A fast-moving consumer goods production and sales whole-process data penetration management method and system

Abstract

The application relates to the technical field of fast consumer product data management, and discloses a fast consumer product production and sales whole-process data penetration management method and system, which comprises the following steps: collecting production data and logistics node data, generating accumulated encoding carrying flow history through exclusive or superposition operation; updating the accumulated encoding at each flow node and obtaining transaction records; analyzing the accumulated encoding to extract flow features, and extracting space-time features from transaction data; aligning the flow features and the transaction features, checking space-time consistency to generate a scoring matrix; comparing the flow path and the authorized channel topology to generate a topology anomaly score; constructing a logistics fund joint graph by using a graph neural network, identifying abnormal node pairs, and calculating transaction authenticity confidence; fusing multidimensional scores to determine abnormalities, reversely analyzing the accumulated encoding to locate a responsible subject, and generating a traceability report. The application realizes whole-process data penetration management.

Description

Technical Field

[0001] This invention relates to the field of fast-moving consumer goods (FMCG) data management technology, and more specifically, to a method and system for comprehensive data management throughout the entire FMCG production and sales process. Background Technology

[0002] In managing the entire production and sales process, beverage companies need to integrate and manage production line data, logistics and transportation data, warehousing data, distributor transaction data, and fund settlement data.

[0003] In existing technologies, the system uses static coding to mark product batches. Data is collected independently at each stage and uploaded to the management platform. Data is correlated and queried through batch numbers, and production reports, logistics reports, and sales reports are output respectively, realizing process monitoring of beverages from production to sales.

[0004] However, the above solutions have the following technical problems: static encoding only carries product identity information and cannot carry circulation history; insufficient data reporting rate in intermediate links leads to data gaps; even if the terminal queries, it can only obtain the first and last node information; production data, logistics data, and transaction data come from different systems, and the data formats are heterogeneous, making it difficult to perform cross-link consistency verification; each system operates independently, making it impossible to detect logical contradictions between logistics location and transaction record location; when the terminal detects data anomalies, it is difficult to trace back and locate the specific data forgery link and the responsible party. Summary of the Invention

[0005] This invention provides a method and system for data penetration management throughout the entire process of fast-moving consumer goods production and sales, solving the technical problem in related technologies that it is impossible to accurately trace and identify anomalies in the entire process of beverage products from production to sales.

[0006] This invention discloses a method for data penetration management throughout the entire process of fast-moving consumer goods (FMCG) production and sales, comprising the following steps: Collect the filling timestamp, production batch number, and production line number of the production line. Use a hash algorithm to generate a basic dynamic identification code from the above data. Simultaneously collect the sensor sampling value, sampling timestamp, and node geographical location of the first logistics node. Perform an XOR operation on the node triplet data and the basic dynamic identification code to generate a cumulative code carrying the initial production state. When a product flows to each logistics node and sales node, the current node data is collected and the cumulative code is subjected to a multi-level XOR operation to generate an updated cumulative code. Transaction records are obtained simultaneously to generate a transaction-settlement matching dataset. Differential parsing is performed on the cumulative encoding to calculate the state change rate, time interval, and spatial displacement between adjacent nodes, generating a flow feature vector sequence, and extracting the transaction spatiotemporal feature set from the transaction-settlement pairing dataset; The flow feature vector sequence is aligned with the transaction spatiotemporal feature set on the time axis, and the spatiotemporal consistency verification algorithm is used to detect geographical deviation and logical consistency, generating a spatiotemporal consistency scoring matrix. Input the flow feature vector sequence into the channel topology constraint rule engine, compare the actual flow path with the authorized channel network topology, and generate a topology anomaly score; A graph neural network model is used to jointly model the relationship between logistics flow and capital flow, identify abnormal node pairs, and calculate the confidence level of transaction authenticity. The system uses topological anomaly scores, spatiotemporal consistency scores, and transaction authenticity confidence levels to make joint decisions to determine anomalies, and reverse-analyzes the cumulative coding to locate the responsible party and generate a traceability report.

[0007] Furthermore, the XOR superposition operation includes: concatenating sensor sample values, sampling timestamps, and node geographical locations into a data string according to a fixed format; converting the data string into a binary code of the same length as the basic dynamic identifier code; using zero padding if the data string length is insufficient; and using segmented XOR folding if the data string length exceeds the limit; and performing a bitwise XOR operation between the converted binary code and the basic dynamic identifier code.

[0008] Furthermore, the differential analysis includes: utilizing the reversibility of the XOR operation, performing an XOR operation on the cumulative code of the current node and the binary code of the known collected data of the current node to restore the cumulative code of the previous node, recursively executing the restoration process to extract the original data of each node layer by layer, calculating the difference value of the extracted original data between adjacent layers to form differential features; the state change rate is the ratio of the sensor reading difference between adjacent nodes to the time interval.

[0009] Furthermore, the spatiotemporal consistency verification algorithm includes: calculating the distance between the geographical location of each flow node and the transaction location in the corresponding transaction record; when the distance exceeds a geographical threshold, the geographical deviation score of the node is the ratio of the geographical threshold to the distance value; calculating the time difference between the logistics arrival time and the transaction time; when the time difference exceeds a time threshold, the time consistency score of the node is calculated according to the ratio of the time threshold to the time difference; and verifying the logical consistency between the logistics status change trend and the transaction status; when the logistics status shows that the goods are in transit while the transaction record shows that the settlement has been completed, the logical consistency score is zero.

[0010] Furthermore, the geographical threshold is determined according to the node type, with different geographical thresholds set for warehousing nodes, transportation nodes, distributor nodes, and terminal sales nodes; the time threshold is determined according to the standard logistics timeliness between adjacent nodes, with different time thresholds set for nodes within the same city and nodes across provinces.

[0011] Furthermore, the calculation process of the topology anomaly score includes: matching the node type identifier in the flow feature vector sequence with the nodes in the authorized channel network topology to generate the actual flow path; comparing the actual flow path with the compliant flow path library; using the edit distance algorithm to calculate the edit distance between the actual flow path and the most similar compliant path; calculating the topology anomaly score based on the edit distance; when the edit distance is zero, the score is full marks; for each increase in edit distance, the score decreases by a preset decay coefficient.

