Big data empowered consumer product quality traceability and risk early warning management and control system

Abstract

The application discloses a big data enabled consumer product quality traceability and risk early warning management and control system, and belongs to the technical field of consumer product quality control. The big data enabled consumer product quality traceability and risk early warning management and control system comprises a multi-source heterogeneous data processing module, a full-chain dynamic traceability module, an intelligent dynamic early warning module and a risk prediction module. The application solves the problem that the prior art cannot avoid quality risks in advance, and can only find problems during spot checks, and the timeliness and foresight of management and control are seriously insufficient. The application realizes granular accurate traceability from raw material procurement to terminal consumption full life cycle, can quickly lock the source node when a quality problem occurs, improves disposal efficiency and accuracy, configures differentiated plans according to early warning levels, optimizes supervision resources, analyzes historical trends by using a time sequence prediction algorithm, predicts potential risks, promotes quality control from passive post-disposal to active pre-prevention, and improves risk control initiative and effectiveness.

Description

Technical Field

[0001] This invention relates to the field of consumer product quality control technology, specifically a consumer product quality traceability and risk early warning control system empowered by big data. Background Technology

[0002] With the rapid development of the consumer goods market, consumers are paying increasing attention to product quality, and traceability and risk warning management of consumer goods quality have become core links in safeguarding market order and consumer rights.

[0003] Chinese Patent Publication No. CN106296230A discloses a quality and safety monitoring system and method for daily consumer goods. The system includes: a business entity identification subsystem, a product subsystem, a sales subsystem, and a quality supervision subsystem. Each subsystem has its own communication module, enabling decentralized connections and forming a communication network, and communicating via broadcast. Each subsystem has its own synchronization module and database. When a database update occurs, the synchronization module broadcasts the updated content to other subsystems. Additionally, each subsystem has an authentication module that authenticates received updates and updates its own database upon successful authentication.

[0004] In actual use, the aforementioned patents can only provide information on sampled products based on existing data, which is a passive monitoring mode of post-inspection. There is no proactive risk warning design, and it is impossible to issue an alarm at the incipient stage of quality problems. As a result, quality risks cannot be avoided in advance, and problems can only be discovered during sampling. The timeliness and foresight of control are seriously insufficient. Therefore, it does not meet the existing needs. In response, we have proposed a big data-enabled consumer product quality traceability and risk warning control system. Summary of the Invention

[0005] The purpose of this invention is to provide a big data-enabled consumer product quality traceability and risk early warning management system. By constructing a full-chain dynamic traceability and risk transmission path identification mechanism, and utilizing industrial internet identifier resolution technology to assign a unique dynamic code to each consumer product, it achieves precise traceability at the granular level throughout the entire lifecycle from raw material procurement to end-consumer use. Combined with dynamic coding, a risk transmission model is built to accurately locate the source of risk and clarify the transmission path. When quality problems occur, the source node can be quickly identified, the diffusion path traced, and the responsibility of each link clarified, improving the efficiency and accuracy of handling. Simultaneously, an intelligent dynamic early warning and hierarchical management mechanism is constructed. A random forest algorithm is used to dynamically calculate risk values ​​to improve assessment accuracy. Differentiated contingency plans are configured according to the early warning level to optimize regulatory resources. Time series prediction algorithms are used to analyze historical trends and predict potential risks, promoting quality control from passive post-event handling to proactive pre-event prevention, improving the initiative and effectiveness of risk management, and solving the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a big data-enabled consumer product quality traceability and risk early warning management system, comprising:

[0007] The multi-source heterogeneous data processing module is used to acquire multi-source consumer product data to form a global data pool. After structured processing and standardization, the data is stored in a blockchain-encrypted distributed manner.

[0008] The full-chain dynamic traceability module is used to dynamically encode consumer products using industrial internet identifiers, build a risk transmission model, and use dynamic coding and risk transmission model to trace the source of consumer products with quality problems;

[0009] The intelligent dynamic early warning module integrates three types of risk indicators: basic, dynamic, and management. It uses the random forest algorithm to dynamically calculate risk values, generate early warning levels, and trigger differentiated contingency plans.

[0010] The risk prediction module is used to analyze historical data trends through time-series prediction algorithms to predict potential risks in advance.

[0011] Preferably, the multi-source heterogeneous data processing module specifically includes:

[0012] Acquire industrial internet identifier resolution data, IoT real-time sensing data, multi-channel feedback data from consumers, regulatory spot check data, and industry standard data to form a comprehensive data pool;

[0013] The acquired structured data undergoes format validation and consistency correction. For unstructured text data, natural language processing algorithms are used for word segmentation, entity recognition, keyword extraction, and semantic normalization to convert it into structured data.

[0014] By using dynamic data mapping technology, a field mapping relationship library for different data sources is established, and the corresponding mapping rules are automatically matched according to the data source type to perform unified and standardized processing on data of different formats.

[0015] After processing, standardized data is distributed and encrypted using blockchain technology.

[0016] Preferably, the standardization process for data in different formats specifically includes:

[0017] Construct a cross-domain data element model to define the core data elements, data types, data lengths, and value ranges of data from different domains;

[0018] Establish a dynamic mapping rule base and formulate mapping rules between data elements for heterogeneous data formats of different enterprises and departments;

[0019] A dynamic mapping engine is used to automatically map and transform heterogeneous data according to the rules in the mapping rule base, converting data of different formats into standardized data that conforms to the cross-domain data meta-model.

