Supply chain dynamic evaluation method and device based on multi-modal perception and storage medium

Abstract

The invention relates to the technical field of supply chain management, in particular to a supply chain dynamic evaluation method, device and equipment based on multi-modal perception and a computer storage medium. According to the supply chain dynamic evaluation method based on multi-modal perception, through deployment of a multi-source heterogeneous perception equipment matrix and protocol adaptive conversion, real-time and unified acquisition and preprocessing of supply chain whole-link multi-modal data are realized, and the problems of single data source and data islands are effectively solved; furthermore, a demand-driven dynamic fusion mechanism is introduced, data weights are adaptively distributed according to correlation coefficients, fusion data meeting evaluation demands are generated, and the defect that traditional fixed weight fusion is insufficient in pertinence is overcome. And finally, by constructing a hierarchical evaluation model, accurate mapping and calculation from index standardization to links and even the overall score of the supply chain are realized, and the accuracy and practicability of an evaluation result are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, and in particular to a method, apparatus, equipment, and computer storage medium for dynamic evaluation of supply chains based on multimodal perception. Background Technology

[0002] In today's era of deep integration of digitalization and intelligence, supply chain management is undergoing a transformation from traditional to intelligent models. With increasing market complexity and competition, the demand for "precise sensing and efficient decision-making" has significantly increased across all links of the supply chain. Traditional supply chain data collection relies on single sensors or manual input, resulting in limited data dimensions and an inability to fully reflect the true operational status—for example, in logistics and transportation, relying solely on GPS location information cannot simultaneously provide real-time information on vehicle load, road conditions, and cargo status, leading to delayed identification of the root cause of problems.

[0003] Existing sensing technologies lack adaptability in complex scenarios: In warehouse management, densely packed goods and poor lighting conditions can prevent vision devices from accurately identifying labels; in the manufacturing process, strong electromagnetic interference can reduce sensor accuracy, causing deviations in the monitoring of key parameters and affecting product quality and production efficiency. Furthermore, data from different stages is stored in independent systems with inconsistent formats and interfaces, creating "data silos" that hinder data flow and comprehensive analysis, thus failing to support overall supply chain optimization.

[0004] Traditional data analysis algorithms struggle to meet the demands for "real-time processing" and "in-depth mining" of multi-source heterogeneous data: they suffer from low computational efficiency when dealing with massive amounts of data, leading to delayed decision-making; and they lack an effective weight allocation mechanism during data fusion, failing to highlight key data based on the personalized needs of the evaluation subject, resulting in insufficient relevance and accuracy of evaluation results—for example, for companies that focus on cost control, the algorithm cannot prioritize the analysis of data such as procurement costs and logistics costs, making it difficult to support cost optimization decisions.

[0005] In summary, current supply chain evaluation methods have significant shortcomings in data collection, scenario adaptation, and algorithm analysis, making it impossible to achieve efficient and intelligent operation of the supply chain. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to overcome the problems of incomplete data collection, poor adaptability to multiple scenarios, and low accuracy of evaluation algorithms in the existing supply chain evaluation methods.

[0007] To address the aforementioned technical problems, this invention provides a supply chain dynamic evaluation method based on multimodal perception, comprising: Deploy a matrix of multi-source heterogeneous sensing devices, adapt protocols and network each device, and use the matrix of multi-source heterogeneous sensing devices to collect multimodal data from each link of the supply chain at a preset frequency, and upload the data to the edge computing node for data preprocessing. Determine the evaluation subjects and demand types, extract demand keywords, and calculate the correlation coefficient between each modal data and the demands of each evaluation subject. Calculate the fusion weight of each modal data based on the correlation coefficient, and calculate the fused data of each modal data based on the fusion weight. The evaluation index system for each stage is determined based on the needs of the evaluation subject, and the standardized score of each evaluation index is calculated. The evaluation score of each stage is calculated based on the standardized score, and the overall evaluation score of the supply chain is calculated based on the evaluation score of each stage.

[0008] Preferably, the deployment of a multi-source heterogeneous sensing device matrix, and the protocol adaptation and networking of each device, includes: Deploy a multi-source heterogeneous sensing device matrix throughout the entire supply chain, from production to warehousing, transportation and distribution. This matrix includes, but is not limited to, environmental sensing devices, status sensing devices and visual sensing devices. Through adaptive protocol conversion technology, data from all devices is uploaded to the edge computing node in real time.