[0012] Furthermore, the graph neural network model includes: a node feature encoding layer, which encodes the attribute features of each node into a fixed-dimensional feature vector; an edge feature encoding layer, which encodes the attribute features of each edge into a fixed-dimensional feature vector; a message passing layer, which uses a graph attention mechanism to aggregate the feature information of its neighboring nodes and combine it with edge features to perform weighted updates to the node embedding vector; a joint modeling layer, which concatenates the embedding vectors of logistics flow edges and capital flow edges to generate a joint embedding representation; and an anomaly detection layer, which calculates the consistency score between the logistics direction and the net capital flow of each node pair based on the joint embedding representation and outputs a set of abnormal node pairs.

[0013] Furthermore, the calculation of transaction authenticity confidence includes: extracting the time series of historical transaction records of abnormal node pairs from the transaction-settlement pairing dataset, extracting the dominant frequency component by performing a Fourier transform on the time series, determining the existence of a periodic hedging transaction pattern and reducing the transaction authenticity confidence when the periodic hedging transaction pattern is within a preset suspicious period range; when no periodic hedging transaction pattern is detected, the transaction authenticity confidence is equal to the consistency score, and when a periodic hedging transaction pattern is detected, the transaction authenticity confidence is the consistency score multiplied by the periodic penalty coefficient.

[0014] Furthermore, the reverse parsing cumulative coding for locating the responsible party adopts a binary search strategy: the flow path is divided into a first half and a second half according to the time sequence, and the cumulative coding of the end node of the first half and the beginning node of the second half are checked for consistency. If the cumulative coding of the end node of the first half passes the check but the cumulative coding of the beginning node of the second half fails the check, the anomaly occurs in the second half. The division and verification process is recursively performed on the half where the anomaly is located until the specific anomaly node is located, and the responsible parties of the node and all subsequent nodes are included in the responsible party chain.

[0015] This invention provides a data penetration management system for the entire process of fast-moving consumer goods (FMCG) production and sales, comprising: The cumulative code generation module is used to collect production line data to generate basic dynamic identification codes, and perform an XOR operation with the data of the first logistics node to generate a cumulative code carrying the initial production state. The data acquisition module is used to collect data at each transfer node and update the cumulative code, and synchronously obtain transaction records to generate a transaction-settlement matching dataset; The feature extraction module is used to perform differential parsing on the cumulative encoding to generate a sequence of flow feature vectors, and to extract the spatiotemporal feature set of transactions from the transaction-settlement pairing dataset; The spatiotemporal consistency verification module is used to align the flow feature vector sequence with the transaction spatiotemporal feature set on the time axis and execute the spatiotemporal consistency verification algorithm to generate a spatiotemporal consistency scoring matrix. The topology anomaly scoring module is used to compare the actual flow path with the authorized channel network topology to generate a topology anomaly score; The joint modeling module is used to jointly model the relationship between logistics flow and capital flow using graph neural network models, identify abnormal node pairs, and calculate the confidence level of transaction authenticity. The traceability report generation module is used to integrate multi-dimensional scores for joint decision-making to identify anomalies and reverse-analyze cumulative codes to locate the responsible party and generate a traceability report.

[0016] This invention solves the data discontinuity problem caused by static coding by using a multi-layer XOR stacking mechanism of cumulative coding to enable product codes to carry a complete circulation history; it solves the difficulty of cross-stage consistency verification by uniformly converting heterogeneous data into feature vectors through differential parsing; it solves the problem of independent systems being unable to cross-verify by detecting the geographical deviation and logical consistency of logistics and transaction data through a spatiotemporal consistency verification algorithm; it improves the ability to identify fraudulent transactions by jointly modeling the relationship between logistics and capital flow and detecting periodic hedging transaction patterns through graph neural networks; and it achieves the location of abnormal links and the tracing of responsible parties through a joint decision-making mechanism and cumulative coding reverse parsing, forming a full-process data penetration management capability. Attached Figure Description

[0017] Figure 1 This is a flowchart of the beverage production and sales process data penetration management method provided in this embodiment of the invention; Figure 2 This is the temperature change trend of the flow node provided in the embodiment of the present invention; Figure 3 This is the cumulative encoding generation process data stream provided in the embodiments of the present invention; Figure 4 This refers to the flow of transaction amounts at each node as provided in the embodiments of the present invention; Figure 5 This is the multidimensional analysis of flow feature vectors provided in the embodiments of the present invention; Figure 6 This is a heatmap of the spatiotemporal consistency scoring matrix provided in an embodiment of the present invention; Figure 7 This refers to the consistency score of abnormal node pairs provided in this embodiment of the invention. Figure 8 This is a comprehensive anomaly score calculation component provided in the embodiments of the present invention. Detailed Implementation

[0018] Beverage companies need to manage production line data, logistics data, warehousing data, distributor transaction data, and financial settlement data across the entire production and sales process. Current technology uses static coding to mark product batches, with each stage independently collecting data and uploading it to the management platform. Data is then linked and queried using batch numbers, generating production reports, logistics reports, and sales reports to monitor the beverage process from production to sales.

[0019] However, the above solutions have the following technical problems: static encoding only carries product identity information and cannot carry circulation history; insufficient data reporting rate in intermediate links leads to data gaps; even if the terminal queries, it can only obtain the first and last node information; production data, logistics data, and transaction data come from different systems, and the data formats are heterogeneous, making it difficult to perform cross-link consistency verification; each system operates independently, making it impossible to detect logical contradictions between logistics location and transaction record location; when the terminal detects data anomalies, it is difficult to trace back and locate the specific data forgery link and the responsible party.

[0020] At least one embodiment of the present invention discloses a method for data penetration management of the entire process of fast-moving consumer goods production and sales, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect production data and the first logistics node data to generate a cumulative code carrying the initial production status; Obtain the filling timestamp, production batch number, and production line number of the beverage production line. Use a hash algorithm to generate a basic dynamic identifier code from the above data. Simultaneously collect the sensor sampling values, sampling timestamp, and node geographical location of the first logistics node. Perform an XOR operation on the node triplet data and the basic dynamic identifier code to generate a cumulative code carrying the initial production state.

[0021] Furthermore, the input to the hash algorithm is a string concatenated from the filling timestamp, production batch number, and production line number in a preset order, and the output is a fixed-length basic dynamic identifier code.

