omission of true anomalies due to the offset of the baseline average. The Z-scores of suspected emulsified nodes in the node absorption intensity set are converted with a confidence weight of 0.7 and then included in the node intensity deviation set. When the absolute value of the converted Z-score still exceeds 2.0, an emulsification conversion deviation label is marked. The nodes marked with emulsification conversion deviation are processed separately and run in parallel with normal confidence nodes. The two deviation paths are independently merged into the deviation symbol vector, and the overall judgment confidence of the node intensity deviation set is not diluted by nodes with high uncertainty.

[0020] Based on the nodal intensity deviation set, the positions of band deviations that violate the thermodynamic order of degradation are identified and a deviation sign vector is constructed. Ester bonds (1746 cm⁻¹) in the normal frying degradation pathway... -1 ) preceding the inverted key (965cm) -1 Significant changes occurred; after inferior oil was mixed with fresh oil, the concentration of trans fatty acids suddenly increased while the ester bonds had not yet significantly degraded, and the node strength deviation was concentrated at 965 cm. -1 The absolute value of the peak Z-score exceeds 2.0 and is 1746 cm⁻¹. -1 A peak Z-score absolute value below 1.0 constitutes a thermodynamic disorder, and this node is marked with a thermodynamic disorder identifier in the deviation sign vector. The positive and negative signs of the four-dimensional Z-scores in the node intensity deviation set are arranged sequentially to form a four-digit sign sequence. Of the 16 possible arrangements, only 3 correspond to normal degradation pathways, while the remaining 13 are all offset states. The deviation sign vector records the sign sequence of each node and its corresponding pathway type. For example, a restaurant's inbound node detected a value of 965 cm⁻¹. -1 The peak Z-score is +2.8, 1746 cm⁻¹. -1 The peak Z-score is only -0.6, and the symbol sequence (+--+) belongs to one of the 13 abnormal pathways, which is consistent with the typical deviation characteristics of inferior oil and fresh oil mixed in a ratio of approximately 3:7. The deviation symbol vector records this abnormal pathway type at this node for use in direction reversal determination. The node intensity deviation set strengthens the weight of the deviation identifier node symbol sequence in the deviation symbol vector by 1.3 times, reducing the missed direction reversal due to high deviation nodes having the same weight as ordinary nodes. When the proportion of thermodynamic disorder identifier nodes in the entire batch exceeds 25%, the deviation symbol vector metadata is supplemented with a thermodynamically abnormal batch warning. The direction reversal determination threshold for the warning batch is tightened by 20% during the fingerprint offset direction amount aggregation stage, which matches the actual characteristics of the high abnormal degradation distribution density of the batch.

[0021] The fingerprint offset direction is generated based on the reversal event of the pathway type between adjacent nodes identified by the deviation symbol vector. Among the 16 combinations of four-band symbol sequences, there are only 3 normal degradation pathways. When the adjacent node of the deviation symbol vector switches from a normal pathway to an abnormal pathway, it is determined as a direction reversal event. The fingerprint offset direction records the reversal starting node number, the pathway type before and after the reversal, and the average Z-score of each band. The reversal magnitude is quantified by the average absolute value of the Z-score of the involved bands. For example, in a certain batch of edible oil, nodes 1 to 6 all show normal pathways (symbol sequence ---+), nodes 7 to 11 show consecutive (++-+) abnormal pathways, and node 12 returns to normal. The fingerprint offset direction records two direction reversals at nodes 6→7 and 11→12. The spatial coordinates of the adulteration behavior concentrated in the segment from node 7 to 11 are thus locked, providing accurate fraud link location for the direction anomaly marker identification stage. The reversal threshold for thermodynamic disorder marker nodes in the deviation sign vector is lowered by one level: a 1-bit sign difference between adjacent nodes is considered a direction reversal, while a regular node requires at least 2 bits. Disordered nodes are already high-confidence anomalous signals, and a strict threshold would miss early doping information carried by single-band directional changes. After the deviation sign vector is collected across the entire batch, the fingerprint offset direction is statistically analyzed to determine the total number of reversals within the batch and the dominant anomalous pathway type. When the dominant anomalous pathway type events account for more than 60%, a dominant anomalous type field is added to the fingerprint offset direction data. This field serves as a priority verification signal during the direction anomaly marker identification stage, reducing the number of localization iterations when multiple anomalies coexist by approximately 40%.

[0022] The oxidized sterol ratio is determined by collecting changes in sterol oxidation rates at each node based on chemical detection data. The oxidation pathway of β-sitosterol follows a fixed order, generating 7α-hydroxysterol, 7β-hydroxysterol, and 7-ketosterol sequentially. The ratio of the sum of the peak areas of these three products to the peak area of ​​β-sitosterol is the current oxidized sterol ratio at that node. The difference between this ratio and the upstream adjacent node in the chemical detection data is divided by the actual flow time between the two nodes to obtain the measured oxidation rate of that node. The oxidized sterol ratio (i.e., the oxidation rate baseline deviation multiple sequence) is generated by collecting the ratio of the measured oxidation rate of each node in the entire batch to the natural baseline of 0.02% / day according to the supply chain time sequence, forming a complete evolution record of the oxidation rate along the flow chain. The flow time between nodes is taken from the supply chain flow ledger. When the time is missing, the average flow time of the batch is used as a substitute. The baseline deviation multiple of the substitute node is downgraded in the oxidized sterol ratio and a time estimation label is added. Under normal flow conditions, the oxidized sterol ratio at each node in the chemical detection data is typically below 0.08. When it exceeds 0.12, the accumulated amount has exceeded the natural degradation limit by 50%. When the deviation from the baseline at a node exceeds 3 times, a high oxidation warning label is added to that node, and the adjacent sampling volume for the warning node is increased to 1.5 times the original. When the high oxidation warning node accounts for more than 30% of the total number of nodes in the batch, a high-risk batch label is added to the oxidized sterol ratio data. During the oil source mixing marker establishment stage, the drift identification window is shortened from 5 nodes to 3 nodes. When a batch experiences a sudden increase in the ratio of adjacent nodes due to a change in oil source, the 3-node window can effectively capture such short-range concentrated drifts, while the 5-node window will miss the report due to the dilution effect of normal nodes before and after.

[0023] Step S102: Based on the fingerprint offset direction, identify the direction reversal between nodes to form a direction anomaly mark, and use the oxidized sterol ratio to identify the multi-node progressive drift feature of the ratio to establish an oil source mixing mark.

[0024] Specifically, directional anomaly markers are formed by identifying directional reversals between nodes based on fingerprint offset direction quantities. The abnormal pathways generated by real doping events typically cover 2 to 4 consecutive nodes through which the doped batch flows. The false alarm rate for isolated reversals at a single node exceeds 40%. Therefore, the fingerprint offset direction quantity must be used as the criterion for establishing a directional anomaly marker based on continuity rather than single-node jumps. The pathway types recorded for each reversal event in the fingerprint offset direction quantity are compared one by one between adjacent nodes. When the current node and the preceding node have the same pathway type and are both abnormal pathways, a continuous anomaly determination is triggered. When the continuous determination spans more than two nodes, the starting position is confirmed as a valid directional anomaly segment. The segment's start and end node numbers and the corresponding abnormal pathway type constitute the basic entries for the directional anomaly marker. Isolated reversals at a single node in the fingerprint offset direction quantity are included in the suspected labeling field of the directional anomaly marker and stored. Suspected labeling nodes do not participate in the fraud confidence level calculation, but when they have a spatial proximity to a valid directional anomaly segment in the same batch, their deviation direction characteristics can be included as supplementary evidence in the chemical evidence chain rating reference. For example, if a batch of edible oil exhibits a continuous (++-+) abnormal pathway from node 5 to node 8, and returns to normal at node 9, the directional anomaly marker identifies nodes 5 to 8 as a valid directional anomaly segment. If only node 5 shows an anomaly and node 6 recovers, it is classified as a suspected anomaly and does not constitute a valid directional anomaly marker entry on its own. For batches where the fingerprint offset direction is the dominant anomaly type, dense warning markers are added to high-density reversal positions during the directional anomaly marker identification stage. The range of the dense warning segment extends to adjacent suspected marker nodes to prevent the valid directional anomaly segment from being truncated due to boundary effects.

[0025] In some embodiments, the step of using the oxidized sterol ratio to identify the multi-node progressive drift characteristics of the ratio and establishing an oil-source mixing label includes: extracting the node ratio increment sequence of adjacent batches using the oxidized sterol ratio to generate a node increment sequence; calculating the cumulative increment mean within a sliding window to determine the cumulative drift amount of the window based on the node increment sequence; identifying accelerated drift windows exceeding the natural degradation limit based on the cumulative drift amount of the window to form an over-limit drift segment; and collecting the drift direction and amplitude of each over-limit drift segment to establish an oil-source mixing label based on the over-limit drift segment.

[0026] A node increment sequence is generated by extracting the ratio increment sequence of adjacent batches of oxidized sterols from the oxidized sterol ratio. The flow time between nodes in the supply chain can vary by several times; the batch change interval within a warehouse is typically 12 to 24 hours, while the interval between nodes in cross-city delivery can be 3 to 5 days. The same original ratio increment at short-interval nodes indicates high-speed abnormal oxidation, while at long-interval nodes it represents normal accumulation. The node increment sequence requires dividing the original ratio increment of oxidized sterols at each node by the corresponding flow time, normalizing it to a daily average increment, and then arranging it chronologically. This eliminates the regular bias in drift rate estimation caused by differences in node time intervals. The confidence weight of the daily average increment of the oxidized sterol ratio indicator node in the node increment sequence is reduced to 0.7 times, thus quantitatively compressing the rate calculation bias introduced by the time estimation error. The location of the high oxidation sterol ratio warning node is marked with a high increment source identifier in the node increment sequence. The weight of the high increment source node is increased by 1.2 times during the sliding window statistical stage to ensure that the contribution of the local high-speed oxidation node to the window mean is not diluted by the surrounding normal nodes. For example, if the oxidation rate of the 6th node in a batch suddenly increases due to deterioration of storage conditions, the node increment sequence retains the amplification effect of the node on the window mean by the high increment source identifier, avoiding the omission of the high-speed drift segment after dilution by the surrounding normal nodes. After the node increment sequence is generated, the whole batch quantile statistics are performed. Nodes with the daily average increment exceeding the 90th percentile are marked with a high quantile identifier. When the number of nodes with high quantile identifiers exceeds 20% of the total number of batches, the node increment sequence metadata is appended with a high-speed oxidation batch warning. During the window cumulative drift calculation stage, the warning batch applies an additional 1.1 times correction coefficient to the natural drift upper limit to match the baseline offset of the overall increase in the over-limit signal in the high-speed oxidation batch.