[0009] Preferably, the step of collecting multimodal data from each link of the supply chain at a preset frequency using a multi-source heterogeneous sensing device matrix and uploading it to an edge computing node for data preprocessing includes: Utilizing a matrix of multi-source heterogeneous sensing devices, multimodal data from each link of the supply chain is collected at a preset frequency; After uploading the multimodal data to the edge computing node, the wavelet transform algorithm is used to remove outliers from the data and transform the original data of different dimensions into the range of 0-1 to obtain the normalized value of each modality data.

[0010] Preferably, the steps of determining the evaluation subject and demand type, extracting demand keywords, calculating the correlation coefficient between each modal data and the demands of each evaluation subject, calculating the fusion weight of each modal data based on the correlation coefficient, and calculating the fused data of each modal data based on the fusion weight include: Identify the evaluation subjects and their needs, extract the corresponding needs keywords based on the needs types, and obtain all the needs keywords for each evaluation subject; Based on the matching degree between the i-th modal data and the k-th demand keyword of the current evaluation subject, as well as the importance coefficient of the k-th demand keyword, the correlation coefficient between the i-th modal data and the current evaluation subject's needs is calculated. Based on the correlation coefficient between the i-th modal data and the current evaluation subject's needs, the weight of the i-th modal data in the fusion process is calculated; Based on the weights of various modal data in the fusion process and the normalized values ​​of various modal data, the fused data after weighted fusion of each modal data is calculated.

[0011] Preferably, the step of determining the evaluation index system for each stage based on the needs of the evaluation subject, calculating the standardized score of each evaluation index, calculating the evaluation score for each stage based on the standardized score, and calculating the overall evaluation score of the supply chain based on the evaluation score for each stage includes: Based on the needs of the evaluation subject, the evaluation indicators for each link of the supply chain are determined, and the optimal and worst values ​​are set for each indicator, and the weight of each indicator and each link of the supply chain is determined. The raw values ​​of each indicator are standardized to a score range of 0-1, and the standardized score of each indicator is calculated using the positive / negative indicator standardization formula according to the nature of the indicator. The standardized scores of all indicators within a supply chain link are weighted and summed to obtain the evaluation score for that link. The evaluation scores of all stages are weighted and summed to obtain the final overall evaluation score of the entire supply chain.

[0012] Preferably, the supply chain dynamic evaluation method based on multimodal perception further includes: Weakness analysis is performed based on the individual evaluation scores and the overall evaluation scores, and a visual report is generated and pushed to the evaluation subject. Based on feedback from the evaluation subjects, the importance coefficients of each requirement keyword or the weights of each indicator are adjusted, and the matrix layout of multi-source heterogeneous sensing devices is optimized.

[0013] Preferably, the step of analyzing weaknesses based on single-stage evaluation scores and overall evaluation scores, generating a visual report, pushing it to the evaluation subject, adjusting the importance coefficients of each requirement keyword or the weights of each indicator based on the feedback from the evaluation subject, and optimizing the matrix layout of multi-source heterogeneous sensing devices includes: Identify the weak links as those with evaluation scores lower than the historical average or a preset threshold, and drill down into the indicator scores within those links to pinpoint the specific reasons. The overall score, scores for each stage, scores for each indicator, and analysis conclusions on weaknesses are presented in a visual format using score trend charts and indicator radar charts, and are automatically pushed to the evaluation subjects. After receiving the visualization report, the evaluator can adjust the importance coefficient of the demand keywords or the weight of the evaluation indicators in each stage through the visualization interface based on the actual business experience, and the adjustment will take effect immediately in the next evaluation calculation. Regularly analyze the historical correlation coefficients of each modality of data. For data sources with correlation coefficients that are consistently below the preset value, determine that their collection value is low and reduce or remove the corresponding sensing devices. For data sources with correlation coefficients that are above the preset value, increase the deployment density of similar devices in key areas.