[0022] Furthermore, the XOR superposition operation involves converting the sensor sample values, sampling timestamps, and node geographical locations into binary codes, and then XORing them with the corresponding bits of the basic dynamic identification code to generate a new encoding sequence. The specific implementation process is as follows: First, the sensor sample values, sampling timestamps, and node geographical locations are concatenated into a data string according to a fixed format. Then, the data string is converted into a binary code of the same length as the basic dynamic identification code. If the data string length is insufficient, zero padding is used; if the data string length exceeds the limit, segmented XOR folding is used. Finally, the converted binary code is XORed bit by bit with the basic dynamic identification code.

[0023] Furthermore, the node triplet data is a data structure formed by splicing three data items—sensor sample value, sampling timestamp, and node geographical location—in a fixed format. The sensor sample value includes at least one of the following: temperature sensor reading, humidity sensor reading, and weight sensor reading.

[0024] Step 2: Collect data at each transfer node and update the cumulative code, and synchronously obtain transaction records to generate a transaction-settlement matching dataset; When a product flows to each logistics node and sales node, the sensor readings, geographic coordinates, node type identifier, and arrival timestamp of the current node are collected. The collected data is then XORed with the cumulative code in chronological order to generate an updated cumulative code that carries the complete flow history. The transaction records of the node are obtained simultaneously, including the delivery time, arrival time, transaction amount, and fund transfer account information, to generate a transaction-settlement matching dataset.

[0025] Furthermore, the multi-layer XOR overlay operation involves converting the current node's data into binary code at each node, then XORing it with the cumulative code generated by the previous node. This results in the final cumulative code overlaying all the flow information from the production source to the current node. The specific implementation process is the same as the XOR overlay operation in step 1: the sensor readings, geographic coordinates, node type identifier, and arrival timestamp of the current node are concatenated and converted into binary code of the same length as the cumulative code, and then XORed bit by bit with the cumulative code of the previous node.

[0026] Furthermore, the transaction-settlement matching dataset is a structured data set formed by associating and matching the shipping records, arrival records, transaction amount, and fund account information of the same transaction.

[0027] Furthermore, the node type identifiers include five types: production node, warehousing node, transportation node, distributor node, and terminal sales node. Each type is distinguished by a different coding prefix.

[0028] Step 3: Parse the cumulative encoding to extract flow features and extract spatiotemporal features from the transaction data; Differential parsing is performed on the cumulative encoding to calculate the state change rate, time interval, and spatial displacement between adjacent nodes, generating a sequence of flow feature vectors; transaction location information, delivery and arrival time difference, and the ratio of settlement amount to transaction amount are extracted from the transaction-settlement pairing dataset to generate a transaction spatiotemporal feature set.

[0029] Furthermore, differential parsing involves decomposing the cumulative code in reverse according to the stacking hierarchy to extract the difference information between each layer. The specific process is as follows: First, using the reversibility of the XOR operation, the cumulative code of the current node is XORed with the binary code of the known collected data of the current node to restore the cumulative code of the previous node; then, the above restoration process is recursively executed to extract the original data of each node layer by layer; finally, the difference value of the extracted original data between adjacent layers is calculated to form the differential feature.

[0030] Furthermore, the rate of state change is the ratio of the difference in sensor readings between adjacent nodes to the time interval, calculated using the following formula: in, Represents the rate of change of state. Indicates the first Sensor readings at each node, Indicates the first Sensor readings at each node, Indicates the first The arrival timestamp of each node. Indicates the first The arrival timestamp of each node. Indicates the node sequence number.

[0031] Furthermore, spatial displacement is calculated using Euclidean distance to determine the distance between the geographical coordinates of adjacent nodes.

[0032] Furthermore, each feature vector in the flow feature vector sequence contains five dimensions: state change rate, time interval, spatial displacement, node type encoding, and number of hours of arrival timestamp.

[0033] Furthermore, each feature record in the transaction spatiotemporal feature set contains four dimensions: the latitude and longitude of the transaction location, the time difference between delivery and arrival, the settlement transaction amount ratio, and the time period identifier of the transaction occurrence.

[0034] Step 4: Align the circulation features with the transaction features, verify spatiotemporal consistency, and generate a scoring matrix; The flow feature vector sequence is aligned with the transaction spatiotemporal feature set on the time axis. The spatiotemporal consistency verification algorithm is used to detect the geographical deviation between the logistics location and the transaction record location, and the logical consistency between the logistics status change trend and the transaction status, and a spatiotemporal consistency scoring matrix is ​​generated.

[0035] Furthermore, timeline alignment involves matching the arrival timestamps in the flow feature vector sequence with the transaction times in the transaction spatiotemporal feature set to establish a time correspondence between logistics events and transaction events. Specifically, the process involves traversing each node in the flow feature vector sequence, finding the transaction record with the closest timestamp in the transaction spatiotemporal feature set, and establishing a correspondence when the time difference is within a preset matching window range, which is set to 6 hours before and after.

[0036] Furthermore, the geographical deviation is the distance between the geographical location of the logistics node and the transaction location in the transaction record.

[0037] Furthermore, the spatiotemporal consistency scoring matrix is ​​a two-dimensional matrix, where the row index corresponds to the sequence number of the flow node, and the column index corresponds to the verification dimension, including the geographical deviation score, the temporal consistency score, and the state logic consistency score. The matrix element values ​​are normalized score values, ranging from 0 to 1.

[0038] Furthermore, the spatiotemporal consistency verification algorithm includes the following processing steps: Step 4.1: Calculate the distance between the geographical location of each transfer node and the transaction location in the corresponding transaction record. When the distance exceeds the preset geographical threshold, the geographical deviation score of the node is the ratio of the geographical threshold to the distance value; otherwise, the score is 1. Step 4.2: Calculate the time difference between the logistics arrival time and the transaction shipment time or delivery time. When the time difference exceeds the preset time threshold, the time consistency score of this node is calculated as the ratio of the time threshold to the time difference; otherwise, the score is 1. Step 4.3: Verify the logical consistency between the trend of logistics status changes and the transaction status. When the logistics status shows that the goods are in transit but the transaction record shows that the settlement has been completed, the logical consistency score is 0; otherwise, it is 1.

[0039] Furthermore, the geographical threshold is determined based on the node type. The geographical threshold is set to 500 meters for warehousing nodes, 5000 meters for transportation nodes, and 1000 meters for distributor nodes and terminal sales nodes.

[0040] Furthermore, the time threshold is determined based on the standard logistics timeliness between adjacent nodes. The time threshold is set to 24 hours for nodes within the same city and 72 hours for nodes across provinces.