[0027] The cumulative drift of the window is determined by the average cumulative increment within a sliding window of the node increment sequence. The daily average increment of a single node fluctuates by approximately ±30% due to sampling error. The standard error of the average of the 5-node window is only 1 / √5 (approximately 13%) that of a single node. The statistical smoothing effect significantly reduces the threshold for the true oxidation acceleration signal to emerge from the noise background. The node increment sequence is shifted forward node by node with a standard window of 5 nodes (narrowed to 3 nodes for high-risk batches). The average daily increment of each window serves as a robust estimate of the average daily oxidation rate of oils during the corresponding period for the window's cumulative drift. Nodes with high increment sources in the node increment sequence have been weighted by 1.2 times during the window statistical stage. The estimated window position of the cumulative drift in the window containing nodes with high increment sources is closer to the true acceleration level than the unweighted version, reflecting the actual contribution of local high-speed oxidation nodes. When a high-speed oxidation batch warning exists in the node incremental sequence, the natural drift upper limit is multiplied by a correction factor of 1.1 before being compared with the window cumulative drift amount. This offsets the interference of the overall rise in the high-speed oxidation batch benchmark on the over-limit identification. After correction, the over-limit judgment sensitivity of the window cumulative drift amount sequence is restored to the same level as the normal batch. After the window cumulative drift amount is arranged with the mean values ​​of each window, when the difference between adjacent windows exceeds twice the standard deviation of the batch mean, the window jump position is marked. The jump position indicates that the node may be a node of oil source switching or sudden change in sampling conditions. The window cumulative drift amount is marked with a jump source identifier at the jump position to distinguish between normal continuous drift and false over-limit signals caused by sudden changes when verifying the boundary of the over-limit drift section. Since the number of adjacent nodes at the beginning and end of the node incremental sequence batch is insufficient for the window length, the actual mean of the number of available nodes is used to calculate the window cumulative drift amount. When the sample size of the edge window is insufficient, an edge window identifier is added.

[0028] For example, the step of identifying accelerated drift windows exceeding the natural degradation limit based on the cumulative drift amount of the window to form an over-limit drift segment includes: analyzing the oil type characteristics of the starting node of each sliding window based on the cumulative drift amount of the window to obtain an oil type identifier set; using the oil type identifier set to retrieve the maximum natural rate of sterol oxidation for the corresponding oil type to determine the natural drift limit; calculating the proportion by which the cumulative drift amount of the window exceeds the limit based on the natural drift limit to obtain a drift over-limit index; and collecting continuous window segments with a continuously greater than zero drift over-limit index based on the drift over-limit index to form an over-limit drift segment.

[0029] Oil type identifier sets were obtained by analyzing the oil type characteristics of the starting nodes of each sliding window based on the cumulative drift amount. Soybean oil has a β-sitosterol content of approximately 120 to 140 mg / 100g, while palm oil has a total sterol content of only approximately 50 to 70 mg / 100g. The oxidation rate benchmark at unit concentration differs by approximately 2 to 3 times. If the same natural drift upper limit is used uniformly without distinguishing between oil types, a large number of exceeding-limit segments will be missed in palm oil batches due to the higher upper limit, while a large number of misclassifications will occur in soybean oil batches due to the lower upper limit. The oil type identifier set completes the oil type classification by extracting the sterol composition characteristics of the starting nodes of each window, and establishes a type index for accurate matching of the natural drift upper limit. For cumulative drift in each window, soybean oil is classified as follows: when the peak area ratio of stigmasterol to β-sitosterol at the starting node of each window is greater than 0.25; rapeseed oil is classified as follows: when the ratio of campesterol to β-sitosterol is greater than 0.30; and palm oil is classified as follows: when β-sitosterol is dominant and the total sterol content is less than 70 mg / 100 g. These three main oil types cover more than 90% of common supply chain adulteration scenarios. Nodes whose sterol composition characteristics do not meet any of the above rules are marked as mixed oil types in the oil type identifier set. When the oil type identifier sets of the starting nodes of adjacent windows are inconsistent, an oil type switching event is marked. One window on each side of the switching position is weighted by interpolation of the two oil type upper limits during the natural drift upper limit calculation stage. The oil type identifier set adds an oil type switching event identifier to the switching position to distinguish between normal oil batch replacement and sterol composition changes caused by adulteration. The natural drift limit of the mixed oil type identifier node in the oil type identifier set is conservatively estimated by the minimum value of the upper limit of all identified oil types in the batch. The conservative strategy reduces the probability of the mixed oil type node being missed due to the window cumulative drift exceeding the limit caused by the upper limit being selected too high.

[0030] The natural drift upper limit is determined by retrieving the maximum natural rate of sterol oxidation for the corresponding oil type using an oil type identifier set. The natural drift upper limit for soybean oil stored at 25°C is approximately 0.020% / day, for rapeseed oil approximately 0.018% / day, and for palm oil approximately 0.035% / day, a difference of about two times. Incorrect classification of the oil type identifier set will cause the natural drift upper limit to deviate by the corresponding magnitude. The natural drift upper limit is obtained by matching the oil type-temperature lookup table window by window using the oil type identifier set node type identifier and storage temperature as a two-dimensional index. Nodes without temperature records are substituted with the batch average storage temperature. The lookup table is segmented with a 5°C step size, and the natural drift upper limit is obtained by linear interpolation of the actual window temperature within the step size. When a temperature sensor malfunction causes the entire batch temperature to be missing, the natural drift upper limit is switched to a conservative estimate of the minimum value of all temperature ranges for the corresponding oil type. A batch-wide temperature missing coverage indicator is added to the corresponding batch. The natural drift upper limit performs additional weighting on the drift exceedance index corresponding to the temperature missing node to prevent underestimation of the upper limit from leading to a generally high exceedance judgment ratio. The natural drift upper limit of the oil type identifier for the oil type switching event identifier node is taken as the weighted average of the corresponding values ​​of the two oil types. The weight is determined by the oil type confidence of the nearest measured nodes on both sides of the switching position. The weighted result falls between the upper limits of the two oil types to prevent sudden changes in the upper limit at the switching point from distorting the local continuity of the drift over-limit index sequence. After the natural drift upper limit is collected for each window in the entire batch, a monotonicity check is performed. When the difference between the upper limits of adjacent windows exceeds 0.005% / day, an upper limit jump mark is marked. The drift over-limit index calculation at the jump position uses the upper limit of the average of the two windows before and after to replace the original value, eliminating the interference of the upper limit jump on the over-limit boundary positioning.

[0031] The drift exceedance index is obtained by calculating the percentage of cumulative drift exceeding the natural drift limit within a given window. The drift exceedance index E is calculated using the following formula: E = (DL) / L, where D is the cumulative drift within a single window (% / day), and L is the natural drift limit for the corresponding window (% / day). A positive E value indicates that the measured oxidation rate in the current window is higher than the natural degradation limit. Using L as the denominator, rather than the absolute difference, ensures direct comparability of the drift exceedance index between different oil types, such as rapeseed oil (L≈0.018% / day) and palm oil (L≈0.035% / day). When E reaches 0.5, the measured oxidation rate exceeds the natural limit of the same type of oil by 50%. For nodes lacking the natural drift limit temperature indicator, a conservative and lower upper limit estimate has been used, resulting in a higher drift exceedance index. For these nodes, the drift exceedance index is reduced to 0.8 times during the exceedance drift segment identification stage to prevent the underestimation of the upper limit due to temperature deficiency from universally increasing the exceedance judgment ratio. For natural drift upper limit abrupt changes, the drift exceedance index is calculated using the upper limit of the average of one window before and after the abrupt change, maintaining consistency with the calculation logic for non-abrupt changes. This ensures that the drift exceedance index at abrupt changes does not introduce artificially created local extremes due to sudden changes in the upper limit. For batches with missing natural drift upper limit temperatures, the drift exceedance index is calculated using a conservative upper limit across the entire batch. This results in an overall larger drift exceedance index sequence. During the exceedance drift segment identification stage, the exceedance judgment threshold for such batches must be adjusted upwards to offset the overall offset introduced by the conservative calculation. The adjustment amount is determined by the estimated value of the conservatively larger average of the drift exceedance index for the entire batch. When the proportion of positive windows in the entire batch of drift exceedance index exceeds 50%, a high exceedance rate warning is added to the metadata. The exclusion rules for isolated exceedance windows in high exceedance rate warning batches are relaxed. When the drift exceedance index of one window on each side of an isolated window is positive, it is no longer excluded but merged into the adjacent exceedance drift segment.

[0032] Over-limit drift segments are formed by continuously collecting over-limit indexes with an over-limit index that is consistently greater than zero. Sampling errors may cause individual windows to have random positive values ​​in the over-limit index within the range of ±0.1. When the over-limit index is consistently greater than zero for two or more consecutive windows, the cumulative signal strength exceeds the statistical interpretation range of random fluctuations. The boundary start point of the over-limit drift segment is determined by the starting position of this continuous positive value sequence. The current over-limit drift segment terminates when the over-limit index turns negative or zero. Over-limit drift segment boundaries are re-evaluated using the original, unweighted value sign when the over-limit index is reduced (temperature missing markers are weighted to 0.8). This prevents the weighting operation from changing the over-limit attribution of boundary nodes. The over-limit drift segment boundary is located based on the original over-limit direction, not the adjusted magnitude after weighting. Isolated positive values ​​of the drift over-limit index are classified as suspected over-limit entries in regular batches, while batches with high over-limit rate warnings are handled according to relaxed rules. When both sides of the isolated window have positive drift over-limit indices, they are merged into adjacent over-limit drift segments. The two handling logics are distinguished by the source field of each entry in the over-limit drift segment. During tracing, it is possible to distinguish between regular continuous over-limit segments and relaxed judgment segments of high over-limit rate batches. After the over-limit judgment threshold is increased in the batch over-limit drift segment identification stage of the drift over-limit index missing temperature full batch coverage, the actual drift over-limit index threshold participating in boundary confirmation is increased from 0 to the value corresponding to the batch correction coefficient. The number of over-limit drift segment entries after the increase is usually lower than that of the uncorrected version, and the difference is the conservative estimate of the number of misjudged over-limit entries introduced. When the average drift over-limit index of each segment exceeds 0.5, a high-intensity over-limit marker is added. The high-intensity over-limit marker segment corresponds to acute doping classification in the oil source mixing mark establishment stage, while the segment with an average drift over-limit index below 0.5 corresponds to gradual mixing classification. The two types of classification are weighted by a differentiated mixing intensity index, and the difference in mixing type is thus accurately reflected in the oil source mixing mark.