[0014] The present invention also provides a supply chain dynamic evaluation device based on multimodal perception, comprising: The data acquisition module is used to deploy a matrix of multi-source heterogeneous sensing devices, adapt protocols and connect the devices, and use the matrix of multi-source heterogeneous sensing devices to collect multimodal data from each link of the supply chain at a preset frequency, and upload the data to the edge computing node for data preprocessing. The data fusion module is used to determine the evaluation subjects and demand types, extract demand keywords, calculate the correlation coefficient between each modal data and the demands of each evaluation subject, calculate the fusion weight of each modal data based on the correlation coefficient, and calculate the fused data of each modal data based on the fusion weight. The scoring and evaluation module is used to determine the evaluation indicator system for each link according to the needs of the evaluation subject, calculate the standardized score of each evaluation indicator, calculate the evaluation score of a single link based on the standardized score, and calculate the overall evaluation score of the supply chain based on the evaluation score of a single link.

[0015] The present invention also provides a supply chain dynamic evaluation device based on multimodal perception, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the above-described supply chain dynamic evaluation method based on multimodal perception when executing the computer program.

[0016] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described supply chain dynamic evaluation method based on multimodal perception.

[0017] The technical solution of the present invention has the following advantages compared with the prior art: The supply chain dynamic evaluation method based on multimodal perception described in this invention achieves real-time, unified collection and preprocessing of multimodal data across the entire supply chain by deploying a matrix of multi-source heterogeneous sensing devices and adaptive protocol conversion, effectively solving the problems of single data sources and "data silos." Furthermore, by introducing a demand-driven dynamic fusion mechanism, data weights are adaptively allocated based on correlation coefficients to generate fused data that fits the evaluation requirements, overcoming the shortcomings of traditional fixed-weight fusion methods in terms of specificity. Finally, by constructing a hierarchical evaluation model, precise mapping and calculation from indicator standardization to the scores of each link and even the entire supply chain are achieved, significantly improving the accuracy and practicality of the evaluation results. Attached Figure Description

[0018] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of a multimodal perception-based dynamic evaluation method for supply chains provided by this invention. Figure 2 This is a structural block diagram of a supply chain dynamic evaluation device based on multimodal perception provided in an embodiment of the present invention. Detailed Implementation

[0019] The core of this invention is to provide a method, device, equipment, and computer storage medium for dynamic evaluation of the supply chain based on multimodal perception, which effectively realizes comprehensive data collection, personalized fusion process, and accurate evaluation results.

[0020] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely 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.

[0021] Please refer to Figure 1. Figure 1 The flowchart illustrates the implementation of a multimodal sensing-based dynamic evaluation method for supply chains provided by this invention; the specific operation steps are as follows: S101: Deploy a multi-source heterogeneous sensing device matrix, perform protocol adaptation and networking for each device, and use the multi-source heterogeneous sensing device matrix to collect multimodal data from each link of the supply chain at a preset frequency, and upload it to the edge computing node for data preprocessing. S102: Determine the evaluation subject and demand type, extract demand keywords, and calculate the correlation coefficient between each modal data and the demand of each evaluation subject. Calculate the fusion weight of each modal data based on the correlation coefficient, and calculate the fused data of each modal data based on the fusion weight. S103: Determine the evaluation index system for each stage based on the needs of the evaluation subject, calculate the standardized score of each evaluation index, calculate the evaluation score of a single stage based on the standardized score, and calculate the overall evaluation score of the supply chain based on the evaluation score of a single stage.

[0022] Based on the above embodiments, this embodiment will provide a detailed description of step S101: In some embodiments, deploying a matrix of multi-source heterogeneous sensing devices and performing protocol adaptation and networking on each device includes: • Deploy a multi-source heterogeneous sensing device matrix across the entire supply chain, including but not limited to: environmental sensing devices, status sensing devices, and visual sensing devices; It should be noted that, in response to the physical characteristics and monitoring needs of the entire supply chain—from production to warehousing to transportation to distribution—three types of sensing devices are deployed to form a multi-source heterogeneous device matrix covering the entire chain, ensuring that data collection is comprehensive and without blind spots. Environmental sensing equipment: Temperature and humidity sensors are deployed in the warehousing process to monitor the storage environment of goods, and dust sensors are deployed in the production process to monitor air quality; Status sensing devices: Vibration sensors and torque sensors are deployed on key equipment in the production process to monitor the health status of the equipment in real time; similar sensors are deployed in the transportation process to monitor vehicle bumps and load status. Visual perception equipment: 3D industrial cameras are deployed at warehouse entrances and exits to automatically identify the volume and quantity of goods; infrared cameras are deployed inside transport vehicles to monitor the stacking status of goods in poor lighting or nighttime conditions.