[0041] Step 5: Compare the circulation path with the authorized channel topology, identify deviating nodes, and generate a topology anomaly score; The flow feature vector sequence is input into the preset channel topology constraint rule engine. By comparing the actual flow path with the authorized channel network topology, abnormal flow nodes that deviate from the authorized path are identified, and a topology anomaly score is generated.

[0042] Furthermore, the channel topology constraint rule engine includes topology data of authorized channel networks and a library of compliant flow paths.

[0043] Furthermore, the topology data of the authorized channel network is stored in the form of a directed graph, where nodes represent authorized distributors or logistics nodes and edges represent permitted flow directions.

[0044] Furthermore, the compliance circulation path library stores a set of all compliance paths from the source of production to each terminal sales point.

[0045] Furthermore, the calculation process for the topological anomaly score includes: Step 5.1: Match the node type identifiers in the flow feature vector sequence with the nodes in the authorized channel network topology to generate the actual flow path; Step 5.2: Compare the actual workflow path with the compliant workflow path library, and use the edit distance algorithm to calculate the edit distance between the actual workflow path and the most similar compliant path. The input of the edit distance algorithm is the node sequence of the actual workflow path and the compliant path, and the output is the minimum number of node insertion, deletion and replacement operations required to convert the actual workflow path into a compliant path. Step 5.3: Calculate the topology anomaly score based on the edit distance. The score is 1 when the edit distance is 0, and decreases by a preset decay coefficient for each increase of 1 in the edit distance. The formula for calculating the topology anomaly score is as follows: in, Indicates topological anomaly score, This indicates a preset attenuation coefficient with a value between 0 and 1. Indicates the edit distance.

[0046] Furthermore, a preset attenuation coefficient is used. The value is 0.8.

[0047] Step 6: Construct a joint logistics fund graph, identify abnormal node pairs, and calculate the confidence level of transaction authenticity; By using a graph neural network model to embed the channel network, the logistics flow relationship and the capital flow relationship are jointly modeled to identify abnormal node pairs where the logistics direction and the net capital flow are inconsistent. For abnormal node pairs, the time regularity features of their transactions are extracted to detect periodic hedging transaction patterns and calculate the confidence level of transaction authenticity.

[0048] Furthermore, the input to the graph neural network model is a heterogeneous graph constructed based on the channel network, where nodes include logistics nodes and transaction entity nodes, and edges include logistics flow edges and capital flow edges.

[0049] Furthermore, the logistics direction refers to the flow of goods from upstream nodes to downstream nodes, while the net capital flow direction refers to the flow of funds from downstream nodes to upstream nodes. When the logistics direction and the net capital flow direction are inconsistent on the same node pair, the node pair is marked as an abnormal node pair. The specific determination process is as follows: For two connected nodes, if the logistics flow edge shows goods flowing from node A to node B, then under normal circumstances, the capital flow edge should show funds flowing from node B to node A. When the capital flow edge shows funds flowing from node A to node B or there is no corresponding capital flow edge, the node pair is determined to be an abnormal node pair.

[0050] Furthermore, before inputting the graph neural network model, data preprocessing is performed on node attribute features and edge attribute features. For geographic location coordinates in node attribute features, mean normalization based on range is used; for numerical features such as historical transaction size, Z-score standardization is used; for categorical features such as node type, one-hot encoding is used to convert them into numerical vectors; for transaction amount in edge attribute features, Z-score standardization is used; for transaction time, it is converted into timestamps and then mean normalization based on range is used; for categorical features such as edge type, one-hot encoding is used to convert them into numerical vectors.

[0051] Furthermore, the graph neural network model includes the following components and their data transmission relationships: Node feature encoding layer: Encodes the attribute features of each node into a fixed-dimensional feature vector. The attribute features include node type, geographical location, and historical transaction size. The node feature encoding layer uses a fully connected neural network to map the original attribute features into a fixed-dimensional embedding vector. The output of the node feature encoding layer is passed to the message passing layer. Edge feature encoding layer: Encodes the attribute features of each edge into a fixed-dimensional feature vector. The attribute features include edge type, transaction amount, and transaction time. The edge feature encoding layer uses a fully connected neural network to map the original edge attribute features into a fixed-dimensional embedding vector. The output of the edge feature encoding layer is passed to the message passing layer. Message Passing Layer: Employs a graph attention mechanism, where each node aggregates the feature information of its neighbors and combines it with edge features for weighted summation, updating its own node embedding vector. Specifically, for each node... Calculate its relationship with each neighbor node Attention weights between The formula for calculating attention weights is: in, Represents a node For neighboring nodes Attention weights Represents a node with neighboring nodes The original attention scores between them Represents a node The set of neighboring nodes, Represents the set of neighboring nodes Node index in Represents an exponential function. Original attention score. The calculation formula is: in, Represents a node with neighboring nodes The original attention scores between them This represents a learnable weight vector. Indicates transpose. Represents a node eigenvectors, Representing neighboring nodes eigenvectors, Indicates the connection node and nodes The eigenvectors of the edges, This represents a vector concatenation operation. This represents the LeakyReLU activation function.

[0052] Then, attention weights are used to perform a weighted summation of the features of neighboring nodes, and the nodes are updated. The embedding vector is updated using the following formula: in, Represents a node Updated embedding vector, Represents a non-linear activation function. Represents a node For neighboring nodes Attention weights Representing neighboring nodes eigenvectors, Represents a node The set of neighboring nodes, This represents the neighbor node index.

[0053] The output of the message passing layer is passed to the joint modeling layer; Joint Modeling Layer: This layer concatenates the embedding vectors of logistics flow edges and capital flow edges to generate a joint embedding representation for connecting nodes. and nodes The node pairs are jointly embedded as the concatenation of the logistics flow edge embedding vector and the capital flow edge embedding vector, and the output of the joint modeling layer is passed to the anomaly detection layer. Anomaly Detection Layer: Based on joint embedding representation, the consistency score between the logistics direction and the net capital flow of each node pair is calculated. The specific process is as follows: the joint embedding representation is input into a fully connected neural network, and the consistency score is output. The fully connected neural network contains two hidden layers and the last layer is the output layer. The Sigmoid activation function is used to map the output to between 0 and 1, and the set of abnormal node pairs is output.