[0033] Oil source mixing markers are established by collecting the drift direction and amplitude of each out-of-limit drift segment. When the peak value of an out-of-limit drift segment is located in the first 1 / 3 of the segment, it indicates that a large amount of doping has been injected at the beginning of the flow segment and is subsequently diluted. A peak value located in the middle or later segment indicates that the doping proportion accumulates gradually along the flow path. The peak position field and drift amplitude sequence of the out-of-limit drift segment together determine the classification of mixing type and mixing intensity in the oil source mixing marker. The product of the out-of-limit amplitude of the peak value in each out-of-limit drift segment and the number of persistent nodes constitutes the mixing intensity index. When the mixing intensity index is more than twice the historical average of the batch, it indicates that the proportion or scale of highly oxidized oil mixed in that segment exceeds historically common levels. The oil source mixing marker uses the mixing intensity index as the quantitative feature of each segment to collect out-of-limit drift information for the entire batch. When the drift direction within the excess drift range continuously increases, it indicates that the contamination ratio is increasing as the supply chain progresses. For example, in one batch, the ratio monotonically increases along the distribution chain due to long-term contamination with catering oil in a downstream distribution warehouse. The drift direction field of the excess drift range captures this monotonically increasing trend, and the oil source contamination marker adds a progressively expanding contamination indicator to this type of range. For excess drift ranges with acute contamination, an acute contamination indicator is added to the oil source contamination marker. Segments with acute contamination indicators correspond to a higher intentional fraud score in the chemical evidence chain rating stage, distinct from the batch management negligence score corresponding to progressive contamination types. After the oil source mixing marker is generated, the total number of drift segments exceeding the limit and the average mixing intensity index of the entire batch are counted. When the average value exceeds twice the historical average value of the batch, the oil source mixing marker metadata is appended with a high-intensity mixing batch identifier. The high-intensity mixing batch triggers mandatory manual verification during the traceability chain block recording stage. When the gradual expansion mixing identifier and the acute doping identifier coexist in a single batch, the oil source mixing marker records the fraud evolution stage according to the time sequence of the two types of identifiers. The acute first and then gradual corresponds to the initial injection and continuous dilution mode, and the gradual first and then acute corresponds to the long-term mixing and then clearing and replenishing mode.

[0034] Step S103: The deviation of the directional anomaly marker is compared with the degradation rate over time to obtain the degradation timestamp deviation. The trace pollutant combination characteristics of each node are analyzed for oil source mixed markers to generate processing imprint. Based on the degradation timestamp deviation and processing imprint, the overlap rate of dual signal triggering at the same node is identified to establish a chemical evidence chain rating.

[0035] Specifically, the degradation timestamp deviation is obtained by comparing the offset amplitude of the directional anomaly marker with the degradation rate over time. When oils degrade at a natural rate within the predetermined circulation time of the supply chain, the offset amplitude recorded by the directional anomaly marker should be equivalent to the product of the natural degradation rate and circulation time of that type of oil. If the actual offset amplitude exceeds this expected value, it indicates that the degree of degradation at the node does not match the claimed circulation time. The degradation timestamp deviation is calculated as δ = A / (V×T)-1, where A is the relative offset amplitude of the node recorded by the directional anomaly marker (the sum of the absolute values ​​of the relative change rates of the intensities of each flip wave band, %), V is the natural degradation rate of the corresponding oil type (% / h), and T is the circulation time of the node (h). A positive δ value indicates that the actual degradation exceeds the natural expectation. When δ exceeds 0.5, the probability of the corresponding oil age being falsely reported or mixed with highly aged oils exceeds 70%. Within the valid directional anomaly segment, each node is substituted into the formula. For nodes with a degradation timestamp deviation δ exceeding 0.5, anomaly entries are generated, categorized into strong deviation (δ>1.0) and moderate deviation (0.5<δ≤1.0). Nodes with δ below 0.5 do not generate anomaly entries but retain their values ​​for dual-signal joint verification. The thermodynamic disorder marker V for directional anomalies is conservatively substituted with 80% of the natural rate of the normal pathway to prevent underestimation of V by non-standard pathway nodes, which could lead to artificially low degradation timestamp deviations. Strong deviation marker nodes correspond to the highest fraud confidence level in the chemical evidence chain rating stage. When strong deviation nodes account for more than 40% of the total number of nodes in the valid directional anomaly segment across the entire batch, a high-frequency strong deviation warning is added to the degradation timestamp deviation metadata, raising the fraud confidence threshold for the chemical evidence chain rating of the warning batch from C to B.

[0036] In some embodiments, the step of generating processing imprints by analyzing the trace pollutant combination characteristics of each node in the oil source mixing marker includes: screening the sampling node set corresponding to the mixing event in the oil source mixing marker to form an event node set; using the event node set to calculate the ratio of metal to organic pollutant concentrations at each node to obtain the pollutant element ratio; assessing the degree of matching of facility source characteristics based on the pollutant element ratio to determine the facility matching confidence level; and collecting the facility attribution judgment conclusions from the facility matching confidence level to generate processing imprints.

[0037] An event node set is formed by screening sampling nodes corresponding to mixed events based on oil source mixing markers. Oil source mixing events are typically concentrated in the supply chain's warehousing, loading, or distribution operations. The detection value of contaminants varies significantly among different types of nodes: production-end nodes are mainly characterized by equipment-specific contamination, distribution-end nodes are mainly characterized by container migration contamination, and retail-end nodes usually have diluted contaminant concentrations. The start and end node intervals of each exceeding drift range of the oil source mixing markers precisely cover the temporal coverage of these operational stages. The event node set accurately screens nodes within the drift range, excluding nodes that are in normal circulation before and after the drift, ensuring that trace contaminant detection targets the actual location of the mixing event. The oil source mixing markers progressively expand the mixing marker nodes and mark key collection points in the event node set. The sampling volume of metal contaminants at key collection nodes is increased to twice the standard amount, resulting in a corresponding improvement in the signal-to-noise ratio, making it easier to distinguish trace facility residues from background noise. Oil source mixing marker drift peak nodes typically correspond to the operational stage with the highest mixing ratio. For example, if a batch's drift peak occurs at node 5 (during the wholesale warehouse transfer operation), the trace contaminant concentration at this node is 3.2 times the average of the nodes before and after it. The event node set assigns priority to the drift peak node for sampling. When batch sampling resources are limited, sampling quotas are allocated in the order of key sampling → drift peak → ordinary acceleration zone nodes. After the event node set is generated, the sampling coverage is calculated. When the coverage is below 15%, a low coverage warning is added to the metadata. For batches with low coverage warnings, the sampling location of the event node set must be extended to two nodes before and after the drift peak for supplementary sampling. Nodes where the coverage still does not meet the standard after the extension sampling are marked with a sampling gap mark.

[0038] The pollutant element ratios were obtained by statistically analyzing the concentration ratios of metals and organic pollutants at each node using an event node set. The iron-to-copper concentration ratio is effective in identifying the material of oil storage containers. Stainless steel oil tanks have an iron / copper ratio of approximately 8 to 12, aluminum alloy containers have an iron / copper ratio below 3, and ordinary carbon steel drums have an iron / copper ratio above 15. These three material ranges do not overlap. After quantifying the five metals (iron, copper, zinc, lead, and nickel) and organic residues (methyl palmitate and siloxanes) in the samples from each node in the event node set, the multidimensional concentration ratio vectors of the pollutant element ratios were extracted for each node as a quantitative expression of the facility fingerprint. The absolute concentration of metals at key nodes in the event node set was approximately twice that of non-key nodes. The pollutant element ratios were used to construct feature vectors based on concentration ratios rather than absolute concentrations, ensuring that the fingerprint vector directions of high-concentration and low-concentration samples are consistent, eliminating the interference of sampling volume differences on facility matching. In some event node sets, due to oil emulsification leading to metal ion complexation and sedimentation, the detected concentration is lower than expected. Nodes with a total metal concentration below 30% of the batch average are marked with a low-concentration suspected marker in the pollutant element ratio. The feature vector of nodes with low-concentration suspected markers has its confidence weight reduced to 0.7 times during the facility matching stage to prevent the overall low concentration caused by emulsification from lowering the facility identification confidence. When siloxane residues in organic pollutants are used in conjunction with the metal ratio, the probability of both types of pollutants pointing to the same facility is significantly higher than the probability of either single type pointing to the same facility alone. For nodes with significantly high levels of both metals and organic pollutants, the pollutant element ratio adds a multi-pollutant consistency marker. This marker triggers a 0.05 reduction in the matching threshold during the facility matching stage to reflect the increased confidence of the combined evidence from multiple types.

[0039] The confidence level of facility matching is determined by assessing the degree of matching of facility source characteristics based on pollutant element ratios. Cosine similarity only reflects the similarity of concentration ratio structure and is not sensitive to absolute concentration levels. When used for facility fingerprint matching, it can effectively resist the overall concentration scaling caused by changes in doping ratios. The feature vectors of each node of pollutant element ratios are matched with the representative vectors of each known facility in the facility fingerprint database using cosine similarity. The highest matching score and the corresponding facility number constitute the core field of facility matching confidence. A similarity greater than 0.85 confirms strong facility attribution, between 0.65 and 0.85 confirms weak facility attribution, and below 0.65 indicates questionable attribution. The facility fingerprint database is continuously updated from historical detection records. Each type of facility is represented by the average vector of sampling over the past 12 months. When the average update cycle exceeds 6 months, a timeliness indicator is added to the representative vector of the facility. The facility matching confidence result using the representative vector with the timeliness indicator is downgraded by 0.05. The matching threshold for suspected low-concentration marker nodes of pollutant element ratios is increased from 0.65 to 0.75. Low-concentration nodes have higher uncertainty, and a loose threshold may misjudge random ratio noise as weak facility attribution. Nodes with abrupt changes in pollutant element ratios within a given range undergo independent attribution verification. If the attribution facility differs from the main facility in the same range, a facility change indicator is added to the facility matching confidence score. For example, if nodes 6 through 9 of a batch belong to a Class A oil processing facility, and node 10 changes to a Class B transport container, the change indicator suggests that node 10 experienced an independent facility contamination event and must be included separately in the chemical evidence chain rating analysis rather than being merged with previous nodes. Nodes with questionable attribution trigger amplified sampling and retesting. If the retesting still indicates questionability, a facility matching confidence score is marked as pending manual verification. Nodes pending verification are not included in the range's attribution statistics until the manual verification conclusion is added.