[0020] Preferably, the construction process of the risk transmission model specifically includes:

[0021] Acquire full-chain data from the global data pool, and construct a node relationship graph of the industrial chain with each link of the industrial chain as a node and the data relationship between each link as an edge;

[0022] The graph neural network algorithm is used to train the node association graph of the industrial chain, explore the potential association between nodes, and determine the risk transmission weight of each node.

[0023] When a quality issue is reported, the dynamic coding information of the problematic consumer product is extracted to locate the corresponding core node.

[0024] Based on the trained risk transmission model, the upstream related nodes of the core node are traced back to determine the risk source node;

[0025] The system traces downstream related nodes of the core node in a forward direction, sorts out the entire path of risk spread, and outputs information on the risk source node, risk spread path information, and the responsible entity information for each node.

[0026] Preferably, the intelligent dynamic early warning module specifically includes:

[0027] Obtain basic risk indicators, dynamic risk indicators, and management risk indicators, and integrate these indicators.

[0028] The random forest algorithm is used to dynamically calculate the risk value of the fused risk indicators. Based on the calculated dynamic risk value, an early warning level signal is automatically generated, and a differentiated response plan is triggered for different early warning levels.

[0029] Preferably, the process of dynamically calculating the risk value specifically includes:

[0030] Define the basic risk indicators, dynamic risk indicators, and management risk indicators, including their definitions, data sources, and quantification values;

[0031] The subjective weights of basic risk indicators, dynamic risk indicators, and management risk indicators are determined using the analytic hierarchy process (AHP).

[0032] The objective weights of each indicator are calculated based on historical data using the entropy weight method. The subjective weights and objective weights are then weighted and fused together to obtain the comprehensive weights of each indicator.

[0033] A risk value calculation model is constructed based on the random forest algorithm. The quantitative values ​​of each indicator and their corresponding comprehensive weights are input into the model to calculate the real-time risk value of consumer products.

[0034] Preferably, the step of automatically generating early warning level signals based on the calculated dynamic risk value and triggering differentiated response plans for different early warning levels specifically includes:

[0035] Preset risk value threshold range, determine the warning level based on the threshold range where the risk value is located, and trigger the corresponding response plan for different warning levels;

[0036] By collecting and processing sampling data, complaint data, and public opinion data, the risk value is recalculated to determine whether the risk has been effectively controlled.

[0037] If the risk value does not drop below the corresponding warning level threshold, the emergency response plan will be adjusted and the response measures will be strengthened.

[0038] Preferably, the process of making advance predictions of potential risks specifically includes:

[0039] Collect historical data from the global data pool, clean, deduplicate, and standardize the historical data, and filter out feature data related to potential risks;

[0040] A potential risk prediction logic is constructed using a time series prediction algorithm. The processed historical feature data is divided into training data and validation data according to the time series. The prediction logic parameters are optimized using the training data, and the effectiveness of the prediction logic is verified using the validation data.

[0041] The real-time collected dynamic data is input into the optimized time series prediction logic to predict the trend of risk value changes over a period of time and identify potential risk types.

[0042] When potential risks are predicted, forward-looking early warning information is sent to regulatory authorities and enterprises, and specific prevention and control recommendations are provided for potential risks.

[0043] Preferably, after acquiring multi-source data on consumer products, the following are also included:

[0044] The multi-source data is standardized, and the standardized multi-source data is then sliced ​​into time series according to the data source and business process to form an initial time series set.

[0045] A causal relationship discovery algorithm is applied to construct a correlation network among the sequences in the initial time series set, where nodes represent sequences and edges represent statistically significant guiding or influencing relationships.

[0046] Based on the association network, a collaborative quality check is performed on the associated sequence groups in the initial time series set to identify low-quality sequences and output a high-quality sequence dataset.

[0047] Mining key risk indicators and dynamic risk patterns from high-quality sequence datasets, and retrieving historical risk events based on key risk indicators;

[0048] Historical risk events are marked as key points on the timeline, and all sequence data fragments within the time window before and after the event are extracted as analysis samples.

[0049] Using time-series pattern mining algorithms and regression analysis, we identify frequently occurring risk indicator combination change patterns that are strongly correlated with risk events from the analyzed samples, which serve as dynamic risk patterns.

[0050] Calculate the fitting coefficient between each individual sequence element in the high-quality sequence dataset and the dynamic risk pattern, and select the set of key sequence elements based on the fitting coefficient.

[0051] For each element in the set of key sequence elements, the statistical control limits are calculated on a rolling basis based on its historical performance data, and the specification limits are set in combination with business rules.

[0052] Real-time detection of the sequence value of key elements; generation of graded early warning signals based on the comparison results between the sequence value and the specification limit of the key element; and generation of early abnormality prompts based on the situation of reaching the control limit.

[0053] Based on early anomaly alerts and historical trends of key elements, a time-series prediction algorithm is used to comprehensively predict potential risks.

[0054] Preferably, before acquiring multi-source consumer product data to form a comprehensive data pool, and after structured processing and standardization, and before employing blockchain-based encrypted distributed storage, the following steps are also included:

[0055] Determine the historical production batches and outgoing defect parameters of consumer products, and identify multiple potential abnormal quality events of consumer products based on the historical production batches and outgoing defect parameters;

[0056] Obtain the influence propagation intensity of each potential abnormal quality event at each node in the industry chain, and determine the risk observation value of each potential abnormal quality event based on the influence propagation intensity of each node;

[0057] The developmental impact factors for each potential abnormal quality event are determined based on risk observations.