[0054] Furthermore, the input layer of the graph neural network model receives heterogeneous graph data, which includes node feature matrices and edge feature matrices. Each row of the node feature matrix corresponds to the attribute features of a node, and each row of the edge feature matrix corresponds to the attribute features of an edge.

[0055] Furthermore, the output layer of the graph neural network model is the output layer of the fully connected neural network in the anomaly detection layer. The output layer uses the Sigmoid activation function to output the set of abnormal node pairs and their corresponding consistency scores. The consistency scores range from 0 to 1. The lower the score, the higher the degree of inconsistency between the logistics direction and the net capital flow of the node pair.

[0056] Furthermore, the graph neural network model is trained using supervised learning. The training dataset includes labeled normal node pairs and abnormal node pairs. The loss function is the binary cross-entropy loss function, calculated as follows: in, This represents the value of the loss function. Indicates the number of training samples. Indicates the first The true label of each sample Indicates the first The prediction consistency score of each sample. This indicates the sample index.

[0057] Furthermore, the Adam optimization algorithm is used to update the parameters during the training of the graph neural network model.

[0058] Furthermore, the embedding vectors output by the fully connected neural networks in both the node feature encoding layer and the edge feature encoding layer are 128-dimensional.

[0059] Furthermore, the number of neurons in the two hidden layers of the fully connected neural network in the anomaly detection layer is 64 and 32, respectively, and the ReLU activation function is used.

[0060] Furthermore, the detection process for cyclical hedging trading patterns includes: Step 6.1: For abnormal node pairs, extract the time series of their historical transaction records from the transaction-settlement pairing dataset; Step 6.2: Extract the main frequency component by performing Fourier transform on the time series. The input of the Fourier transform is the transaction time series, and the output is the main frequency component represented in the frequency domain. Step 6.3: When the cycle corresponding to the main frequency component is within the preset suspicious cycle range, it is determined that there is a periodic hedging transaction pattern, which reduces the confidence of the transaction authenticity. The suspicious cycle range is an abnormal cycle interval determined according to the normal business transaction cycle. When the transaction shows a fixed periodic repetition and the cycle falls into this interval, it is determined to be suspicious.

[0061] Furthermore, the suspicious period range is set between 7 and 30 days, which corresponds to the typical time interval of artificially designed periodic hedging transactions.

[0062] Furthermore, the confidence level of transaction authenticity is calculated as follows: when no periodic hedging transaction pattern is detected, the confidence level of transaction authenticity is equal to the consistency score output by the graph neural network model; when a periodic hedging transaction pattern is detected, the confidence level of transaction authenticity is the consistency score multiplied by a preset periodic penalty coefficient, and the periodic penalty coefficient ranges from 0 to 1.

[0063] Furthermore, the periodic penalty coefficient is set to 0.5.

[0064] Step 7: Integrate multi-dimensional scoring to determine anomalies, reverse analyze cumulative codes to locate the responsible party, and generate a traceability report; The topological anomaly score, the scores of each dimension in the spatiotemporal consistency score matrix, and the confidence level of transaction authenticity are used for joint decision-making. When the comprehensive anomaly score is lower than the preset threshold, it is judged as a data anomaly. Based on the spatiotemporal characteristics of the anomaly node, the superimposed information of each layer in the cumulative encoding is reverse-analyzed to extract the complete flow timeline, the time of anomaly occurrence, and the chain of responsible entities, and generate a full-process data penetration and traceability report.

[0065] Furthermore, the joint decision-making process employs a weighted summation method, summing the mean scores of each dimension in the topological anomaly score and the spatiotemporal consistency score matrix, along with the transaction authenticity confidence score, according to preset weights, to obtain a comprehensive anomaly score. The formula for calculating the comprehensive anomaly score is as follows: in, Indicates the overall anomaly score. , , Represent the preset weights corresponding to the topological anomaly score, the mean of the spatiotemporal consistency score, and the confidence level of the transaction authenticity, respectively, and satisfy the following conditions: , Indicates topological anomaly score, This represents the mean of all elements in the spatiotemporal consistency score matrix. This indicates the confidence level in the authenticity of the transaction.

[0066] Furthermore, preset weights , , The values ​​are 0.3, 0.4, and 0.3 respectively.

[0067] Furthermore, the preset threshold is set to 0.6. When the comprehensive anomaly score is lower than 0.6, it is judged as data anomaly.

[0068] Furthermore, the reverse analysis of the cumulative encoding, based on the reversibility of the XOR operation, utilizes the known data collected from each node to reconstruct the encoding state of each intermediate step layer by layer from the final cumulative encoding, thereby determining the specific step where the anomaly occurred. The specific implementation process is as follows: starting from the cumulative encoding of the final node, the cumulative encoding is XORed with the binary encoding of the known collected data of the final node to reconstruct the cumulative encoding of the second-to-last node. Then, the cumulative encoding of the second-to-last node is XORed with the binary encoding of the known collected data of the second-to-last node to reconstruct the cumulative encoding of the third-to-last node, and so on recursively until the production source node. During the reconstruction process, the cumulative encoding reconstructed from each node is compared with the theoretical cumulative encoding recalculated based on the known collected data of that node and all previous nodes. The first node where inconsistency occurs is the step where the anomaly occurred.

[0069] Furthermore, the chain of responsible parties is an ordered set of all nodes and their corresponding responsible parties, starting from the node where the data anomaly first appears and ending at the terminal node where the data anomaly is detected.

[0070] Furthermore, to improve tracing efficiency, the reverse parsing process employs a binary search strategy: Step 7.1: Divide the flow path into the first half and the second half according to the time sequence; Step 7.2: Perform consistency verification on the cumulative codes of the end nodes of the first half and the beginning nodes of the second half respectively. The consistency verification is to compare the cumulative code of the node with the theoretical code value calculated based on the known collected data. Step 7.3: If the cumulative encoding verification of the end node of the first half passes but the cumulative encoding verification of the beginning node of the second half fails, the exception occurs in the second half; otherwise, the exception occurs in the first half. Step 7.4: Recursively perform the above division and verification process on the half segment where the anomaly is located until the specific anomaly node is located. Mark the node as the node corresponding to the time when the anomaly occurred, and include the responsible parties of the node and all subsequent nodes into the responsible party chain.