[0040] Processing imprints are generated by summarizing the facility attribution judgments for each node based on facility matching confidence. The facility confirmation is highest when multiple nodes within the same segment belong to the same facility and all have strong attributions. The confidence of a strong attribution for an isolated single node is approximately 40% lower than that of three or more consecutive strong attributions. The facility matching confidence judgments for each node are summarized by the segment number marked with oil source mixing. The processing imprints are organized using the segment number as the primary key, comprising three components: the facility attribution judgment for each segment, the total number of nodes within the segment, and the percentage of nodes with strong attributions. For segments with facility jump indicators, the processing imprints are marked with multi-facility source identifiers. The processing imprints also include the percentage distribution of each facility node in multi-facility source segments. When each type of facility accounts for approximately 50% of the nodes in a segment, the processing imprints include a mixed facility source identifier, indicating that the segment has undergone alternating processing by two different types of facilities. Nodes with facility matching confidence pending manual verification are categorized into the pending verification field in the processed imprint quantity. After manual verification, the corresponding nodes are added and the pending verification label is removed. When the number of pending verification nodes exceeds 30% of the total number of nodes in a segment, the processed imprint quantity is marked with a high pending verification ratio for that segment, indicating that there is significant uncertainty in the facility attribution conclusion for that segment, and it must be downgraded during the chemical evidence chain rating stage. After the processed imprint quantity is generated, the proportion of strongly attributable nodes in the entire batch to the total number of event node sets is calculated. When the proportion exceeds 50%, a high attribution certainty label is added to the metadata of the processed imprint quantity. For batches with high attribution certainty, the matching threshold for the facility fingerprint database of subsequent supplementary sampling or retesting of nodes is tightened from 0.85 to 0.88, and a stricter standard is applied to review the data to maintain high consistency of the rating input.

[0041] In some embodiments, the step of establishing a chemical evidence chain rating based on the overlap rate of dual-signal triggering at the same node by identifying the degradation time stamp deviation and the processing imprint amount includes: extracting time stamp anomaly markers and deviation amplitudes of each node from the degradation time stamp deviation to construct a time stamp anomaly set; matching the processing imprint amount anomaly markers of the corresponding nodes to the time stamp anomaly set to determine a dual-marker set at the same node; judging the joint triggering ratio of dual-signal triggering and abnormally clean nodes based on the dual-marker set at the same node; and establishing a chemical evidence chain rating by performing node fraud confidence grading based on the joint triggering ratio.

[0042] The timestamp anomaly set is constructed by extracting timestamp anomaly markers and deviation amplitudes from each node based on the degradation timestamp deviation. Nodes with a deviation amplitude exceeding 0.5 have exceeded the upper limit of the 70% confidence interval for natural degradation. Under normal operating conditions, the probability of such nodes appearing naturally is approximately 7%. If such nodes appear densely in a batch, it indicates deliberate intervention. After extracting the node numbers and δ values ​​of nodes with strong and moderate deviations in the degradation timestamp deviation, the two types of markers are distinguished by the deviation level field, which together constitute the core entries of the timestamp anomaly set. Batches with high-frequency strong deviation warnings in the degradation timestamp deviation are appended with a batch-level strong deviation warning transmission marker in the timestamp anomaly set. This marker triggers a stricter synchronization judgment threshold during the matching stage of the dual marker set of the same node, preventing the overall mean of high-deviation batches from being raised and masking the true anomalies of medium-deviation nodes. Nodes with a degradation timestamp deviation value (δ) below 0.5 are retained and incorporated into the edge reference field of the timestamp anomaly set. These nodes are distinguished from formal anomaly entries by their edge source identifier. When the processing imprint of an edge-source node indicates strong facility attribution, an edge upgrade identifier is added, elevating it to a formal entry. This upgrade mechanism can capture low-intensity adulteration cases where "degradation deviation is not significant but facility fingerprints are clear." The entries corresponding to nodes with degradation timestamp deviation duration estimation identifiers are labeled in the timestamp anomaly set with a 0.8 confidence weight to prevent the introduction of regular deviations due to shift duration estimation errors. After the timestamp anomaly set is generated, the number of nodes at each deviation level is counted. When the mean δ value of a strong deviation node exceeds 1.5, an extreme deviation warning is added to the timestamp anomaly set metadata. In the extreme deviation warning batch joint trigger ratio calculation stage, the weight α for abnormally net nodes is increased from 0.4 to 0.6, reflecting the higher strength of fraudulent evidence from net nodes under extreme deviation conditions.

[0043] For example, the step of matching the processing imprint quantity anomaly flags of corresponding nodes to the timestamp anomaly quantity set to determine the dual-mark set of the same node includes: calculating the degradation rate deviation amplitude of each abnormal node based on the timestamp anomaly quantity set to obtain the timestamp deviation amplitude; calculating the two-dimensional normalized product based on the timestamp deviation amplitude and the processing imprint quantity of the same node to obtain the imprint consistency deviation; judging the degree of two-dimensional anomaly of the timestamp and processing imprint of the same node based on the imprint consistency deviation to establish a two-dimensional triggering ratio; and using the two-dimensional triggering ratio to collect the two-dimensional synchronous triggering node set to determine the dual-mark set of the same node.

[0044] The timestamp deviation magnitude is obtained by calculating the degradation rate deviation of each anomalous node based on the timestamp anomaly set. δ quantifies the relative degree of deviation but is linked to the circulation duration T. Directly comparing δ across nodes can introduce confusion due to different T values ​​for each node. For example, with δ=0.6, the absolute deviation magnitude of a node with T=2 days is only 1 / 5 of that of a node with T=10 days, indicating a significant difference in the strength of fraud evidence. The timestamp deviation magnitude is converted into an absolute deviation magnitude by multiplying the δ values ​​of each node in the timestamp anomaly set by V×T. The absolute deviation magnitude is directly comparable between nodes with different circulation durations. The absolute deviation magnitude of edge source identifier nodes in the timestamp anomaly set is labeled with a 0.6 confidence weight in the timestamp deviation magnitude and stored separately from the formal anomalous nodes to prevent edge nodes from diluting the magnitude statistics of formal anomalous nodes. The absolute deviation of the duration estimation marker nodes in the time stamp anomaly concentration is subject to error due to the uncertainty of T. The time stamp deviation magnitude adds an uncertainty range estimate (calculated based on the average duration standard deviation) to the duration estimation marker nodes. Nodes with an uncertainty range exceeding 50% of the absolute deviation magnitude are marked with a high uncertainty marker. The confidence weight of high uncertainty nodes is further reduced to 0.6 times during the imprint consistency deviation calculation stage. After the time stamp deviation magnitude is generated, quantile statistics are performed. Nodes exceeding the 75th percentile are marked with a high magnitude marker. High magnitude markers trigger an increase in the verification priority of processed imprints for the same node during the imprint consistency deviation calculation stage. The strongest time stamp evidence node receives the most sufficient two-dimensional verification resources, preventing resource sharing from leading to key evidence nodes and marginal nodes receiving the same verification intensity.

[0045] The imprint consistency deviation is calculated by multiplying the timestamp deviation amplitude and the processing imprint quantity at the same node using a two-dimensional normalized product. Band intensity difference and cosine similarity belong to different dimensions. Directly comparing the one with the larger dimension will dominate the joint judgment and mask the abnormal signal of the other. The timestamp deviation amplitude and the processing imprint quantity must each be normalized to the 0-1 range using the maximum value within the batch as the denominator. After normalization, the two-dimensional values ​​are comparable for nodes in the same batch. The imprint consistency deviation quantifies the degree of consistency deviation between the two types of signals at the same node using the product of the two-dimensional normalized values. A high product of the two-dimensional normalized values ​​effectively widens the gap compared to a high value in only one dimension. When both dimensions are 0.8, the product is 0.64; when only one dimension is 0.9 and the other is 0.2, the product is only 0.18. This product structure naturally suppresses the influence of single-signal noise on the joint judgment. The verification priority of processing imprint quantity for nodes with high timestamp deviation amplitude has been increased. After verification, if the normalized value of the processing imprint quantity for the same node also exceeds 0.6, the imprint consistency deviation will be marked with a double-high value. Nodes marked with double-high values ​​constitute the most certain counting source for subsequent two-dimensional trigger ratios. The normalized deviation of nodes with high uncertainty in timestamp deviation amplitude is included in the calculation of the imprint consistency deviation product with a weight of 0.6. If the product after conversion still exceeds 0.36, it is classified as a two-dimensional significant node, ensuring that nodes with high uncertainty are not completely excluded when the evidence is significant. Nodes with the imprint consistency deviation product of all nodes exceeding 0.36 (the product threshold when both two-dimensional normalizations reach 0.6) are marked with a two-dimensional significant label. The ratio of the number of two-dimensional significant nodes to the total number of event nodes in the batch constitutes the basis for calculating the two-dimensional trigger ratio.