[0058] Obtain abnormal observation data for each potential abnormal quality event from multi-source consumer product data, and obtain user evaluation data for key abnormal data;

[0059] Determine the user's confidence and support for the abnormal key data based on the evaluation data;

[0060] The uncertainty factor of the source parameters of multi-source consumer product data is calculated based on users' confidence and support for the key abnormal data and the development impact factors of each potential abnormal event:

[0061]

[0062] Where Q represents the uncertainty factor of the source parameter of the multi-source data for consumer goods, N represents the number of potential abnormal quality events, and i represents the i-th potential abnormal quality event. Let be the development impact factor of the i-th potential abnormal quality event. This represents the confidence level related weight, with a value of 0.6. This represents the user's confidence level regarding the critical data related to potential abnormal quality events (i items). This represents the support-related weight, with a value of 0.4. Let represent the user's support for the key data of the i-th potentially abnormal quality event, ln represent the natural logarithm, k represent the baseline fluctuation factor of the consumer goods supply chain, and S represent the steady-state connectivity index of the consumer goods supply chain. This refers to the controllable factors in the development of consumer product quality incidents;

[0063] The uncertainty factor of the source parameter of the multi-source data of consumer products is compared with a preset threshold. If the uncertainty factor is higher than the preset threshold, an enhanced data collection instruction is generated, and the specific data source that needs to be collected is determined based on the enhanced data collection instruction.

[0064] Collect relevant parameters from a specific data source and insert them into multi-source data.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] This invention achieves precise traceability and clear identification of risk transmission paths throughout the entire lifecycle of consumer products by constructing a full-chain dynamic traceability and risk transmission path identification mechanism. Based on industrial internet identifier resolution technology, each consumer product is assigned a unique dynamic code, enabling precise traceability at the granular level from raw material procurement to end consumption. By constructing a risk transmission model combined with dynamic coding, it effectively solves the problems of inaccurate risk source positioning and unclear risk transmission path identification. When quality problems occur, it can quickly pinpoint the risk source node, trace the complete path of risk spread in a forward direction, clarify the responsible parties at each stage, and significantly improve the efficiency and accuracy of quality problem handling. By constructing an intelligent dynamic early warning and hierarchical control mechanism, it achieves dynamic risk assessment, hierarchical control, and forward-looking prediction. By using the random forest algorithm to dynamically calculate risk values, it significantly improves the accuracy of risk assessment. Differentiated handling plans are configured for different early warning levels to achieve precise allocation of regulatory resources. By using time-series prediction algorithms to analyze historical data change trends, it focuses on predicting potential risks, realizing the transformation of quality risk management from post-event handling to pre-event prevention, effectively improving the initiative and effectiveness of risk management. Attached Figure Description

[0067] Figure 1 Module diagram of the consumer product quality traceability and risk early warning and control system empowered by big data for the present invention;

[0068] Figure 2 Flowchart of the consumer product quality traceability and risk early warning control system empowered by big data for this invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] To address the issue that existing technologies cannot proactively mitigate quality risks and can only detect problems during random inspections, resulting in a severe lack of timeliness and foresight in control, please refer to [link to relevant documentation]. Figures 1-2 This embodiment provides the following technical solution:

[0071] A big data-enabled consumer product quality traceability and risk early warning management system includes:

[0072] The multi-source heterogeneous data processing module is used to acquire multi-source consumer product data to form a global data pool. After structured processing and standardization, the data is stored in a blockchain-encrypted distributed manner.

[0073] The full-chain dynamic traceability module is used to dynamically encode consumer products using industrial internet identifiers, construct a risk transmission model, and trace the source of consumer products with quality problems using dynamic coding and the risk transmission model. Based on industrial internet identifier resolution technology, each consumer product is assigned a unique dynamic code. The graph neural network algorithm is used to mine the correlation of data at each node of the industrial chain to construct a risk transmission model. When a quality problem occurs, the dynamic coding is used to locate the full life cycle data of the risky consumer product. The risk transmission model is used to quickly locate the risk source node and trace the entire path of risk spread in a forward direction, clarifying the responsible parties at each stage.

[0074] The intelligent dynamic early warning module integrates three types of risk indicators: basic, dynamic, and management. It uses the random forest algorithm to dynamically calculate risk values, generate early warning levels, and trigger differentiated contingency plans.

[0075] The risk prediction module is used to analyze historical data trends through time-series prediction algorithms to predict potential risks in advance.

[0076] The multi-source heterogeneous data processing module specifically includes:

[0077] The system acquires industrial internet identifier resolution data, IoT real-time sensing data, multi-channel consumer feedback data, regulatory inspection data, and industry standard data to form a comprehensive data pool. Among these, IoT real-time sensing data includes cold chain temperature and humidity data, warehousing environment data, and transportation trajectory data; multi-channel consumer feedback data includes e-commerce evaluation data, 12315 complaint data, and public opinion data.

[0078] The acquired structured data undergoes format validation and consistency correction. For unstructured text data, natural language processing algorithms are used for word segmentation, entity recognition, keyword extraction, and semantic normalization to convert it into structured data.

[0079] By using dynamic data mapping technology, a field mapping relationship library for different data sources is established, and the corresponding mapping rules are automatically matched according to the data source type to perform unified and standardized processing on data of different formats.