[0071] According to this embodiment, the product code carries a complete production and sales history through a multi-layer XOR overlay mechanism of cumulative coding. Since the cumulative code is generated by XORing the current node data with the historical cumulative code at each flow node, even if the data reporting rate of intermediate links is insufficient, the cumulative code of the terminal node still contains complete flow information. The complete trajectory can be restored through reverse parsing, thereby overcoming the data discontinuity problem caused by static coding.

[0072] By using differential parsing, logistics data and transaction data are uniformly converted into a sequence of flow feature vectors and a spatiotemporal feature set of transactions, eliminating the differences in data format between heterogeneous systems. This enables cross-process data from different systems to be compared and verified in a unified feature space, solving the difficulty of cross-process consistency verification caused by heterogeneous data formats.

[0073] By using a spatiotemporal consistency verification algorithm to detect the geographical deviation between the logistics location and the transaction record location, as well as the logical consistency between the logistics status change trend and the transaction status, it is possible to discover geographical and temporal logical contradictions between logistics data and transaction data, thus solving the problem that cross-validation cannot be performed when each system is running independently.

[0074] By jointly modeling the logistics flow relationship and the capital flow relationship through a graph neural network model, the authenticity of transactions can be verified from the capital dimension. It can identify abnormal node pairs where the logistics direction and the net capital flow are inconsistent. Furthermore, by using Fourier transform to detect periodic hedging transaction patterns, it improves the ability to identify collusion among dealers to conduct fraudulent transactions.

[0075] By using a joint decision-making mechanism to comprehensively assess topological anomalies, spatiotemporal consistency scores, and transaction authenticity confidence, and combining the reverse analysis capability of cumulative coding with a binary search strategy, the system has been able to locate anomalies and trace the responsible parties, thus forming a full-process data penetration management capability from production to sales.

[0076] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

[0077] In May 20XX, a beverage company implemented end-to-end data management for its carbonated beverage product, "Soda Type A". This batch of products, after leaving its production base in East China, passed through a regional warehousing center, provincial distributors, city-level distributors, and finally reached end retailers. The company needed to implement data traceability management for this batch of products from production line bottling to end-sales, monitoring the authenticity of data at each stage of logistics, warehousing, and transactions. This batch of products, numbered batch20XX0515 and production line number line03, was filled on May 15, 20XX at 08:23:45, with its initial logistics node being the finished product temporary storage area at the production base.

[0078] like Figure 2-8 As shown, it includes the following steps: Upon completion of filling on the production line, the system collects three basic data points: filling timestamp, production batch number, and production line number. Simultaneously, IoT sensors deployed in the finished product storage area collect temperature readings, sampling timestamps, and geographic coordinates of the first logistics node. The basic data is processed using a hash algorithm to generate a basic dynamic identifier code. Then, the sensor triplet data is XORed with the basic dynamic identifier code to generate a cumulative code containing the initial production state.

[0079] Table 1 Production Basic Data and First Node Data: The filling timestamp, production batch number, and production line number are concatenated into the string "20XX051508234500batch20XX0515line03" in the order of "timestamp + batch number + production line number". This string is then used to generate a 256-bit basic dynamic identifier using the SHA-256 hash algorithm. The temperature reading of 18.5°C, the sampling timestamp (converted to a Unix timestamp of 1715744712), and the geographic coordinates (converted to numerical codes) are concatenated into a data string. This string is then converted to a 256-bit binary code and XORed bitwise with the basic dynamic identifier to generate a cumulative code code_v1 carrying the initial production status.

[0080] Table 2. Cumulative encoding generation process data; The product flows through four nodes: regional warehousing center (node ​​2), provincial distributor warehouse (node ​​3), city-level distributor (node ​​4), and terminal retailer (node ​​5). At each node, sensor readings, geographical location, node type, and arrival timestamp are collected. These are then XORed with the accumulated code of the previous node to progressively update the accumulated code. Simultaneously, transaction records for each node are acquired, including shipping time, arrival time, transaction amount, and fund transfer account.

[0081] Table 3 Data collected at each circulation node: Table 4 Transaction-Settlement Matching Dataset: At node 2, the temperature 16.8, timestamp 1715766622, and position 121.456 and 31.678 are concatenated and converted into a 256-bit binary code. This code is then XORed with code_v1 to generate code_v2. Nodes 3, 4, and 5 perform the same XOR concatenation process sequentially, ultimately generating code_v5, which contains the complete flow history.

[0082] Differential parsing is performed on the cumulative encoding code_v5. Utilizing the invertibility of the XOR operation, code_v4, code_v3, code_v2, and code_v1 are reconstructed layer by layer from code_v5. The original data between each layer is extracted, and the state change rate, time interval, and spatial displacement between adjacent nodes are calculated to generate a sequence of flow feature vectors. Simultaneously, transaction location, delivery-arrival time difference, and settlement transaction amount ratio are extracted from the transaction-settlement pairing dataset to generate a transaction spatiotemporal feature set.

[0083] Table 5: Flow Feature Vector Sequence Example of calculating the rate of change of state (node ​​1 → node 2): Table 6: Spatiotemporal Feature Set of Transactions The arrival timestamps in the flow feature vector sequence are matched with the delivery times in the transaction spatiotemporal feature set to establish a correspondence between logistics events and transaction events. The distance deviation between the geographical location of the logistics node and the location of the transaction record is calculated to verify the consistency between the logistics arrival time and the transaction delivery time, and to verify the logical consistency between the logistics status and the transaction status, generating a spatiotemporal consistency scoring matrix.

[0084] Table 7 Spatiotemporal Consistency Scoring Matrix: Geographic deviation calculation for node 2: The logistics location (121.456°, 31.678°) is completely consistent with the transaction location (121.456°, 31.678°), with a distance of 0 meters, which is less than the geographical threshold of 500 meters for warehouse nodes. The geographical deviation score is 1.00.

[0085] Time consistency calculation for node 3: The logistics arrival time is 20XX-05-16 09:15:38, the transaction arrival time is 20XX-05-16 09:15:38, the time difference is 0 hours, which is less than the cross-province node time threshold of 72 hours. Since the time is completely consistent, but considering that the actual logistics time of 18.75 hours is within a reasonable range, the time consistency score is 0.95.

[0086] The node type identifier sequence "Production Node → Warehousing Node → Distributor Node → Distributor Node → Terminal Sales Node" in the circulation feature vector sequence is compared with the authorized channel network topology. The compliant path defined by the authorized channel network is "Production Node → Warehousing Node → Distributor Node → Distributor Node → Terminal Sales Node". The difference between the actual circulation path and the compliant path is calculated using an edit distance algorithm.