[0046] A two-dimensional triggering ratio was established to assess the degree of two-dimensional anomaly between the timestamp and processing imprint at the same node based on the imprint consistency deviation. The product of the normalized deviations of the two dimensions is significantly greater when both dimensions are high than when only one dimension is high. When both dimensions are 0.6, the product is 0.36, while when one dimension is 0.9 and the other is 0.1, the product is only 0.09. An imprint consistency deviation exceeding 0.36 is used as the threshold for significant two-dimensional synchronization, enabling the two-dimensional triggering ratio to distinguish between genuine two-signal synchronous triggering and occasional deviations dominated by single-signal noise. The two-dimensional triggering ratio η is obtained by dividing the number of nodes with an imprint consistency deviation product exceeding 0.36 by the total number of nodes in the event node set. When η exceeds 0.4, it indicates that more than 40% of the nodes in the event node set exhibit two-dimensional synchronization anomalies. This proportion occurs with a probability of approximately 3% in normal batches. An η exceeding 0.4 indicates that the batch exhibits a statistically significant two-dimensional chemical anomaly pattern. Nodes with high values ​​for both imprint consistency deviation are directly included in the number of nodes exceeding the limit. Nodes with non-high values ​​but a product between 0.36 and 0.50 are included with a weight of 0.8, and nodes with a product exceeding 0.50 are included with a weight of 1.2. After weighting, the transition of the two-dimensional trigger ratio between different levels of evidence strength is smoother, and the contribution of strong evidence nodes is appropriately amplified. Conflicting nodes with opposite normalization directions (one dimension is too high and the other is too low, resulting in a product below 0.09) are excluded from the total number of event nodes. When the proportion of contradictory nodes exceeds 15%, a dense contradictory signal identifier is added to the metadata of the two-dimensional trigger ratio. The two-dimensional significance threshold for batches with dense contradictory signals is increased from 0.36 to 0.45 to reduce misjudgments. For example, if the confidence of the processing imprint facility matching is generally low due to the mixture of oil types in a certain batch, the increased threshold can effectively filter out occasional artificially high product caused by facility fingerprint failure.

[0047] The dual-trigger ratio is used to aggregate the set of dual-trigger nodes that trigger synchronously in the dual-dimensional environment, thus determining the dual-label set of the same node. When η exceeds 0.4, all nodes whose product has passed the 0.36 threshold are included in the dual-label set of the same node. When η is between 0.2 and 0.4, only the top 30% of nodes with the highest product are included in the dual-label set of the same node. This ensures that low-η batches do not lose valid fraud clues due to sparse evidence leading to an empty dual-label set of the same node. The top 30% strategy retains the set of nodes with the most sufficient dual-dimensional evidence in the current batch, even if their absolute evidentiary strength is limited. The threshold for batches with dense contradictory signals in the dual-trigger ratio has been increased to 0.45, and the number of nodes included in the dual-label set of the same node has been reduced accordingly. The metadata records the actual threshold used, which can be used to trace the strictness of the dual-label set selection during the joint trigger ratio calculation stage. When η is below 0.2, the dual-label set of the same node only contains the strongest evidence nodes with a product exceeding 0.50. The metadata of the dual-label set of the same node is marked with low-η sparse evidence identifiers. The batch of sparse evidence identifiers is rated as B at the highest level in the chemical evidence chain rating. For example, if a batch has only 2 nodes with a product exceeding 0.50, these 2 nodes have clear evidence, but they account for a very small proportion of the total nodes. After the dual-label set of the same node is marked with sparse evidence identifiers, the rating upper limit is limited to B to prevent fraud by inferring the entire batch as A based on a very small number of nodes. After the dual-label set of the same node is determined, the number and proportion of the three types of sources are counted: nodes with double high value identifiers, weighted inclusion nodes, and edge upgrade identifier nodes. The proportions of the three types of sources and the η value are jointly included in the metadata of the dual-label set of the same node. The joint trigger ratio calculation and the traceable review of the chemical evidence chain rating can both obtain a complete evidence tracing record from the metadata.

[0048] The joint triggering ratio is determined based on the dual-marker set of the same node and the joint triggering ratio of dual-signal triggering and abnormally clean nodes. Abnormally clean nodes refer to nodes where all surrounding nodes are high-risk but the current node has not triggered either of the dual signals. Their abnormal absence may be due to targeted cleaning by the adulterer or dilution with a large amount of normal oil to cover up evidence. The dual-marker set of the same node must simultaneously identify such nodes and include them in the calculation of the joint triggering ratio R, calculated as follows: R=(M+α×C) / N, where M is the number of nodes in the batch that are simultaneously triggered by dual signals, C is the number of abnormally clean nodes (two nodes before and after which are high-risk but the current node has not triggered either of the dual signals), α is the weight of the clean nodes (baseline value 0.4, adjusted to 0.6 for extreme deviation warning batches), and N is the total number of nodes participating in the joint verification. When R exceeds 0.5, it is judged as a high-confidence fraud batch. The identification of abnormally clean nodes relies on the spatial distribution of high-risk nodes in the same node's dual-label set. Isolated, excessively clean nodes within densely populated high-risk node sections often indicate targeted cleanup operations by the adulterer. For example, in a certain batch, nodes 7 to 14 are all high-risk nodes triggered by dual signals, while node 10 does not trigger either signal. Node 10 is thus identified as an abnormally clean node and included in the C-side with a weighted average of α=0.4. The identification window for abnormally clean nodes in batches with high coverage exceeding 0.4 in the same node's dual-label set is expanded from two nodes before and after to three nodes on each side to prevent missed identification due to narrow buffers on both sides of the cleanup node when fraudulent links are dense. For the R formula, nodes with a weighted average of 0.36 to 0.50 in the same node's dual-label set product are assigned to the M-side with a weight of 0.7 (independent in purpose from the 0.8 weight used in the same interval in the η calculation). Their contribution is lower than that of full-strength nodes but is still included in the synchronous triggering side rather than the excessively clean side, reflecting the true state of these nodes where there is evidence but doubt.

[0049] A chemical evidence chain rating system is established based on the joint trigger ratio to classify the confidence level of fraud at each node. A rating of A indicates sufficient fraud evidence (R≥0.5), B indicates suspected fraud (0.25≤R<0.5), and C indicates insufficient evidence (R<0.25). The three-tier boundary is determined by the overlap of the R distribution between historical fraudulent batches and normal batches in the same supply chain. The misclassification rate for each tier is controlled to be below 5%. An R of 0.5 means that one in two participating verification nodes exhibits synchronous dual-signal triggering or anomaly over-triggered phenomena. This density occurs approximately 2% in normal batches, resulting in a high statistical significance for a Grade A rating. An R between 0.25 and 0.5 corresponds to one in four nodes, indicating a strong tendency towards fraud. When the proportion of C in the joint trigger ratio calculation exceeds M, it indicates that the evidence mainly comes from anomalous absence rather than synchronous triggering. The chemical evidence chain rating system marks such batches with an evidence structure bias indicator, suggesting that the fraudster may have strong trace manipulation capabilities. If an evidence structure bias exists in a Grade A rating, a manual review suggestion is added during the traceability chain recording stage. When the α factor in the joint trigger ratio is increased from 0.4 to 0.6 due to extreme deviation warning, the chemical evidence chain rating will include an additional parameter adjustment flag for the batch rating conclusion. This prevents rating jumps caused by parameter adjustments from misleading cross-batch longitudinal comparisons. A-level batches in the chemical evidence chain rating trigger cross-batch tracking during the traceability block recording stage; B-level batches are marked as pending observation; if the next batch rating remains B or higher, it will be upgraded to A-level. For C-level batches, if any abnormal signal item, such as degradation timestamp deviation or processing mark amount, exceeds 10%, a single-signal suspected warning will be added to prevent C-level from masking genuine fraud clues.

[0050] Step S104: Based on the chemical evidence chain rating, the fingerprint offset direction quantity is combined with the preceding hash encoding to obtain the forward differential hash packet. The forward differential hash packet is used to verify the consistency voting of the chemical conclusions of multiple parties to form a chemical consensus record. The consistency of the drift dynamics of the verification node is determined from the chemical consensus record to determine the traceability chain block record.

[0051] In some embodiments, the step of collecting the fingerprint offset direction quantity based on the chemical evidence chain rating to obtain a forward differential hash packet by jointly encoding the preceding hash value includes: determining the current node symbol sequence in the fingerprint offset direction quantity based on the chemical evidence chain rating to generate a node symbol sequence; using the node symbol sequence and the preceding node hash value as the input to a hash function to form a joint hash input; collecting the fingerprint offset direction quantity weighted by the evidence level based on the joint hash input to establish a node chain hash value; and encapsulating compliance credentials and a geographic timestamp according to the node chain hash value to obtain a forward differential hash packet.

[0052] Based on the chemical evidence chain rating, the node symbol sequence is generated by determining the current node symbol sequence in the fingerprint offset direction quantity. A-level nodes in the chemical evidence chain rating have sufficient fraud evidence; all four band deviation directions in the corresponding fingerprint offset direction quantity are statistically significant. The node symbol sequence for A-level nodes extracts the complete four-bit symbol sequence as hash input, and any modification to a single band direction can be captured by hash verification. B-level nodes in the chemical evidence chain rating are triggered by only one type of signal. The reliability of band symbols in the fingerprint offset direction quantity that are not jointly confirmed by two signals is lower than that of A-level nodes. The node symbol sequence for B-level nodes replaces the non-double-signal confirmed band bits with neutral "0" symbols, preserving the node's structural position in the hash chain while avoiding the introduction of potential hash sensitivity due to unreliable direction symbols. C-level nodes in the chemical evidence chain rating are not triggered by either signal, and the node deviation direction corresponding to the fingerprint offset direction quantity lacks evidentiary confidence. The node symbol sequence for C-level nodes uses a four-bit sequence of all "0"s to maintain chain continuity but does not contribute effective chemical information. In the chemical evidence chain rating, the symbol sequence of the node corresponding to the contradictory node is replaced by a special contradictory identifier "X" to replace the corresponding band position. The contradictory identifier has a unique hash feature in the joint hash input stage, which makes the contradictory node clearly distinguishable from the normal deviation node in the hash space. For example, if the spectral deviation direction of a node indicates adulteration while the facility fingerprint shows that it has not been processed, the two types of signals contradict each other. The corresponding band "X" makes the hash value of the node fall into the exclusive hash range of the contradictory node. During the multi-party voting stage, each institution can identify the source of the contradiction based on this.