[0080] Blockchain encryption technology is used for distributed storage of standardized data, enabling traceability and tamper-proofing throughout the entire data collection, transmission, and usage process. Data anonymization algorithms are used to process enterprise trade secrets and consumer privacy data, ensuring data security. Unstructured data cleaning specifically includes: preprocessing the collected unstructured text data, including removing special characters and stop words, and performing text segmentation and part-of-speech tagging.

[0081] Based on natural language processing technology, a text classification model is constructed to classify preprocessed text data. The classification categories include consumer product quality complaints, consumer product functional failures, service quality feedback, positive evaluations, and neutral evaluations.

[0082] Construct a named entity recognition model to extract key entity information from classified text data, including product name, product model, fault location, fault phenomenon, complaint time, complainant's region, and company name involved.

[0083] The extracted key entity information is structured according to a preset data structure to generate standardized structured data, which is then stored in the global data pool.

[0084] Standardize and unify data of different formats, specifically including:

[0085] Construct a cross-domain data element model to define the core data elements, data types, data lengths, and value ranges of data in different domains, including production, circulation, consumption, and regulation.

[0086] Establish a dynamic mapping rule base to formulate mapping rules between data elements for heterogeneous data formats of different enterprises and departments; the mapping rules include one-to-one mapping rules, one-to-many mapping rules, and many-to-one mapping rules.

[0087] A dynamic mapping engine is used to automatically map and transform heterogeneous data according to the rules in the mapping rule base, converting data of different formats into standardized data that conforms to the cross-domain data meta-model.

[0088] The process of constructing a risk transmission model specifically includes:

[0089] Acquire full-chain data from the global data pool, including raw material procurement data, production process data, warehousing data, transportation data, sales data, consumer feedback data, and regulatory data. Construct a chain node relationship graph with each link in the chain as a node and the data relationships between each link as edges. The nodes include raw material supplier nodes, production workshop nodes, warehousing nodes, transportation nodes, sales terminal nodes, consumer nodes, and regulatory nodes.

[0090] The graph neural network algorithm is used to train the node association graph of the industrial chain, explore the potential association between nodes, and determine the risk transmission weight of each node. The risk transmission weight is determined based on the closeness of business association between nodes, the frequency of data interaction, and historical risk transmission records.

[0091] When a quality issue is reported, the dynamic coding information of the problematic consumer product is extracted to locate the corresponding core node.

[0092] Based on the trained risk transmission model, the upstream related nodes of the core node are traced back to determine the risk source node;

[0093] By tracing the downstream related nodes of the core node in a forward direction, the entire path of risk spread is sorted out, and information on the risk source node, risk spread path information, and the responsible entity information of each node is output, providing a basis for subsequent handling.

[0094] The intelligent dynamic early warning module specifically includes:

[0095] Acquire basic risk indicators, dynamic risk indicators, and management risk indicators, and integrate them. Basic risk indicators include inherent attribute indicators such as consumer goods business format, raw material characteristics, and production process complexity. Dynamic risk indicators include variable indicators such as real-time monitoring data, complaint density, public opinion heat value, and sampling failure rate. Management risk indicators include management-related indicators such as enterprise self-inspection and rectification rate, historical violation records, and quality system certification status.

[0096] The random forest algorithm is used to dynamically calculate the risk value of the fused risk indicators. Based on the calculated dynamic risk value, an early warning level signal is automatically generated, and a differentiated response plan is triggered for different early warning levels.

[0097] The process of dynamically calculating the risk value specifically includes:

[0098] Define the basic risk indicators, dynamic risk indicators, and management risk indicators, including their definitions, data sources, and quantification values. The basic risk indicators are quantified using a combination of qualitative and quantitative methods, the dynamic risk indicators are quantified using real-time data, and the management risk indicators are quantified using a combination of historical and real-time data.

[0099] The subjective weights of basic risk indicators, dynamic risk indicators, and management risk indicators are determined using the analytic hierarchy process (AHP).

[0100] The objective weights of each indicator are calculated based on historical data using the entropy weight method. The subjective weights and objective weights are then weighted and fused together to obtain the comprehensive weights of each indicator.

[0101] A risk value calculation model is constructed based on the random forest algorithm. The quantitative values ​​of each indicator and their corresponding comprehensive weights are input into the model to calculate the real-time risk value of consumer products. The random forest algorithm predicts the risk value through multiple decision trees and takes the average of the prediction results of multiple decision trees as the final risk value.

[0102] The system automatically generates early warning level signals based on the calculated dynamic risk values ​​and triggers differentiated response plans for different early warning levels, including:

[0103] A preset risk threshold range is established, and the warning level is determined based on the threshold range in which the risk value falls. Specifically, a risk value ≥ 85 is a red warning, 70 ≤ risk value < 85 is a yellow warning, 50 ≤ risk value < 70 is a blue warning, and a risk value < 50 is a green warning. Corresponding contingency plans are triggered for different warning levels: a red warning activates a cross-departmental joint enforcement mechanism, conducts a comprehensive inspection of the involved company, recalls the same batch of consumer products, and suspends the market sales of the involved consumer products; a yellow warning strengthens spot checks at key points, increases the frequency and sample size of spot checks, and requires companies to conduct self-inspection and submit rectification reports; a blue warning sends risk alerts to companies, requiring them to strengthen control over key aspects and report control status regularly; and a green warning implements annual random inspections and reduces the frequency of routine supervision.

[0104] By collecting and processing sampling data, complaint data, and public opinion data, the risk value is recalculated to determine whether the risk has been effectively controlled.

[0105] If the risk value does not drop below the corresponding warning level threshold, the emergency response plan will be adjusted and the response measures will be strengthened.