[0087] The actual workflow path perfectly matches the compliant path, with edit distance. According to the topological anomaly scoring formula: The topology anomaly score is 1.00, indicating that the circulation path fully conforms to the authorized channel topology structure and no path deviation was found.

[0088] Step 6 Implementation Example: Identify abnormal node pairs and calculate the confidence level of transaction authenticity. A heterogeneous graph is constructed, consisting of five logistics nodes (nodes 1 to 5) and corresponding transaction entity nodes: ACC_prod01, ACC_store02, ACC_dealer03, ACC_dealer04, and ACC_retail05. The logistics flow edges are: Node 1 → Node 2 → Node 3 → Node 4 → Node 5. The fund flow edges are established based on the transaction-settlement pairing dataset, with directions: ACC_store02 → ACC_prod01, ACC_dealer03 → ACC_store02, ACC_dealer04 → ACC_dealer03, and ACC_retail05 → ACC_dealer04.

[0089] Node attribute features (node ​​type, geographical location, historical transaction size) and edge attribute features (edge ​​type, transaction amount, transaction time) are input into the graph neural network model. The node feature encoding layer and the edge feature encoding layer map the original features into 128-dimensional embedding vectors. The message passing layer uses a graph attention mechanism to aggregate neighbor information.

[0090] Table 8. Detection results of abnormal node pairs: All node pairs showed consistent logistics flow and net capital flow, with consistency scores all above 0.85, and no abnormal node pairs were detected.

[0091] For the time series of historical transaction records extracted from each node, Fourier transform was used to extract the dominant frequency component. The transaction time series of nodes 3-4 showed that the dominant frequency component corresponds to a period of 45 days, which is not within the suspicious period range (7 to 30 days), and no periodic hedging transaction pattern was detected.

[0092] Transaction authenticity confidence calculation: Since no periodic hedging transaction patterns were detected, the transaction authenticity confidence score is equal to the average of the consistency scores. Topological anomaly scoring Mean of each element in the spatiotemporal consistency scoring matrix, and confidence level of transaction authenticity. Perform weighted summation to calculate the comprehensive anomaly score.

[0093] Calculation of the mean of each element in the spatiotemporal consistency scoring matrix: Comprehensive anomaly score calculation: The overall anomaly score was 0.966, which is higher than the preset threshold of 0.6. Therefore, the data for this batch of products was determined to be normal, and no data anomalies were found.

[0094] If the overall anomaly score is below 0.6, the system will initiate a reverse analysis process. A binary search strategy is used to divide the flow path into a first half (node ​​1 → node 2 → node 3) and a second half (node ​​3 → node 4 → node 5). A consistency check is performed on the cumulative code_v3 of node 3, comparing it with the theoretical cumulative code recalculated based on the known collected data from nodes 1, 2, and 3. If the check fails, the anomaly occurs in the first half; if the check passes but the code_v5 check fails, the anomaly occurs in the second half. The division and check process is recursively executed until the specific node where the coding inconsistency first occurs is located. This node is marked as the node corresponding to the time of the anomaly, and the responsible parties for this node and all subsequent nodes are included in the responsibility chain, generating a full-process data penetration and traceability report.

[0095] Throughout the process, data starts with three basic data points: the filling timestamp, batch number, and production line number from the production line. These are combined with sensor sampling values ​​from the first logistics node to generate an initial cumulative code, code_v1. As the product flows to regional warehousing centers, provincial distributors, city-level distributors, and terminal retailers, the sensor data and geographic location information at each node are updated layer by layer through XOR overlay operations, forming an evolutionary sequence from code_v2 to code_v5, allowing the cumulative code to carry the complete flow history. Simultaneously collected transaction records form a transaction-settlement pairing dataset, recording the capital flow information at each stage. Differential analysis restores the cumulative code to a flow feature vector sequence, aligns it with the transaction spatiotemporal feature set on the time axis, calculates geographical deviation, temporal consistency, and state logic consistency, and generates a spatiotemporal consistency scoring matrix. A topology anomaly score is calculated by comparing the actual flow path with the authorized channel topology. A graph neural network model is used to jointly model the logistics flow relationship and the capital flow relationship, identifying the consistency between the logistics direction and the net capital flow. Combined with periodic hedging transaction pattern detection, the confidence level of transaction authenticity is calculated. Finally, the topological anomaly score, spatiotemporal consistency score, and transaction authenticity confidence score are weighted and fused to obtain a comprehensive anomaly score of 0.966, indicating that the data is normal. If an anomaly is detected, reverse parsing and cumulative encoding are used, along with a binary search strategy, to locate the specific link where the anomaly occurred and the responsible party, generating a traceability report. The entire data flow process achieves end-to-end data penetration management from production to sales, ensuring data continuity and traceability.

Claims

1. A method for managing data penetration throughout the entire production and sales process of fast-moving consumer goods, characterized in that, Includes the following steps: Collect the filling timestamp, production batch number, and production line number of the production line. Use a hash algorithm to generate a basic dynamic identification code from the above data. Simultaneously collect the sensor sampling value, sampling timestamp, and node geographical location of the first logistics node. Perform an XOR operation on the node triplet data and the basic dynamic identification code to generate a cumulative code carrying the initial production state. When a product flows to each logistics node and sales node, the current node data is collected and the cumulative code is subjected to a multi-level XOR operation to generate an updated cumulative code. Transaction records are obtained simultaneously to generate a transaction-settlement matching dataset. Differential parsing is performed on the cumulative encoding to calculate the state change rate, time interval, and spatial displacement between adjacent nodes, generating a flow feature vector sequence, and extracting the transaction spatiotemporal feature set from the transaction-settlement pairing dataset; The flow feature vector sequence is aligned with the transaction spatiotemporal feature set on the time axis, and the spatiotemporal consistency verification algorithm is used to detect geographical deviation and logical consistency, generating a spatiotemporal consistency scoring matrix. Input the flow feature vector sequence into the channel topology constraint rule engine, compare the actual flow path with the authorized channel network topology, and generate a topology anomaly score; A graph neural network model is used to jointly model the relationship between logistics flow and capital flow, identify abnormal node pairs, and calculate the confidence level of transaction authenticity. The system uses topological anomaly scores, spatiotemporal consistency scores, and transaction authenticity confidence levels to make joint decisions to determine anomalies, and reverse-analyzes the cumulative coding to locate the responsible party and generate a traceability report.