[0053] The joint hash input is constructed by concatenating the node symbol sequence with the hash value of the preceding node as the input to the hash function. This concatenation of the preceding node's hash value and the current node's symbol sequence makes the hash output of each node dependent on the historical state of the entire forward chain. Any tampering with the preceding node propagates downstream, causing a break in the hash chain. The joint hash input achieves full-chain tamper-proof protection through this chain-like dependency. The current node's four-bit symbol sequence is converted into a byte string using ASCII encoding and directly concatenated with the preceding node's hash value (256-bit SHA-256 output) to form the original byte sequence of the joint hash input. The concatenation order is fixed as "previous hash value || current symbol sequence," and this order specification must be explicitly recorded in the joint hash input metadata to ensure the reproducibility of hash calculation results from different participating institutions. Before concatenation, the joint hash input of the node containing the conflicting identifier "X" in the node symbol sequence undergoes masking processing. The conflicting bits are replaced with a fixed feature string, making the distribution characteristics of conflicting nodes in the hash space distinctly different from normal and abnormal nodes, facilitating the identification of the conflict source during the hash verification stage. The first node of a supply chain batch has no preceding node hash value. The combined hash input uses the hash value of the batch's unique identifier as the preceding node hash value for the first node. The batch's unique identifier is generated by combining the production license number, batch date, and geographical code of origin, ensuring that the first node hash value of the combined hash input for different batches is not repeated. When concatenating the combined hash input of nodes with timestamp anomaly sets in the node symbol sequence, a timestamp anomaly flag byte is appended. This timestamp anomaly flag byte is parsed as an entry to be verified during the hash verification stage, which does not block the chain verification but triggers the manual timestamp verification process.

[0054] A chained hash value for nodes is established based on the weighted fingerprint offset direction of the joint hash input. Evidence level weighting is achieved by adjusting the repetition factor of each band symbol bit in the hash calculation. For chemical evidence chain rating, the key band symbol of a Level A node is repeated 3 times in the hash input, Level B nodes are repeated 2 times, and Level C nodes are not repeated. This repetition factor makes the chained hash value for nodes much more sensitive to data changes in high-confidence fraud nodes than in low-confidence nodes, thus increasing the detection cost of tampering with high-confidence nodes. The repetition factor is appended to the end of the joint hash input byte string in the order of each band symbol bit, ensuring that the concatenation order is consistent with the original band order of the node symbol sequence. This ensures that different detection agencies obtain identical input sequences when performing weighted concatenation on the same node. The joint hash input is calculated using the SHA-256 hash function to obtain a 256-bit chained hash value for nodes, with a SHA-256 collision probability of approximately 2. -128This means that the probability of the same hash value corresponding to different input data is negligible, thus ensuring the ability of the node chain hash value to distinguish the chemical characteristics of different supply chain nodes. The timestamp anomaly flag byte in the joint hash input participates in the hash calculation but does not affect the chain verification logic of the node chain hash value. The timestamp anomaly flag is only recorded as an additional attribute in the hash value metadata and does not interfere with the inheritance relationship of normal nodes. Batches with dense fingerprint offset direction warnings are synchronously recorded in the node chain hash value metadata, enabling each institution to grasp the overall directional anomaly density of fingerprint offset directions when receiving node chain hash values ​​during the multi-party voting stage. This helps each institution to implement a more rigorous review process for high-risk batches before voting.

[0055] A forward differential hash packet is obtained by encapsulating compliance credentials and geographic timestamps using the node chain hash value. The compliance credential contains three pieces of identity information: production license number, testing institution certification number, and batch certification code. These three pieces of information are independently issued by the market supervision department, the testing institution certification management agency, and the production enterprise's batch ledger, respectively, and are not overlapping. If any one of them is tampered with, it can be independently verified through the public archives of the corresponding issuing entity. The joint encapsulation of the node chain hash value and the compliance credential gives the forward differential hash packet traceability of institutional affiliation. Any node on the hash chain can be traced back to the specific testing institution and place of origin information, providing a basis for verifying the source of the conclusions of each institution during the multi-party voting stage. The geographic timestamp is generated jointly by the GPS coordinates of the data collection device and the time synchronization of the national time server. The GPS coordinate accuracy must be no less than 10 meters, and the time server synchronization error must not exceed 60 seconds. The 10-meter accuracy corresponds to the geographic positioning requirements at the supply chain node level (spatial resolution capability for warehouses / vehicles / stores). The 60-second synchronization error accounts for less than 4% of the shortest node circulation time (approximately 30 minutes). These two accuracies together ensure the validity of the geographic timestamp as an additional attribute for node identity. The forward differential hash packet marks any node with a geographic timestamp downgraded flag if any one of the accuracies fails to meet the standards. Nodes with downgraded flags participate in consensus confirmation during the multi-party voting stage but are not used as arbitration benchmark nodes. After the node chain hash value collision probability verification is passed, the 256-bit hash value, compliance certificate, and geographic timestamp are encapsulated into a forward differential hash packet byte string with a fixed format. The byte string format includes a version number field, which is located in the first 2 bytes of the byte string. The high byte indicates the major version, and the low byte indicates the minor version. When the version number changes, the old version of the forward differential hash packet must be backward compatible to ensure that historical traceability records can still be read after the protocol upgrade.

[0056] A chemical consensus record is formed by verifying the consistency of chemical conclusions from multiple parties through forward differential hash packets. The chemical conclusions of a single testing institution may be affected by equipment deviations, human error, or sample contamination. Judgments of adulteration based solely on data from one party lack sufficient credibility in legal proceedings. The forward differential hash packet requires consistency comparison after at least three independent testing institutions submit their respective chemical conclusions. Consensus confirmation is triggered when the conclusions of at least two-thirds of the institutions match, thus forming the chemical consensus record. Each independent testing institution submits its chemical conclusion using its calculated node hash value. The forward differential hash packet performs a consistency check on the difference between the submitted hash value and the hash value recorded on the chain. Differences within a tolerance threshold (Hamming distance in the hash space not exceeding 4 bits) are considered a match. Submissions exceeding the tolerance are marked as abnormal submissions in the chemical consensus record, and the institution submitting the abnormal submission must resubmit or explain the source of the difference within 48 hours. When the consensus voting ratio of all nodes in the forward differential hash packet is less than 2 / 3, the node is determined to be a consensus divergence node. A consensus divergence node triggers the intervention of a fourth-party arbitration body. After submitting its conclusion, the arbitration body re-determines the issue based on majority rule. The arbitration conclusion is marked in the chemical consensus record with an arbitration source identifier, distinguishing it from direct consensus nodes. For example, in a batch, if two of the three testing institutions have matching hash values ​​and one exceeds the tolerance, this node passes the consensus determination with a 2 / 3 voting ratio and is included in the formal entry of the chemical consensus record. The reason for the difference in the third institution must be traced and recorded for auditing purposes. The chemical consensus record organizes the consensus conclusion of each node, the number of participating institutions, and the voting ratio using the node number as the primary key. When the consensus nodes in the same batch account for more than 80%, the chemical consensus record metadata is appended with a high consensus coverage identifier. Batches with high consensus coverage have a higher credibility weight for fraud conclusions during the traceability chain block recording stage.

[0057] The traceability chain record is determined by verifying the drift kinetic consistency of nodes in the chemical consensus record. As oils flow along the supply chain, the increase in the oxidized sterol ratio at each node should follow physical conservation laws. Oxidation products accumulated at upstream nodes cannot disappear without cause at downstream nodes. The drift amplitude and direction of the consensus conclusions at each node in the chemical consensus record must satisfy this monotonicity constraint in time. A node sequence that violates this constraint indicates data forgery or the existence of physically impossible operations in the supply chain record. The traceability chain record uses drift kinetic consistency verification as the final entry threshold. After arranging the consensus drift amplitudes of each node in the chemical consensus record in time sequence, a significant reversal in the drift amplitude between adjacent nodes (the absolute value difference exceeds three times the average of the entire batch) is identified as a kinetically inconsistent node. Kinetically inconsistent nodes must be marked with an abnormal kinetic identifier before the traceability chain record is generated. The chemical consensus record entry corresponding to the node with the abnormal kinetic identifier is transferred for manual review, and the review conclusion affects the final generation of the traceability chain record. When the entire batch of chemical consensus records passes the drift dynamics consistency check and has no pending anomaly dynamics indicators, all consensus nodes in the current batch are packaged into a traceability chain block record according to blockchain standards. The hash values ​​of each node within the block are organized in a Merkle tree structure, with the root hash value serving as the unique identifier for the entire traceability chain block record. Batches with high consensus coverage of chemical consensus records have higher priority for being included in the traceability chain block record packaging process than batches with low consensus coverage. Batches with low consensus coverage must complete all arbitration decisions before entering the traceability chain block record packaging process.

[0058] Step S105: Based on the traceability chain block record, identify the topological association density between abnormal nodes to obtain a fraud cluster set. Extract the offset type and organizational forgery degree from the fraud cluster set to obtain an anomaly type report. Integrate the anomaly type report into the chemical evidence chain rating to output a chemical traceability report.

[0059] In some embodiments, obtaining a fraud cluster set based on the topological association density between abnormal nodes identified by the traceability chain block records includes: determining a set of abnormal nodes in the chain block based on nodes whose chemical evidence chain rating is below a qualified threshold identified by the traceability chain block records; constructing an abnormal node association graph for the abnormal node set based on the supply chain flow adjacency relationship and isolated off-chain nodes; performing time-series fraud behavior evolution analysis on the abnormal node association graph to establish fraud evolution feature quantities; and collecting fraud types and evolution stages from the fraud evolution feature quantities to form a fraud cluster set.

[0060] Based on the traceability chain records, nodes with chemical evidence chain ratings below the acceptable threshold are identified to determine the chain's abnormal node set. The chemical evidence chain rating is defined as follows: A-level nodes have a joint triggering ratio R ≥ 0.5, and B-level nodes have a ratio of 0.25 ≤ R < 0.5. Using R = 0.25 as the lower bound of the acceptable threshold ensures that all nodes below the B-level threshold are included in the chain's abnormal node set. The R value of each node in the traceability chain record serves as the direct basis for determining the acceptable threshold; nodes below 0.25 are identified as unacceptable nodes and included in the chain's abnormal node set. When a Class A node in the traceability chain enters the abnormal node set, its corresponding mean delta value and mean confidence level of the processing imprint facility are simultaneously recorded. These two additional values ​​are used as additional attributes of the node during the abnormal node association graph construction stage for subsequent traceability. Class B nodes are merged into the abnormal node set with the fraud confidence level field and R value. Class B nodes with R values ​​between 0.40 and 0.50 are additionally marked with a critical B- label. The weight of the associated edge of critical B- nodes is increased to 0.8 during the abnormal node association graph construction stage (higher than the weight of 0.7 for regular Class B nodes). Due to reliability issues with the source of contradiction in the traceability chain record, the R value of contradictory nodes is not directly determined using the 0.25 threshold. Contradictory nodes are archived separately in the abnormal node set with a contradiction source label, retaining their topological coordinates but not participating in the reliable node set calculation for the association graph edge weights. After the abnormal node set of the chain block is generated, the proportion of unqualified nodes in the entire batch is calculated. When the proportion exceeds 30%, a high anomaly density warning is added to the metadata of the abnormal node set of the chain block. During the construction of the abnormal node association graph, the upper limit of the buffer for indirectly associated normal nodes in the high anomaly density warning batch is relaxed from 2 to 3. The traceability chain block records chemical evidence chain rating B and nodes with single signal suspected warning are archived in the chain block abnormal node set with single signal source identifier. The weight of the associated edge of the single signal source identifier node is reduced to weight 0.6 during the construction of the abnormal node association graph, forming a gradient layer with the weight range of 0.7 to 1.0 of the dual signal confirmed nodes.