[0106] The process of anticipating potential risks in advance includes:

[0107] Collect historical data from the entire data pool, including historical risk value data, historical quality problem data, historical raw material quality data, historical production process data, and historical consumer feedback data; clean, deduplicate, and standardize the historical data, and filter out characteristic data related to potential risks, with a focus on filtering risk characteristic data;

[0108] A potential risk prediction logic is constructed using a time series prediction algorithm. The processed historical feature data is divided into training data and validation data according to the time series. The prediction logic parameters are optimized using the training data, and the effectiveness of the prediction logic is verified using the validation data.

[0109] The real-time collected dynamic data is input into the optimized time series prediction logic to predict the trend of risk value changes over a period of time and identify potential risk types, including raw material quality fluctuation risk, production process deviation risk, and potential risks.

[0110] When potential risks are predicted, forward-looking early warning information is sent to regulatory authorities and enterprises, and specific prevention and control recommendations are provided for potential risks.

[0111] Working principle: When using the big data-enabled consumer product quality traceability and risk early warning control system of this invention, according to... Figure 1 and Figure 2 This includes the following steps:

[0112] S1: Acquire multi-source consumer product data to form a global data pool, and after structured processing and standardization, use blockchain for encrypted distributed storage;

[0113] S2: Based on industrial internet identifier resolution technology, each consumer product is assigned a unique dynamic code, and the code information is updated in real time during warehousing, handling and transportation.

[0114] S3: When quality problems occur, a risk transmission model constructed through graph neural network algorithms is used to locate the entire life cycle data of risky consumer products, quickly identify the source of risk and its spread path, and clarify the responsible party.

[0115] S4: Dynamically calculate the risk value using the random forest algorithm, generate early warning level signals based on the risk value, and trigger differentiated response plans for different early warning levels;

[0116] S5: Analyze historical data trends through time-series prediction algorithms, predict risk value change trends, identify potential risk types, and push forward-looking early warnings and prevention and control suggestions to regulatory authorities and enterprises.

[0117] In one embodiment, after obtaining multi-source data on consumer products, the method further includes:

[0118] The multi-source data is standardized, and the standardized multi-source data is then sliced ​​into time series according to the data source and business process to form an initial time series set.

[0119] A causal relationship discovery algorithm is applied to construct a correlation network among the sequences in the initial time series set, where nodes represent sequences and edges represent statistically significant guiding or influencing relationships.

[0120] Based on the association network, a collaborative quality check is performed on the associated sequence groups in the initial time series set to identify low-quality sequences and output a high-quality sequence dataset.

[0121] Mining key risk indicators and dynamic risk patterns from high-quality sequence datasets, and retrieving historical risk events based on key risk indicators;

[0122] Historical risk events are marked as key points on the timeline, and all sequence data fragments within the time window before and after the event are extracted as analysis samples.

[0123] Using time-series pattern mining algorithms and regression analysis, we identify frequently occurring risk indicator combination change patterns that are strongly correlated with risk events from the analyzed samples, which serve as dynamic risk patterns.

[0124] Calculate the fitting coefficient between each individual sequence element in the high-quality sequence dataset and the dynamic risk pattern, and select the set of key sequence elements based on the fitting coefficient.

[0125] For each element in the set of key sequence elements, the statistical control limits are calculated on a rolling basis based on its historical performance data, and the specification limits are set in combination with business rules.

[0126] Real-time detection of the sequence value of key elements; generation of graded early warning signals based on the comparison results between the sequence value and the specification limit of the key element; and generation of early abnormality prompts based on the situation of reaching the control limit.

[0127] Based on early anomaly alerts and historical trends of key elements, a time-series prediction algorithm is used to comprehensively predict potential risks.

[0128] In this embodiment, time-series slicing refers to cutting standardized data into continuous time-series segments according to fixed time intervals or based on key events in business processes.

[0129] In this embodiment, the causal relationship discovery algorithm adopts the Granger causality test. At a given significance level (e.g., p<0.05), if the historical value of time series X can significantly predict the current value of time series Y, then it is considered that there is a guiding relationship from X to Y, and a directed edge is established in the association network.

[0130] In this embodiment, the co-quality check refers to checking whether the dynamic changes of a causally related sequence group are temporally coordinated. Specifically, the method involves calculating the dynamic time-normalized distance or cross-correlation coefficient between the dominant and induced sequences within a sliding window; if this distance exceeds a threshold or the correlation coefficient is below a threshold, the induced sequence is determined to be a low-quality sequence because its change pattern does not conform to causal expectations.

[0131] In this embodiment, a low-quality sequence is defined as a sequence that fails the collaborative quality check, or a sequence with a missing rate exceeding 10%, excessively long consecutive constant values, or zero variance. The system marks the low-quality sequences and triggers a data review process. After interpolation, smoothing, or removal, the processed sequences are combined with the original sequences that passed the check to form a high-quality sequence dataset.

[0132] In this embodiment, a time-series pattern mining algorithm and regression analysis are used to determine dynamic risk patterns, including:

[0133] Feature construction: Extract statistical features (mean, variance, slope) and morphological features (peak value, trough value) from each sequence segment.

[0134] Pattern recognition: Using association rule mining or sequence pattern mining, we identify frequently co-occurring feature combinations before the occurrence of risk events, which are denoted as candidate risk patterns.