[0023] • Through adaptive protocol conversion technology, data from all devices is uploaded to the edge computing node in real time.

[0024] It should be noted that an adaptive protocol conversion module is built into the device gateway or edge computing node. This module supports multiple communication protocols such as LoRa (suitable for long-distance, low-power warehousing environments), 5G (suitable for high-speed mobile, high-bandwidth transportation environments), and Ethernet (suitable for fixed, highly stable production environments). It solves the problem of inconsistent interfaces of traditional devices, and encapsulates raw data of different formats into a specified format, and uploads it to the edge computing node in real time and stably.

[0025] In some embodiments, the process of collecting multimodal data from various links in the supply chain at a preset frequency using a multi-source heterogeneous sensing device matrix and uploading it to an edge computing node for data preprocessing includes: • Utilize a matrix of multi-source heterogeneous sensing devices to collect multimodal data from each link of the supply chain at a preset frequency; It should be noted that different data collection frequencies are set according to the criticality of each business step, forming a real-time data stream. For example: Production process: Equipment vibration data is collected every 5 minutes to capture abnormal vibration patterns that may indicate a malfunction.

[0026] During transportation: GPS location data and temperature data inside the vehicle are collected every minute to achieve precise monitoring of the cargo location and environment.

[0027] After uploading multimodal data to the edge computing node, the wavelet transform algorithm is used to remove outliers from the data and transform the original data of different dimensions into the range of 0-1 to obtain the normalized value of each modality data.

[0028] It should be noted that at the edge computing nodes, wavelet transform algorithms are used to clean the original data stream. For example, in real-time analysis of the temperature data stream, if a data point is detected to have a temperature difference exceeding 10°C from the points before and after it, it is identified as an abnormal noise point, and the moving average of the five data points before and after it is used to replace the noise point, thereby smoothing the data.

[0029] To eliminate the influence of different modal data units (such as temperature in °C, vibration amplitude in mm / s, and image data in pixel values), the denoised original modal data are uniformly mapped to the [0,1] interval using the following formula to generate normalized values ​​that can be used for subsequent weighted calculations:

[0030] The original values ​​of the i-th modality data are linearly transformed to the interval [0,1] based on their physically valid range (or historical statistical range). After transformation, 0 represents the worst-case scenario where the data is within an acceptable range, and 1 represents the best-case scenario. : The original value of the i-th modal data after denoising. : The minimum value of the i-th modal data within the valid range, as predefined. : The maximum value of the i-th modal data within the valid range, as preset. : The normalized value of the i-th modal data obtained by calculation.

[0031] Based on the above embodiments, this embodiment will provide a detailed description of step S102: In some embodiments, the evaluation subject and demand type are determined, demand keywords are extracted, and the correlation coefficient between each modal data and the demands of each evaluation subject is calculated. Based on the correlation coefficient, the fusion weight of each modal data is calculated, and the fused data of each modal data is calculated based on the fusion weight, including: • Identify the evaluation subjects and their needs, extract the corresponding needs keywords based on the needs types, and obtain all the needs keywords for each evaluation subject; It should be noted that when initializing or executing a new task, the subject of this evaluation (e.g., "Manufacturer A") and its core demand type (e.g., "Quality Assurance") should be clearly defined. Based on the demand type, a set of core demand keywords should be extracted from a predefined knowledge base. For example, the keyword K corresponding to the "Quality Assurance" demand can be defined as: {Product pass rate, Number of equipment failures, Transportation loss rate}.

[0032] • Based on the matching degree between the i-th modal data and the k-th demand keyword of the current evaluation subject, and the importance coefficient of the k-th demand keyword, calculate the correlation coefficient between the i-th modal data and the current evaluation subject's needs; It should be noted that this process is used to quantify the correlation between each type of modal data (such as temperature data and vibration data) and the current evaluation requirements; Calculate matching degree The semantic similarity between the name / description of the i-th modality data and the k-th demand keyword is calculated using a text semantic matching algorithm (such as cosine similarity calculation based on Word2Vec or BERT). The result is normalized to the interval [0,1]. For example, the matching degree between "temperature data" and "transportation loss rate" may be high (S=0.8), while the matching degree with "number of equipment failures" may be low (S=0.3). Obtaining the importance coefficient The evaluator specifies the relative importance of each keyword when setting requirements, and this must be met. For example, for a "quality assurance type", the following settings can be configured: ; The comprehensive correlation degree of the i-th modality data is calculated using a weighted summation formula. :

[0033] By combining the semantic relevance of the modal data with each keyword and the importance of the keywords themselves, a coefficient representing the overall importance of the modal data to this evaluation task is obtained. : The correlation coefficient of the i-th modal data : The total number of keywords related to evaluation needs The matching degree between the i-th modality data and the k-th demand keyword. : Importance coefficient of the kth demand keyword.