[0061] An anomalous node association graph is constructed for the chain block's anomalous node set based on the supply chain's adjacency relationships and isolated off-chain nodes. After dilution by 3 to 4 circulation nodes, the concentration of trace oil fingerprints decays to the detection limit. When two anomalous nodes are separated by more than two normal nodes, their trace commonalities are below the reliable association inference threshold. Therefore, the upper limit for the normal node buffer for indirect path associations between nodes in the chain block's anomalous node set is set to 2. Two anomalous nodes exceeding this limit are not associated with each other in the anomalous node association graph. Isolated off-chain nodes in the chain block's anomalous node set that have no direct or indirect connection with any other member are included in the vertex set as isolated nodes in the anomalous node association graph, but no associated edges are established. Isolated off-chain nodes may correspond to the source or end of a fraud chain and are treated as traceable targets separately during the fraud evolution analysis phase, without participating in cluster cohesion density calculations. The indirect association buffer for high anomaly density warning batches in the chain block anomaly node set has been relaxed to 3 normal nodes. The number of indirect association edges corresponding to the batch in the anomaly node association graph has increased accordingly, with an edge density approximately 20% to 35% higher than that of the regular batch. The metadata of the anomaly node association graph records the actual buffer parameters used to distinguish the graph structure characteristics of the two batches. Each association edge in the anomaly node association graph is assigned an edge weight, which is the product of the fraud confidence level of the two end nodes (Level A end 1.0, critical B- end 0.8, regular B- end 0.7, single signal end 0.6, contradiction source end 0.3) and then multiplied by the path coefficient (direct connection 1.0, 1 indirect node 0.8, 2 indirect nodes 0.6) to obtain the final edge weight. Association edges of nodes with high evidence level receive higher transit weights in the graph structure. The anomaly node association graph is organized using an adjacency list data structure. The vertex field records the node number, fraud confidence level, and topological coordinates, while the edge field records the end node numbers, path coefficients, and final edge weights. During the isolated off-chain determination stage, contradictory nodes are not used as reference nodes to prevent normal nodes around contradictory nodes from being misjudged as isolated off-chains.

[0062] A fraud evolution feature quantity is established by performing time-series fraud behavior evolution analysis on the abnormal node association graph. The slope of the first-order linear regression of the δ value of each member node in the connected subgraph of the abnormal node association graph along the time direction is a direct indicator for determining the evolution stage. A slope exceeding +0.05 / node is identified as the expansion stage, below -0.05 / node as the decline stage, and an absolute value not exceeding 0.05 as the plateau stage. The fraud evolution feature quantity classifies each cluster into the corresponding evolution stage by extracting the time-series slope of δ value in each connected subgraph. When the slope of the δ value sequence in each connected subgraph of the abnormal node association graph is first positive and then negative (with obvious inflection points within the same subgraph), it is identified as a complete link across stages. The fraud evolution feature quantity labels the complete link clusters with complete link identifiers. The complete link identifier indicates that the cluster covers the entire process of fraud behavior from the introduction stage to the decline stage, and is the cluster type with the most certain location of the fraud start and end point. The evolution stage of isolated, disconnected nodes in the abnormal node association graph cannot be determined by the temporal slope. Fraud evolution characteristics are inferred jointly by the δ value of isolated nodes and the matching confidence of processing imprint facilities: a δ value of 0.5 to 0.8 with the facility's home node located upstream in the supply chain indicates the introduction stage; a δ value exceeding 1.0 with the facility's home node located midstream in the supply chain indicates the expansion stage. This inference is distinguished from the deterministic evolution markers of connected subgraphs by the inferred evolution marker in the fraud evolution characteristics. The fraud evolution characteristics are weighted using chemical evidence chain rating weights (Level A 1.0, Level B 0.7) to calculate the average δ value for each evolution stage. When the weighted average δ value in the expansion stage exceeds twice that of the regression stage, an expansion-dominant marker is added to the fraud evolution characteristics metadata, indicating that the fraudulent activity is still in an active expansion stage and the circulation and penetration of adulterated oils have not yet contracted.

[0063] Fraud types and evolutionary stages are aggregated from fraud evolution features to form a fraud cluster set. Fraud types are determined by a combination of offset type (based on the ratio of spectral-chemical deviation nodes within the cluster, see below) and evolutionary stage. Spectral-dominant type plus an extension period corresponds to continuous infiltration of frying oil; chemical-dominant type plus a decline period corresponds to a dispersed type where highly oxidized oil sources are diluted by fresh oil; dual-signal synergistic type plus a complete link identifier corresponds to an organized fraud across the entire supply chain from source to end. The offset type and evolutionary stage combination of each connected subgraph of the fraud evolution features map to the corresponding fraud type identifier. The fraud cluster set aggregates cluster member nodes based on the fraud type identifier. The complete link identifier cluster of the fraud evolution features is categorized into the organized fraud category within the fraud cluster set; the extended dominant identifier cluster of the fraud evolution features is categorized into the active extended fraud category; and the plateau period cluster without a dominant identifier is categorized into the steady-state continuous fraud category. These three category fields are recorded as fraud category fields in the fraud cluster set, allowing for rapid location of the target cluster type during the anomaly type reporting stage. Isolated nodes identified by the inferred evolutionary features of fraud are maintained separately in the fraud cluster set as pending classification. The fraud type is temporarily marked "Source to be confirmed." After the chemical traceability report is generated, the fraud type conclusion is manually verified and completed. Once the verification conclusion is added, the pending classification status is canceled. Each cluster in the fraud cluster set directly inherits the evolutionary attribute summary from the fraud evolutionary features, including the evolutionary stage type, the number of nodes in each stage, and the weighted average δ value of each stage. When extracting the evolutionary attribute summary in the anomaly type reporting stage, it is directly read from the fraud cluster set without needing to backtrack to the original connected subgraph data of the fraud evolutionary features.

[0064] Anomaly reports are generated by extracting offset types and organizational forgery levels from fraud cluster sets. The offset type is determined by the ratio of spectral bias trigger nodes to chemical bias trigger nodes within each cluster: a ratio greater than 2:1 indicates a spectral-dominant type, less than 1:2 indicates a chemical-dominant type, and a ratio of roughly equal numbers of nodes from both types indicates a dual-signal co-occurrence type. The offset type of each cluster in the fraud cluster set is directly determined by these three mapping relationships. The comprehensive score for organizational forgery level is a weighted sum of cohesion density (40 points maximum), fraud duration (30 points maximum, exceeding 10 nodes for full marks), and geographical span (30 points maximum, exceeding 500 kilometers for full marks). Anomaly reports are assessed as highly organized when the total score of all clusters in the fraud cluster set reaches 80 points, moderately organized when between 50 and 80 points, and sporadic when below 50 points. Fraud clusters, including dual-signal collaborative and highly organized clusters, are appended to the anomaly report with a full-link collaborative fraud warning. These full-link collaborative fraud warning clusters are output at the highest risk level during the chemical traceability report stage, triggering cross-batch tracking. The anomaly report organizes each cluster by its cluster number with two core fields: offset type and degree of forgery organization. For spectrally dominant clusters, a list of dominant anomalous bands is appended; for chemically dominant clusters, the mean oxidosterol ratio of each member node and the node number of the drift peak are appended. Offset type mapping boundary nodes (where the ratio of spectral to chemical trigger nodes is exactly 2:1 or 1:2) are assigned to the type with the larger absolute number of nodes between the two signal types. When the absolute number of nodes is equal, they are classified as dual-signal collaborative clusters. This avoids cross-cluster jitter near the ratio boundary affecting the stability of the anomaly report due to small perturbations.

[0065] The anomaly type report integrates chemical evidence chain ratings to output a chemical traceability report. Cluster-level fraud pattern conclusions and node-level joint triggering ratios (R values) represent evidence of different granularities. The chemical traceability report presents this information in a three-tiered evidence hierarchy: the strongest layer (Level A nodes), the second strongest layer (Level B nodes), and the auxiliary layer (Level C), aligning the information of both granularities. The anomaly type report aligns the cluster offset type field with the chemical evidence chain ratings of the same cluster member nodes node-by-node in the chemical traceability report. For spectrally dominant clusters, the report adds the mean Z-score of the dominant band; for chemically dominant clusters, it adds the mean oxidized sterol ratio and the mean deviation multiple from the baseline; for dual-signal synergistic clusters, both types of summaries are listed simultaneously. Highly organized clusters in the anomaly type report trigger fraudulent logistics path reconstruction in the chemical traceability report. Using the matching confidence level of processing imprint facilities for each member node as a clue, identifiable facility origin nodes are connected to form the fraudulent logistics path. The starting and ending nodes of the path are marked as key monitoring nodes in the chemical traceability report. The chemical traceability report concludes with a summary of three indicators: the distribution of the entire batch's chemical evidence chain rating (the proportion of A / B / C level nodes), the total number of clusters for full-chain collaborative fraud warnings in anomaly type reports, and the total score for the highest degree of organizational forgery. These indicators are weighted and summed using the formula W = 40 × P_A + 30 × min(K / 3,1) + 30 × (S_max / 100) to form the batch risk index W. Here, P_A represents the proportion of A-level nodes, K represents the total number of clusters for full-chain collaborative fraud warnings, and S_max represents the highest total score for organizational forgery in each cluster. The three indicators are capped at 40, 30, and 30 points respectively, corresponding to the batch rating, the breadth of collaborative fraud, and the depth of the highest level of organizational forgery. When the batch risk index exceeds 80, a high-risk batch warning is added to the cover of the chemical traceability report, triggering a priority verification procedure by the competent authority. The chemical traceability report outputs both a machine-readable JSON version and a human-readable PDF version. The JSON version retains the R-value and δ-value for independent calculation by regulatory agencies, while the PDF version visualizes the geographical distribution of each cluster and the fraudulent logistics path for manual verification.