[0135] Pattern selection: Using logistic regression or Cox proportional hazards regression models, analyze the statistical correlation strength between each candidate risk pattern and the occurrence of risk events, and select strongly correlated patterns as the final dynamic risk patterns.

[0136] In this embodiment, the fitting coefficient is obtained by calculating the inverse of the dynamic time-normalized distance between each individual sequence element and the dynamic risk pattern sequence, or the absolute value of the Pearson correlation coefficient. The higher the fitting coefficient, the more similar the sequence element is to the morphology of the risk pattern.

[0137] In this embodiment, the statistical control limits are represented by an exponentially weighted moving average control chart method based on the historical performance data of each element, using the data of the most recent N periods (e.g., N=30) to calculate the center line and upper and lower control limits (e.g., ±3 times the standard deviation).

[0138] In this embodiment, the specification limits refer to the upper and lower limits of permissible fluctuations set for the element according to national mandatory standards, industry standards, or internal quality control documents of the enterprise, which are used to determine whether the product / process is qualified.

[0139] In this embodiment, an early anomaly alert is triggered when the real-time sequence value of any key sequence element exceeds its statistical control limit but is still within the specification limit.

[0140] In this embodiment, based on comprehensive prediction, a gradient boosting tree model or a long short-term memory network model is used as input variables, taking early anomaly warning signals, historical trend features of key elements, and other contextual features as input variables, to output the probability of potential risk events occurring within a future period. When the probability exceeds a preset threshold, a potential risk warning is generated.

[0141] The beneficial effects of the above technical solution are as follows: By setting control limits for key indicators, statistical anomalies in the production or supply chain process can be detected before quality problems cause substantial non-conformities, issuing early anomaly alerts and improving practicality. A causal relationship discovery algorithm constructs a network of correlations between indicators. When an indicator issues an early anomaly alert, the system can immediately locate other causally related indicators for collaborative quality inspection. This not only verifies the authenticity of the anomaly but also helps to pinpoint the possible transmission path of the anomaly, greatly improving the interpretability of alerts and the efficiency of root cause analysis. It achieves a shift in risk perception from macro-level lag to micro-level proactiveness, significantly improving the initiative, accuracy, and intelligence of risk management.

[0142] In one embodiment, before acquiring multi-source consumer product data to form a global data pool, and after structured processing and standardization, and before employing blockchain-based encrypted distributed storage, the following steps are also included:

[0143] Determine the historical production batches and outgoing defect parameters of consumer products, and identify multiple potential abnormal quality events of consumer products based on the historical production batches and outgoing defect parameters;

[0144] Obtain the influence propagation intensity of each potential abnormal quality event at each node in the industry chain, and determine the risk observation value of each potential abnormal quality event based on the influence propagation intensity of each node;

[0145] The developmental impact factors for each potential abnormal quality event are determined based on risk observations.

[0146] Obtain abnormal observation data for each potential abnormal quality event from multi-source consumer product data, and obtain user evaluation data for key abnormal data;

[0147] Determine the user's confidence and support for the abnormal key data based on the evaluation data;

[0148] The uncertainty factor of the source parameters of multi-source consumer product data is calculated based on users' confidence and support for the key abnormal data and the development impact factors of each potential abnormal event:

[0149]

[0150] Where Q represents the uncertainty factor of the source parameter of the multi-source data for consumer goods, N represents the number of potential abnormal quality events, and i represents the i-th potential abnormal quality event. Let be the development impact factor of the i-th potential abnormal quality event. This represents the confidence level related weight, with a value of 0.6. This represents the user's confidence level regarding the critical data related to potential abnormal quality events (i items). This represents the support-related weight, with a value of 0.4. Let represent the user's support for the key data of the i-th potentially abnormal quality event, ln represent the natural logarithm, k represent the baseline fluctuation factor of the consumer goods supply chain, and S represent the steady-state connectivity index of the consumer goods supply chain. This refers to the controllable factors in the development of consumer product quality incidents;

[0151] The uncertainty factor of the source parameter of the multi-source data of consumer products is compared with a preset threshold. If the uncertainty factor is higher than the preset threshold, an enhanced data collection instruction is generated, and the specific data source that needs to be collected is determined based on the enhanced data collection instruction.

[0152] Collect relevant parameters from a specific data source and insert them into multi-source data.

[0153] In this embodiment, a potential abnormal quality event refers to an event that may pose a risk to the quality of consumer products and is identified through any one or a combination of the following methods: (a) batches exceeding control limits identified by applying statistical process control rules to historical production batch data; (b) strong association combinations between specific raw material batches and specific factory defects discovered based on association rule mining; and (c) events corresponding to abnormal samples identified by using unsupervised learning of factory inspection parameters using isolated forest or a class of support vector machine algorithms.

[0154] In this embodiment, the propagation intensity of the influence of the industry chain node is represented by taking the occurrence node of the potential abnormal quality event as the source, and using the random walk algorithm or network contagion model based on the industry chain node association graph to simulate and calculate the probability that the influence of the event will reach other nodes in the graph within a preset number of steps. The probability value is the propagation intensity of the event to the corresponding node.

[0155] In this embodiment, the risk observation is obtained by multiplying the propagation intensity by the inherent risk vulnerability coefficient of each node (set according to the historical quality problem frequency of each node).

[0156] In this embodiment, the development impact factor is used to characterize the relative importance of the i-th potential anomalous quality event. It is calculated by taking a weighted average or maximizing the risk observations of the event across all nodes, and then normalizing the result so that the sum of all events equals 1.