[0034] • Calculate the weight of the i-th modal data in the fusion process based on the correlation coefficient between the i-th modal data and the current evaluation subject's needs; It should be noted that the correlation coefficient is normalized to obtain the final weights used for data fusion:

[0035] The sum of the correlation coefficients of all modal data is used as the denominator to determine the weight of each modal data. It falls within the interval [0,1], and all The sum is 1. This ensures that data highly relevant to the needs accounts for a larger proportion during the fusion process. Among them, : The fusion weights of the i-th modality data The total number of modal data types involved in the fusion. : The correlation coefficient of the j-th modal data.

[0036] • Based on the weights of various modal data in the fusion process and the normalized values ​​of various modal data, calculate the fused data after weighted fusion of each modal data.

[0037] It should be noted that, Using the calculated weights, the preprocessed multimodal data is weighted and fused to generate a single data value representing the overall state of this stage:

[0038] Normalized values ​​of each modal data Its dynamic weight The products are multiplied and summed to obtain a comprehensive score between 0 and 1. This value reflects the overall operational status of this stage under the current evaluation requirements. The combined data value obtained after weighted fusion. : The fusion weights of the i-th modality data : The normalized value of the i-th modal data The total number of modal data types.

[0039] Based on the above embodiments, this embodiment will provide a detailed description of step S103: In some embodiments, the evaluation index system for each stage is determined according to the needs of the evaluation subject, and the standardized score of each evaluation index is calculated. The evaluation score for each stage is then calculated based on the standardized score, and the overall supply chain evaluation score is calculated based on the evaluation score for each stage, including: • Based on the needs of the evaluation subject, determine the evaluation indicators for each link of the supply chain, set the optimal and worst values ​​for each indicator, and determine the weight of each indicator and the weight of each link in the supply chain. It should be noted that, based on the needs of the evaluation subjects, specific and quantifiable evaluation indicators are set for each link in the supply chain. For example, indicators are set for the "production link": product qualification rate (positive), number of equipment failures (negative), and unit energy consumption (negative). Set an expected optimal value for each indicator. And an acceptable worst value For example, the number of equipment failures is set as an indicator. Next / month, Once per month.

[0040] • Standardize the raw values ​​of each indicator to a score range of 0-1, and calculate the standardized score of each indicator using the positive / negative indicator standardization formula according to the nature of the indicator. It should be noted that unifying the measurement standards for different indicators (with different units and directions) involves using the original values ​​of each indicator. Convert to dimensionless standardized score , For positive indicators (the higher the value, the better, such as product qualification rate):

[0041] For inverse metrics (the smaller the value, the better, such as the number of equipment failures):

[0042] These two formulas linearly map the original values ​​of indicators with different properties to the [0,1] interval. When equal hour, ;when equal hour, If the value exceeds this range, truncation can be performed (e.g., less than 0 is considered 0, greater than 1 is considered 1). Among these, The standardized score of the j-th evaluation indicator. The original value of the j-th evaluation indicator. The optimal value set for the j-th evaluation index. : The worst value set for the j-th evaluation index.

[0043] • The standardized scores of all indicators within a supply chain link are weighted and summed to obtain the evaluation score for that link; It should be noted that, Within a single stage (such as the production stage), the standardized scores of each indicator are combined according to their pre-defined importance:

[0044] This formula calculates the weighted sum of all indicators within this stage, multiplied by 100 to convert it into a more intuitive percentage score. A higher score indicates a better performance of this stage. The calculated single-stage evaluation score (out of 100). The weight of the j-th evaluation indicator in this stage (set by domain experts or the evaluation body, and...) , The total number of evaluation indicators in this stage.

[0045] • The evaluation scores of all stages are weighted and summed to obtain the final overall evaluation score of the entire supply chain.

[0046] It should be noted that, By treating the supply chain as a whole and combining the scores of each link, a total score is obtained.