[0066] To implement the oil traceability method based on the chemical composition evidence chain corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2 A structural block diagram of a petroleum traceability system 200 based on a chemical composition evidence chain provided in this application embodiment is shown. For ease of explanation, only the parts relevant to this embodiment are shown. The petroleum traceability system 200 based on a chemical composition evidence chain provided in this application embodiment includes: Data acquisition module 201 is used to acquire optical fingerprint data and chemical detection data, collect the offset direction of each node from the optical fingerprint data to generate fingerprint offset direction amount, and determine the oxidized sterol ratio amount based on the sterol oxidation rate change of each node collected from the chemical detection data. Anomaly detection module 202 is used to identify the direction reversal between nodes based on the fingerprint offset direction amount to form a direction anomaly mark, and to establish an oil source mixing mark by using the oxidized sterol ratio amount to identify the multi-node progressive drift feature of the ratio. The evidence fusion module 203 is used to compare the offset amplitude of the directional anomaly marker with the degradation rate over time to obtain the degradation timestamp deviation, analyze the trace pollutant combination characteristics of each node for the oil source mixed marker to generate the processing imprint, and identify the double signal trigger overlap rate of the same node based on the degradation timestamp deviation and the processing imprint to establish a chemical evidence chain rating. The hash evidence storage module 204 is used to collect the fingerprint offset direction based on the chemical evidence chain rating and obtain a forward differential hash packet by combining the forward differential hash packet with the forward differential hash packet to verify the consensus voting of multiple chemical conclusions to form a chemical consensus record, and to determine the traceability chain block record from the chemical consensus record to verify the drift dynamic consistency of the node. The report output module 205 is used to obtain a fraud cluster set based on the topological association density between abnormal nodes identified by the traceability chain block record, extract the offset type and organizational forgery degree of the fraud cluster set to obtain an anomaly type report, and integrate the anomaly type report with the chemical evidence chain rating to output a chemical traceability report.

[0067] The aforementioned oil traceability system 200 based on a chemical composition evidence chain can implement the oil traceability method based on a chemical composition evidence chain described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application's embodiments can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0068] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. A method for tracing the origin of oils and fats based on a chain of evidence of chemical composition, characterized in that, include: Collect optical fingerprint data and chemical detection data, collect the offset direction of each node from the optical fingerprint data to generate the fingerprint offset direction, and determine the oxidized sterol ratio based on the change of sterol oxidation rate of each node collected from the chemical detection data. Based on the fingerprint offset direction, the direction reversal between nodes is identified to form a direction anomaly mark, and the oxidized sterol ratio is used to identify the multi-node progressive drift feature of the ratio to establish an oil source mixing mark. The degradation timestamp deviation is obtained by comparing the offset amplitude of the directional anomaly marker with the degradation rate over time. The processing imprint is generated by analyzing the combination characteristics of trace pollutants at each node for the oil source mixed marker. Based on the degradation timestamp deviation and the processing imprint, the chemical evidence chain rating is established by identifying the overlap rate of dual signal triggering at the same node. Based on the chemical evidence chain rating, the fingerprint offset direction is collected and combined with the preceding hash encoding to obtain a forward differential hash packet. The forward differential hash packet is used to verify the consensus voting of multiple chemical conclusions to form a chemical consensus record. The traceability chain block record is determined from the chemical consensus record to verify the drift dynamic consistency of the node. Based on the traceability chain block record, a fraud cluster set is obtained by identifying the topological association density between abnormal nodes. An anomaly type report is obtained by extracting the offset type and the degree of organization forgery from the fraud cluster set. The anomaly type report is then integrated with the chemical evidence chain rating to output a chemical traceability report.

2. The method according to claim 1, characterized in that, The step of generating the fingerprint offset direction from the offset directions of each node in the optical fingerprint data collection includes: The node absorption intensity set is obtained by extracting the intensity values ​​of key absorption bands of each node from the optical fingerprint data. The standardized deviations of each band from the mean of normal nodes are statistically analyzed to form a node intensity deviation set; Based on the node intensity deviation set, the positions of band deviations that violate the thermodynamic order of degradation are identified and a deviation sign vector is constructed. Based on the deviation sign vector, the fingerprint offset direction is generated by identifying the path type reversal event between adjacent nodes.

3. The method according to claim 1, characterized in that, The method of establishing an oil-source mixing marker by recognizing the multi-node progressive drift characteristics of the oxidized sterol ratio includes: The node increment sequence is generated by extracting the node ratio increment sequence of adjacent batches using the oxidized sterol ratio; The cumulative drift of the window is determined by calculating the average cumulative increment within the sliding window of the node increment sequence. Based on the cumulative drift amount of the window, accelerated drift windows exceeding the natural degradation limit are identified, forming over-limit drift segments; Based on the drift direction and amplitude of each drift segment exceeding the limit, an oil source mixing mark is established.

4. The method according to claim 1, characterized in that, The generation of processing imprints based on the combined characteristics of trace contaminants at each node of the oil source mixture marker analysis includes: To form an event node set, the set of sampling nodes corresponding to the mixed events is selected based on the oil source mixing markers. The pollutant element ratio is obtained by statistically analyzing the ratio of metal to organic pollutant concentrations at each node using the event node set. The facility matching confidence level is determined by assessing the degree of matching of facility source characteristics based on the pollutant element ratios. The processing imprint quantity is generated by summarizing the facility attribution judgment conclusions of each node based on the facility matching confidence level.

5. The method according to claim 1, characterized in that, The method of establishing a chemical evidence chain rating based on the degradation timestamp deviation and the processing imprint amount to identify the overlap rate of dual signals triggered at the same node includes: Extract the timestamp anomaly flags and deviation amplitudes of each node from the degradation timestamp deviation to construct a timestamp anomaly set; The abnormal timestamp quantity set is matched with the abnormal processing mark quantity flag of the corresponding node to determine the dual-mark set of the same node; Based on the aforementioned dual-marker set of the same node, the joint triggering ratio of dual signals and abnormally clean nodes is determined. Based on the aforementioned joint triggering ratio, a chemical evidence chain rating is established by classifying node fraud confidence.

6. The method according to claim 1, characterized in that, The step of collecting the fingerprint offset direction based on the chemical evidence chain rating and obtaining the forward differential hash packet by combining the preorder hash encoding includes: Based on the chemical evidence chain rating, the current node symbol sequence in the fingerprint offset direction is determined to generate a node symbol sequence; The node symbol sequence is concatenated with the hash value of the preceding node to form a joint hash input for the hash function; Based on the joint hash input, evidence levels are aggregated and the fingerprint offset direction is weighted to establish a node chain hash value; The forward differential hash packet is obtained by encapsulating the compliance credential and the geographic timestamp based on the chained hash value of the nodes.

7. The method according to claim 1, characterized in that, The process of obtaining a fraud cluster set based on the topological association density between abnormal nodes identified by the traceability chain block records includes: Based on the traceability chain block records, nodes whose chemical evidence chain rating is below the qualified threshold are identified to determine the chain block abnormal node set; For the set of abnormal nodes in the chain block, construct an abnormal node association graph with isolated off-chain nodes according to the adjacent relationship of the supply chain flow; The abnormal node association graph is subjected to time-series fraud behavior evolution analysis to establish fraud evolution feature quantities; Fraud clusters are formed by classifying fraud types and evolution stages based on the aforementioned fraud evolution characteristics.

8. The method according to claim 3, characterized in that, The process of identifying accelerated drift windows exceeding the natural degradation limit based on the cumulative drift amount of the window includes: Based on the cumulative drift of the window, the oil type characteristics of the starting node of each sliding window are analyzed to obtain an oil type identifier set; The maximum natural rate of sterol oxidation for the corresponding oil type is retrieved using the oil type identifier set to determine the upper limit of natural drift; The drift overshoot index is obtained by calculating the percentage by which the cumulative drift amount in the window exceeds the upper limit based on the natural drift limit. Based on the drift over-limit index, continuous window segments with an over-limit index that are continuously greater than zero are used to form over-limit drift segments.

9. The method according to claim 5, characterized in that, The step of matching the processing mark anomaly flags of the corresponding nodes to the timestamp anomaly set to determine the dual-mark set of the same node includes: The timestamp deviation magnitude is obtained by calculating the degradation rate deviation magnitude of each abnormal node based on the timestamp anomaly set. The two-dimensional normalized product is calculated based on the timestamp deviation magnitude and the processing mark quantity at the same node to obtain the mark consistency deviation. To assess the degree of anomaly between the timestamp and the processing mark at the same node and establish a two-dimensional trigger ratio based on the consistency deviation of the imprint; The set of dual-trigger nodes with the same node is determined by using the dual-trigger ratio to aggregate the set of dual-synchronous trigger nodes.

10. A traceability system for oils and fats based on a chain of evidence of chemical composition, characterized in that, include: The data acquisition module is used to acquire optical fingerprint data and chemical detection data, to collect the offset direction of each node from the optical fingerprint data to generate the fingerprint offset direction amount, and to determine the oxidized sterol ratio based on the sterol oxidation rate change of each node collected from the chemical detection data. Anomaly detection module is used to identify directional reversal between nodes based on the fingerprint offset direction to form directional anomaly markers, and to establish oil source mixing markers by identifying multi-node progressive drift characteristics of the ratio based on the oxidized sterol ratio. The evidence fusion module is used to compare the offset magnitude of the directional anomaly marker with the degradation rate over time to obtain the degradation timestamp deviation, analyze the trace pollutant combination characteristics of each node for the oil source mixed marker to generate the processing imprint, and identify the double signal trigger overlap rate of the same node based on the degradation timestamp deviation and the processing imprint to establish a chemical evidence chain rating. The hash evidence storage module is used to collect the fingerprint offset direction based on the chemical evidence chain rating and obtain a forward differential hash packet by combining the forward differential hash packet with the forward differential hash packet to verify the consensus voting of multiple chemical conclusions to form a chemical consensus record. The module then uses the chemical consensus record to verify the drift dynamics consistency of the node and determine the traceability chain block record. The report output module is used to obtain a fraud cluster set based on the topological association density between abnormal nodes identified by the traceability chain block records, extract the offset type and organizational forgery degree of the fraud cluster set to obtain an anomaly type report, and integrate the anomaly type report with the chemical evidence chain rating to output a chemical traceability report.

Images