[0157] In this embodiment, critical abnormal data refers to monitoring or detection data extracted from the global data pool that is directly related to each potential abnormal quality event, such as the content of a specific component in the Nth batch of raw materials or the value of a specific process parameter during the production process.

[0158] In this embodiment, the evaluation data is represented by the assessment conclusions issued by the domain expert system or the quality audit rule engine for the abnormal key data.

[0159] In this embodiment, the confidence level is represented by the probability value output by a classification model (such as logistic regression) trained based on historical data, which determines that the current abnormal key data belongs to a true anomaly.

[0160] In this embodiment, support is represented by other co-indicator data associated with each anomalous key data point, which is also the proportion of anomalous data points.

[0161] In this embodiment, the baseline fluctuation factor is represented by the coefficient of variation of the monthly average factory pass rate of various consumer products over the past year. It is used to measure the baseline fluctuation level of industry quality and is directly calculated from historical statistical databases.

[0162] In this embodiment, the steady-state connectivity index is represented as the geometric mean of the data reporting timeliness rates among all critical path nodes, calculated based on the industry chain node association diagram, and is used to measure the stability of the industry chain data flow.

[0163] In this embodiment, the development controllable factor is a coefficient calculated based on the company's internal audit score of the quality system and the rectification closure rate in the previous quarter, which is used to measure the company's ability to control the development of quality incidents.

[0164] The beneficial effects of the above technical solution are as follows: by calculating the uncertainty factor of the source parameters, the uncertainty at the data level of different consumer products and different batches is quantified. Based on the uncertainty factor of the source parameters, data collection items can be dynamically planned (e.g., adding more testing items for high-risk batches, increasing the sensor sampling frequency). This achieves adaptive allocation of data collection resources, meets the needs of risk management with minimal data cost, realizes the optimal configuration of data resources under the risk management objective, and ensures the efficiency and effectiveness of the overall quality traceability and early warning system from the starting point of the data supply chain.

[0165] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0166] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A big data-enabled method for consumer product quality traceability and risk early warning, characterized in that: include: The multi-source heterogeneous data processing module acquires multi-source consumer product data to form a global data pool. After structured processing and standardization, the data is stored in a blockchain-encrypted distributed manner. The full-chain dynamic traceability module uses industrial internet identifiers to dynamically encode consumer products, builds a risk transmission model, and uses dynamic coding and risk transmission model to trace the source of consumer products with quality problems; The intelligent dynamic early warning module integrates three types of risk indicators: basic, dynamic, and management. It uses the random forest algorithm to dynamically calculate risk values, generate early warning levels, and trigger differentiated contingency plans. The risk prediction module analyzes historical data trends using time-series prediction algorithms to predict potential risks in advance. Among them, the statistical control limits are represented by the exponentially weighted moving average control chart method based on the historical performance data of each element, and the center line and upper and lower control limits are calculated on a rolling basis using the data of the most recent N periods. The construction process of the risk transmission model specifically includes: acquiring full-chain data from the global data pool, constructing a chain node association graph with each link in the chain as a node and the data association between each link as an edge; training the chain node association graph based on a graph neural network algorithm to mine the potential association relationships between each node and determine the risk transmission weight of each node; when receiving feedback on quality issues, extracting the dynamic coding information of the problematic consumer products to locate the corresponding core node; based on the trained risk transmission model, tracing back the upstream associated nodes of the core node to determine the risk source node; tracing forward the downstream associated nodes of the core node to sort out the entire path of risk diffusion, and outputting information on the risk source node, risk diffusion path, and the responsible entity information corresponding to each node; After acquiring multi-source data on consumer goods, the process also includes: standardizing the multi-source data; slicing the standardized multi-source data into time series segments according to data source and business process to form an initial time series set; applying a causal relationship discovery algorithm to construct a correlation network among the sequences in the initial time series set, where nodes represent sequences and edges represent statistically significant guiding or influencing relationships; based on the correlation network, performing collaborative quality checks on correlated sequence groups in the initial time series set to identify low-quality sequences and output a high-quality sequence dataset; mining key risk indicators and dynamic risk patterns from the high-quality sequence dataset; retrieving historical risk events based on the key risk indicators; marking historical risk events as key points on the timeline and extracting all sequence data slices within the time window before and after the event. The data is analyzed using time-series pattern mining algorithms and regression analysis to identify frequently occurring risk indicator combination change patterns that are strongly correlated with risk events, which are then used as dynamic risk patterns. The fitting coefficients of each individual sequence element in the high-quality sequence dataset with the dynamic risk patterns are calculated, and key sequence element sets are selected based on these coefficients. For each element in the key sequence element set, statistical control limits are calculated on a rolling basis based on its historical performance data, and specification limits are set in conjunction with business rules. The sequence values ​​of key elements are monitored in real time, and graded early warning signals are generated based on the comparison results between the sequence values ​​and the specification limits of the key element. Early anomaly alerts are generated based on the occurrence of control limits. Based on early anomaly alerts and the historical trends of key elements, a time-series prediction algorithm is used to comprehensively predict potential risks.

2. The method for consumer product quality traceability and risk early warning empowered by big data according to claim 1, characterized in that, The multi-source heterogeneous data processing module specifically includes: Acquire industrial internet identifier resolution data, IoT real-time sensing data, multi-channel feedback data from consumers, regulatory spot check data, and industry standard data to form a comprehensive data pool; The acquired structured data undergoes format validation and consistency correction. For unstructured text data, natural language processing algorithms are used for word segmentation, entity recognition, keyword extraction, and semantic normalization to convert it into structured data. By using dynamic data mapping technology, a field mapping relationship library for different data sources is established, and the corresponding mapping rules are automatically matched according to the data source type to perform unified and standardized processing on data of different formats. After processing, standardized data is distributed and encrypted using blockchain technology.