[0047] This formula reflects that the overall performance of the supply chain is a weighted average of the performance of each link. (The weights of each link are...) This demonstrates the strategic importance of this link in the overall supply chain. Among them, The calculated overall evaluation score of the supply chain. The weight of the t-th link in the supply chain (e.g., the production link). warehousing Transportation Delivery process , The single-link evaluation score of the t-th supply chain link. The total number of links in the supply chain.

[0048] Based on the above embodiments, the supply chain dynamic evaluation method based on multimodal perception also includes: • Weakness analysis is performed based on the scores of individual evaluations and the overall evaluation, generating a visual report that is then pushed to the evaluation subjects; • Adjust the importance coefficients of each requirement keyword or the weights of each indicator based on feedback from the evaluation subjects, and optimize the matrix layout of multi-source heterogeneous sensing devices.

[0049] In some embodiments, weakness analysis is performed based on single-stage evaluation scores and overall evaluation scores, generating a visual report, which is then pushed to the evaluation subject. The importance coefficients of each requirement keyword or the weights of each indicator are adjusted based on feedback from the evaluation subject, and the layout of the multi-source heterogeneous sensing device matrix is ​​optimized, including: • Identify the weak links as those with evaluation scores lower than the historical average or a preset threshold, and drill down into the indicator scores within that link to pinpoint the specific reasons. It should be noted that the scores of each stage are compared. By comparing scores with historical averages or preset thresholds, areas with significantly low scores are identified as weak points. Furthermore, the analysis drills down to the specific metrics within these weak points. The specific reasons for this can be identified (e.g., "the transportation segment scored 72 points, mainly due to temperature indicators"). The value is only 0.6, indicating that the temperature control is unstable.

[0050] • The overall score, scores for each stage, indicator scores, and analysis conclusions on weak points are presented in a visual format using score trend charts and indicator radar charts, and are automatically pushed to the evaluation subjects; It should be noted that the overall score, scores for each stage, indicator scores, and analysis conclusions on weak points will be presented in a visual format, such as a score trend chart (showing historical changes) and an indicator radar chart (showing the distribution of strengths and weaknesses), and will be automatically pushed to the evaluation subject.

[0051] • After receiving the visualization report, the evaluator can adjust the importance coefficient of the requirement keywords or the weight of the evaluation indicators in each stage through the visualization interface based on the actual business experience, and the adjustment will take effect immediately in the next evaluation calculation. It should be noted that after receiving the report, the evaluator can adjust the importance coefficient of the required keywords through a visual interface based on their actual business experience. Or adjust the weights of evaluation indicators within each stage. These adjustments are recorded and immediately applied in the next evaluation calculation, making the evaluation model more closely reflect users' actual preferences.

[0052] • Regularly analyze the historical correlation coefficients of each modality of data. For data sources with correlation coefficients that are consistently below the preset value, determine that their collection value is low and reduce or remove the corresponding sensing devices. For data sources with correlation coefficients that are above the preset value, increase the deployment density of similar devices in key areas.

[0053] It should be noted that the historical correlation coefficients of each modality data should be statistically analyzed periodically (e.g., monthly). For those that have been in a state of low correlation for a long time (such as...) Data sources with a sensitivity consistently below 0.1 are considered to have low collection value, and it is recommended to reduce or remove the corresponding sensing devices. Conversely, for highly correlated data, consider increasing the deployment density of similar devices in key areas to improve data quality and system reliability. This approach enables on-demand allocation of sensing resources and continuously optimizes the return on investment.

[0054] This invention breaks through the traditional "collection-analysis" separation architecture, achieving for the first time a fully integrated and dynamically linked "perception-fusion-evaluation" process for supply chain data. It solves the problems of data silos and rigid evaluation methods. For the first time, it proposes a dynamic weighting mechanism driven by "demand-data correlation," replacing traditional fixed-weight fusion to ensure that fused data aligns with personalized needs and improves the targeting of data fusion. It constructs a hierarchical evaluation model, for the first time distinguishing between standardized processing methods for positive and negative indicators, achieving precise score mapping from "single link to the whole," solving the problem of poor targeting of evaluation results. It translates the architecture and algorithms into an executable full-process solution, achieving for the first time a closed loop of "real-time collection-near real-time analysis-dynamic optimization" for supply chain evaluation, ensuring the technology's feasibility and practicality. It breaks through the protocol barriers of different types of sensing devices, providing key technical support for real-time uploading of multi-source data, and is the core foundation for realizing a "multi-source heterogeneous device matrix."