2. The method for managing data penetration throughout the entire production and sales process of fast-moving consumer goods according to claim 1, characterized in that, The XOR superposition operation includes: concatenating sensor sample values, sampling timestamps, and node geographical locations into a data string according to a fixed format; converting the data string into a binary code of the same length as the basic dynamic identifier code; using zero padding if the data string length is insufficient; and using segmented XOR folding if the data string length exceeds the limit; and performing a bitwise XOR operation between the converted binary code and the basic dynamic identifier code.

3. The method for managing data penetration throughout the entire production and sales process of fast-moving consumer goods according to claim 1, characterized in that, The differential analysis includes: utilizing the reversibility of the XOR operation, performing an XOR operation on the cumulative code of the current node and the binary code of the known collected data of the current node to restore the cumulative code of the previous node, recursively executing the restoration process to extract the original data of each node layer by layer, calculating the difference value of the extracted original data between adjacent layers to form differential features; the state change rate is the ratio of the difference in sensor readings between adjacent nodes to the time interval.

4. The method for data penetration management of the entire production and sales process of fast-moving consumer goods according to claim 1, characterized in that, The spatiotemporal consistency verification algorithm includes: calculating the distance between the geographical location of each circulation node and the transaction location in the corresponding transaction record; when the distance exceeds a geographical threshold, the geographical deviation score of the node is the ratio of the geographical threshold to the distance value; calculating the time difference between the logistics arrival time and the transaction time; when the time difference exceeds a time threshold, the time consistency score of the node is calculated according to the ratio of the time threshold to the time difference; and verifying the logical consistency between the logistics status change trend and the transaction status; when the logistics status shows that the goods are in transit but the transaction record shows that the settlement has been completed, the logical consistency score is zero.

5. The method for data penetration management of the entire production and sales process of fast-moving consumer goods according to claim 4, characterized in that, The geographical threshold is determined based on the node type, with different geographical thresholds set for warehousing nodes, transportation nodes, distributor nodes, and terminal sales nodes; the time threshold is determined based on the standard logistics timeliness between adjacent nodes, with different time thresholds set for nodes within the same city and nodes across provinces.

6. The method for data penetration management of the entire production and sales process of fast-moving consumer goods according to claim 1, characterized in that, The calculation process of the topology anomaly score includes: matching the node type identifier in the flow feature vector sequence with the nodes in the authorized channel network topology to generate the actual flow path; comparing the actual flow path with the compliant flow path library; using the edit distance algorithm to calculate the edit distance between the actual flow path and the most similar compliant path; and calculating the topology anomaly score based on the edit distance. When the edit distance is zero, the score is full marks. For each increase in edit distance, the score decreases by a preset decay coefficient.

7. The method for data penetration management of the entire production and sales process of fast-moving consumer goods according to claim 1, characterized in that, The graph neural network model includes: a node feature encoding layer, which encodes the attribute features of each node into a fixed-dimensional feature vector; an edge feature encoding layer, which encodes the attribute features of each edge into a fixed-dimensional feature vector; a message passing layer, which uses a graph attention mechanism to aggregate the feature information of each node's neighbors and combines it with edge features to weighted update the node embedding vector; a joint modeling layer, which concatenates the embedding vectors of logistics flow edges and capital flow edges to generate a joint embedding representation; and an anomaly detection layer, which calculates the consistency score between the logistics direction and the net capital flow of each node pair based on the joint embedding representation and outputs a set of abnormal node pairs.

8. The method for data penetration management of the entire production and sales process of fast-moving consumer goods according to claim 7, characterized in that, The calculation of transaction authenticity confidence includes: extracting the time series of historical transaction records of abnormal node pairs from the transaction-settlement pairing dataset, extracting the dominant frequency component by performing Fourier transform on the time series, determining the existence of a periodic hedging transaction pattern and reducing the transaction authenticity confidence when the periodic hedging transaction pattern is within a preset suspicious period range; when no periodic hedging transaction pattern is detected, the transaction authenticity confidence is equal to the consistency score, and when a periodic hedging transaction pattern is detected, the transaction authenticity confidence is the consistency score multiplied by the periodic penalty coefficient.

9. The method for data penetration management of the entire production and sales process of fast-moving consumer goods according to claim 1, characterized in that, The reverse parsing cumulative coding for locating the responsible party adopts a binary search strategy: the flow path is divided into a first half and a second half according to the time sequence, and the cumulative coding of the end node of the first half and the beginning node of the second half are checked for consistency. If the cumulative coding of the end node of the first half passes the check but the cumulative coding of the beginning node of the second half fails the check, the anomaly occurs in the second half. The division and verification process is recursively performed on the half where the anomaly is located until the specific anomaly node is located, and the responsible parties of the node and all subsequent nodes are included in the responsible party chain.

10. A data penetration management system for the entire process of fast-moving consumer goods (FMCG) production and sales, used to execute the data penetration management method for the entire process of FMCG production and sales as described in any one of claims 1 to 9, characterized in that, include: The cumulative code generation module is used to collect production line data to generate basic dynamic identification codes, and perform an XOR operation with the data of the first logistics node to generate a cumulative code carrying the initial production state. The data acquisition module is used to collect data at each transfer node and update the cumulative code, and synchronously obtain transaction records to generate a transaction-settlement matching dataset; The feature extraction module is used to perform differential parsing on the cumulative encoding to generate a sequence of flow feature vectors, and to extract the spatiotemporal feature set of transactions from the transaction-settlement pairing dataset; The spatiotemporal consistency verification module is used to align the flow feature vector sequence with the transaction spatiotemporal feature set on the time axis and execute the spatiotemporal consistency verification algorithm to generate a spatiotemporal consistency scoring matrix. The topology anomaly scoring module is used to compare the actual flow path with the authorized channel network topology to generate a topology anomaly score; The joint modeling module is used to jointly model the relationship between logistics flow and capital flow using graph neural network models, identify abnormal node pairs, and calculate the confidence level of transaction authenticity. The traceability report generation module is used to integrate multi-dimensional scores for joint decision-making to identify anomalies and reverse-analyze cumulative codes to locate the responsible party and generate a traceability report.

Images