3. The method for consumer product quality traceability and risk early warning empowered by big data according to claim 2, characterized in that, The standardization process for data in different formats specifically includes: Construct a cross-domain data element model to define the core data elements, data types, data lengths, and value ranges of data from different domains; Establish a dynamic mapping rule base and formulate mapping rules between data elements for heterogeneous data formats of different enterprises and departments; A dynamic mapping engine is used to automatically map and transform heterogeneous data according to the rules in the mapping rule base, converting data of different formats into standardized data that conforms to the cross-domain data meta-model.

4. The method for consumer product quality traceability and risk early warning empowered by big data according to claim 1, characterized in that, The intelligent dynamic early warning module specifically includes: Obtain basic risk indicators, dynamic risk indicators, and management risk indicators, and integrate these indicators. The random forest algorithm is used to dynamically calculate the risk value of the fused risk indicators. Based on the calculated dynamic risk value, an early warning level signal is automatically generated, and a differentiated response plan is triggered for different early warning levels.

5. The big data-enabled consumer product quality traceability and risk warning method according to claim 4, characterized in that, The process of dynamically calculating the risk value specifically includes: Define the basic risk indicators, dynamic risk indicators, and management risk indicators, including their definitions, data sources, and quantification values; The subjective weights of basic risk indicators, dynamic risk indicators, and management risk indicators are determined using the analytic hierarchy process (AHP). The objective weights of each indicator are calculated based on historical data using the entropy weight method. The subjective weights and objective weights are then weighted and fused together to obtain the comprehensive weights of each indicator. A risk value calculation model is constructed based on the random forest algorithm. The quantitative values ​​of each indicator and their corresponding comprehensive weights are input into the model to calculate the real-time risk value of consumer products.

6. The method for consumer product quality traceability and risk early warning empowered by big data according to claim 4, characterized in that, The automatic generation of early warning level signals based on the calculated dynamic risk value, and the triggering of differentiated response plans for different early warning levels, specifically includes: Preset risk value threshold range, determine the warning level based on the threshold range where the risk value is located, and trigger the corresponding response plan for different warning levels; By collecting and processing sampling data, complaint data, and public opinion data, the risk value is recalculated to determine whether the risk has been effectively controlled. If the risk value does not drop below the corresponding warning level threshold, the emergency response plan will be adjusted and the response measures will be strengthened.

7. The method for consumer product quality traceability and risk early warning empowered by big data according to claim 1, characterized in that, The process of making advance predictions of potential risks specifically includes: Collect historical data from the global data pool, clean, deduplicate, and standardize the historical data, and filter out feature data related to potential risks; A potential risk prediction logic is constructed using a time series prediction algorithm. The processed historical feature data is divided into training data and validation data according to the time series. The prediction logic parameters are optimized using the training data, and the effectiveness of the prediction logic is verified using the validation data. The real-time collected dynamic data is input into the optimized time series prediction logic to predict the trend of risk value changes over a period of time and identify potential risk types. When potential risks are predicted, forward-looking early warning information is sent to regulatory authorities and enterprises, and specific prevention and control recommendations are provided for potential risks.

8. The method for consumer product quality traceability and risk early warning empowered by big data according to claim 1, characterized in that, Before acquiring multi-source consumer product data to form a comprehensive data pool, and after structured processing and standardization, and before employing blockchain-based encrypted distributed storage, the process also includes: Determine the historical production batches and outgoing defect parameters of consumer products, and identify multiple potential abnormal quality events of consumer products based on the historical production batches and outgoing defect parameters; Obtain the influence propagation intensity of each potential abnormal quality event at each node in the industry chain, and determine the risk observation value of each potential abnormal quality event based on the influence propagation intensity of each node; The developmental impact factors for each potential abnormal quality event are determined based on risk observations. Obtain abnormal observation data for each potential abnormal quality event from multi-source consumer product data, and obtain user evaluation data for key abnormal data; Determine the user's confidence and support for the abnormal key data based on the evaluation data; The uncertainty factor of the source parameters of multi-source consumer product data is calculated based on users' confidence and support for the key abnormal data and the development impact factors of each potential abnormal event: Where Q represents the uncertainty factor of the source parameter of the multi-source data for consumer goods, N represents the number of potential abnormal quality events, and i represents the i-th potential abnormal quality event. Let be the development impact factor of the i-th potential abnormal quality event. This represents the confidence level related weight, with a value of 0.

6. This represents the user's confidence level regarding the critical data related to potential abnormal quality events (i items). This represents the support-related weight, with a value of 0.

4. Let represent the user's support for the key data of the i-th potentially abnormal quality event, ln represent the natural logarithm, k represent the baseline fluctuation factor of the consumer goods supply chain, and S represent the steady-state connectivity index of the consumer goods supply chain. This refers to the controllable factors in the development of consumer product quality incidents; The uncertainty factor of the source parameter of the multi-source data of consumer products is compared with a preset threshold. If the uncertainty factor is higher than the preset threshold, an enhanced data collection instruction is generated, and the specific data source that needs to be collected is determined based on the enhanced data collection instruction. Collect relevant parameters from a specific data source and insert them into multi-source data.

Images