[0055] Please refer to Figure 2 , Figure 2 A structural block diagram of a supply chain dynamic evaluation device based on multimodal perception provided in an embodiment of the present invention; the specific device may include: The data acquisition module 100 is used to deploy a multi-source heterogeneous sensing device matrix, perform protocol adaptation and networking for each device, and use the multi-source heterogeneous sensing device matrix to collect multimodal data from each link of the supply chain at a preset frequency, and upload it to the edge computing node for data preprocessing. The data fusion module 200 is used to determine the evaluation subject and demand type, extract demand keywords, calculate the correlation coefficient between each modal data and the demand of each evaluation subject, calculate the fusion weight of each modal data based on the correlation coefficient, and calculate the fused data of each modal data based on the fusion weight. The scoring and evaluation module 300 is used to determine the evaluation indicator system for each link according to the needs of the evaluation subject, calculate the standardized score of each evaluation indicator, calculate the evaluation score of a single link based on the standardized score, and calculate the overall evaluation score of the supply chain based on the evaluation score of the single link.

[0056] The multimodal perception-based supply chain dynamic evaluation device of this embodiment is used to implement the aforementioned multimodal perception-based supply chain dynamic evaluation method. Therefore, the specific implementation of the multimodal perception-based supply chain dynamic evaluation device can be found in the embodiment section of the multimodal perception-based supply chain dynamic evaluation method above. For example, the data acquisition module, data fusion module, and scoring evaluation module are used to implement steps S101, S102, S103, S104, and S105 in the aforementioned multimodal perception-based supply chain dynamic evaluation method. Therefore, its specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0057] A specific embodiment of the present invention also provides a supply chain dynamic evaluation device based on multimodal perception, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described supply chain dynamic evaluation method based on multimodal perception.

[0058] A specific embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described supply chain dynamic evaluation method based on multimodal perception.

[0059] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A dynamic evaluation method for supply chains based on multimodal perception, characterized in that, include: Deploy a matrix of multi-source heterogeneous sensing devices, adapt protocols and network each device, and use the matrix of multi-source heterogeneous sensing devices to collect multimodal data from each link of the supply chain at a preset frequency, and upload the data to the edge computing node for data preprocessing. Determine the evaluation subjects and demand types, extract demand keywords, and calculate the correlation coefficient between each modal data and the demands of each evaluation subject. Calculate the fusion weight of each modal data based on the correlation coefficient, and calculate the fused data of each modal data based on the fusion weight. The evaluation index system for each stage is determined based on the needs of the evaluation subject, and the standardized score of each evaluation index is calculated. The evaluation score of each stage is calculated based on the standardized score, and the overall evaluation score of the supply chain is calculated based on the evaluation score of each stage.

2. The supply chain dynamic evaluation method based on multimodal perception according to claim 1, characterized in that, The deployment of a multi-source heterogeneous sensing device matrix, including protocol adaptation and networking of each device, includes: Deploy a multi-source heterogeneous sensing device matrix throughout the entire supply chain, from production to warehousing, transportation and distribution. This matrix includes, but is not limited to, environmental sensing devices, status sensing devices and visual sensing devices. Through adaptive protocol conversion technology, data from all devices is uploaded to the edge computing node in real time.

3. The supply chain dynamic evaluation method based on multimodal perception according to claim 1, characterized in that, The step of collecting multimodal data from each link of the supply chain at a preset frequency using a matrix of multi-source heterogeneous sensing devices and uploading it to an edge computing node for data preprocessing includes: Utilizing a matrix of multi-source heterogeneous sensing devices, multimodal data from each link of the supply chain is collected at a preset frequency; After uploading the multimodal data to the edge computing node, the wavelet transform algorithm is used to remove outliers from the data and transform the original data of different dimensions into the range of 0-1 to obtain the normalized value of each modality data.

4. The supply chain dynamic evaluation method based on multimodal perception according to claim 3, characterized in that, The process involves determining the evaluation subject and demand type, extracting demand keywords, calculating the correlation coefficient between each modal data and the demands of each evaluation subject, calculating the fusion weight of each modal data based on the correlation coefficient, and calculating the fused data of each modal data based on the fusion weight, including: Identify the evaluation subjects and their needs, extract the corresponding needs keywords based on the needs types, and obtain all the needs keywords for each evaluation subject; Based on the matching degree between the i-th modal data and the k-th demand keyword of the current evaluation subject, as well as the importance coefficient of the k-th demand keyword, the correlation coefficient between the i-th modal data and the current evaluation subject's needs is calculated. Based on the correlation coefficient between the i-th modal data and the current evaluation subject's needs, the weight of the i-th modal data in the fusion process is calculated; Based on the weights of various modal data in the fusion process and the normalized values ​​of various modal data, the fused data after weighted fusion of each modal data is calculated.

5. The supply chain dynamic evaluation method based on multimodal perception according to claim 4, characterized in that, The process of determining the evaluation indicator system for each stage based on the needs of the evaluation subject, calculating the standardized score of each evaluation indicator, calculating the evaluation score for each stage based on the standardized score, and calculating the overall evaluation score of the supply chain based on the evaluation score for each stage includes: Based on the needs of the evaluation subject, the evaluation indicators for each link of the supply chain are determined, and the optimal and worst values ​​are set for each indicator, and the weight of each indicator and each link of the supply chain is determined. The raw values ​​of each indicator are standardized to a score range of 0-1, and the standardized score of each indicator is calculated using the positive / negative indicator standardization formula according to the nature of the indicator. The standardized scores of all indicators within a supply chain link are weighted and summed to obtain the evaluation score for that link. The evaluation scores of all stages are weighted and summed to obtain the final overall evaluation score of the entire supply chain.

6. The supply chain dynamic evaluation method based on multimodal perception according to claim 5, characterized in that, Also includes: Weakness analysis is performed based on the individual evaluation scores and the overall evaluation scores, and a visual report is generated and pushed to the evaluation subject. Based on feedback from the evaluation subjects, the importance coefficients of each requirement keyword or the weights of each indicator are adjusted, and the matrix layout of multi-source heterogeneous sensing devices is optimized.

7. The supply chain dynamic evaluation method based on multimodal perception according to claim 6, characterized in that, The process of analyzing weaknesses based on single-stage evaluation scores and overall evaluation scores, generating a visual report, and pushing it to the evaluation subject, adjusting the importance coefficients of each requirement keyword or the weights of each indicator based on the feedback from the evaluation subject, and optimizing the matrix layout of multi-source heterogeneous sensing devices includes: Identify the weak links as those with evaluation scores lower than the historical average or a preset threshold, and drill down into the indicator scores within those links to pinpoint the specific reasons. The overall score, scores for each stage, scores for each indicator, and analysis conclusions on weaknesses are presented in a visual format using score trend charts and indicator radar charts, and are automatically pushed to the evaluation subjects. After receiving the visualization report, the evaluator can adjust the importance coefficient of the demand keywords or the weight of the evaluation indicators in each stage through the visualization interface based on the actual business experience, and the adjustment will take effect immediately in the next evaluation calculation. Regularly analyze the historical correlation coefficients of each modality of data. For data sources with correlation coefficients that are consistently below the preset value, determine that their collection value is low and reduce or remove the corresponding sensing devices. For data sources with correlation coefficients that are above the preset value, increase the deployment density of similar devices in key areas.

8. A supply chain dynamic evaluation device based on multimodal perception, characterized in that, include: The data acquisition module is used to deploy a matrix of multi-source heterogeneous sensing devices, adapt protocols and connect the devices, and use the matrix of multi-source heterogeneous sensing devices to collect multimodal data from each link of the supply chain at a preset frequency, and upload the data to the edge computing node for data preprocessing. The data fusion module is used to determine the evaluation subjects and demand types, extract demand keywords, calculate the correlation coefficient between each modal data and the demands of each evaluation subject, calculate the fusion weight of each modal data based on the correlation coefficient, and calculate the fused data of each modal data based on the fusion weight. The scoring and evaluation module is used to determine the evaluation indicator system for each link according to the needs of the evaluation subject, calculate the standardized score of each evaluation indicator, calculate the evaluation score of a single link based on the standardized score, and calculate the overall evaluation score of the supply chain based on the evaluation score of a single link.

9. A supply chain dynamic evaluation device based on multimodal perception, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of a multimodal perception-based dynamic evaluation method for supply chains as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the supply chain dynamic evaluation method based on multimodal perception as described in any one of claims 1 to 7.

Images