Artificial intelligence-based digital enterprise management data analysis method and system

Abstract

The application relates to the technical field of graphic data processing, and provides a digital enterprise management data analysis method and system based on artificial intelligence, which is used for generating a visual graphic analysis report by fusing dynamic decision logic based on comprehensive and diversified supply chain data, improving the comprehensiveness, dynamics and decision guidance of graphic generation, and effectively guiding the supply chain dynamic adjustment of an enterprise management system. The method comprises the following steps: collecting a supply chain data set of a target enterprise, performing feature extraction on the supply chain data set, obtaining a supply chain trend prediction feature set and a supply chain anomaly detection feature set; based on a preset reinforcement learning decision optimization model, performing dynamic decision path generation processing on the supply chain trend prediction feature set and the supply chain anomaly detection feature set, and generating a multi-stage decision optimization path; generating a visual graphic analysis report according to the multi-stage decision optimization path and the supply chain data set, and feeding back the visual graphic analysis report to an enterprise management system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of graphic data processing, and particularly relates to a digital enterprise management data analysis method and system based on artificial intelligence. BACKGROUND

[0002] With the promotion of enterprise digital transformation, visual graphic generation has become increasingly critical in the field of enterprise data analysis. Visual graphics can convert massive complex data into intuitive and easy-to-understand forms such as charts and graphs, greatly improving the efficiency of data understanding and analysis, assisting enterprise managers in quickly obtaining key information and making accurate decisions. Effective visual graphics can clearly show each link of enterprise operation, helping enterprises grasp the overall situation and optimize resource allocation.

[0003] However, the current visual graphic generation technology has obvious shortcomings. Most existing technologies are relatively single in data sources, usually relying on only one type of data for graphic generation, resulting in one-sided graphic content that cannot reflect the overall picture of enterprise operation. Moreover, existing graphic generation lacks deep integration of complex decision-making logic, and the generated graphics are mostly static displays that cannot generate dynamic decision-oriented visual graphics based on real-time changing data and dynamic decision-making needs of enterprises. SUMMARY

[0004] The present application provides a digital enterprise management data analysis method and system based on artificial intelligence, which generates visual graphic analysis reports by integrating dynamic decision-making logic based on comprehensive and diverse supply chain data, thereby improving the comprehensiveness, dynamics and decision-making guidance of graphic generation, and effectively guiding the supply chain dynamic adjustment of the enterprise management system.

[0005] In a first aspect, the present application provides a digital enterprise management data analysis method based on artificial intelligence, applied to a digital enterprise management data analysis system. The method comprises: collecting a supply chain data set of a target enterprise; the supply chain data set includes text-form order interaction record data and image-form logistics node monitoring data; performing feature extraction on the supply chain data set to obtain a supply chain trend prediction feature set and a supply chain anomaly detection feature set; based on a preset reinforcement learning decision optimization model, performing dynamic decision path generation processing on the supply chain trend prediction feature set and the supply chain anomaly detection feature set to generate a multi-stage decision optimization path; generating a visual graphic analysis report based on the multi-stage decision optimization path and the supply chain data set, and feeding back the visual graphic analysis report to an enterprise management system; wherein the visual graphic analysis report is used to instruct the enterprise management system to perform supply chain dynamic adjustment operations.

[0006] In a second aspect, an embodiment of the present application provides a digital enterprise management data analysis system, comprising a processor and a memory, wherein the memory stores a computer program which, when executed by the processor, causes the processor to perform the steps of the method described above.

[0007] In a third aspect, an embodiment of the present application provides a computer-readable storage medium comprising a computer program which, when executed on a digital enterprise management data analysis system, causes the digital enterprise management data analysis system to perform the steps of the method described above.

[0008] The embodiment of the present application generates a visual text analysis report based on comprehensive and diverse supply chain data and dynamic decision-making logic, which can improve the comprehensiveness, dynamics and decision-making guidance of the generated text and images, thereby effectively guiding the supply chain dynamic adjustment of the enterprise management system.

[0009] In detail, the embodiment of the present application provides comprehensive and diverse data cornerstone for text and image generation by collecting supply chain data sets containing text and image forms; the supply chain trend prediction feature set and the supply chain anomaly detection feature set obtained by feature extraction inject depth and pertinence into the text and image content; the multi-stage decision optimization path generated based on the reinforcement learning decision optimization model further provides a clear logic framework for text and image generation, ensuring that the generated visual text analysis report is not a simple data list, but a deep analysis result with logical coherence and decision-oriented guidance. The visual text analysis report not only intuitively presents the supply chain status, but also indicates the enterprise management system to perform supply chain dynamic adjustment operations according to the multi-stage decision path in a clear and understandable text and image form, thereby improving the visualization and intelligent level of enterprise decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A flowchart of a digital enterprise management data analysis method based on artificial intelligence provided by an embodiment of the present application.

[0011] Figure 2 A structural diagram of a digital enterprise management data analysis system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0012] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the technical scheme of the present application, rather than all the embodiments. Based on the embodiments described in the present application document, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the technical scheme of the present application.

[0013] Referring to Figure 1 It is a digital enterprise management data analysis method based on artificial intelligence provided in an embodiment of the application. The method can be applied to a digital enterprise management data analysis system, and the specific process is as shown in steps 110-140.

[0014] Step 110: Collecting a supply chain data set of a target enterprise.

[0015] In this embodiment, a target enterprise with a complex business structure is taken as an example. The supply chain of the enterprise covers various business scenarios and links. The collection of the supply chain data set is realized through a data collection architecture built internally by the enterprise. The architecture connects multiple data sources and collects order interaction record data from the order information management system of the enterprise. These data are in the form of text and contain various order-related information, such as order identifier, order time, cargo category identifier, and cargo quantity identifier.

[0016] For logistics node monitoring data, the monitoring equipment distributed at various logistics nodes is used for collection. The real-time status of the logistics nodes is recorded in the form of images, including logistics node identifier, cargo storage state image, and transportation tool state image. The collection process strictly follows the data collection standard specification to ensure that the collected data are accurate, complete, and representative.

[0017] Step 120: Feature extraction is performed on the supply chain data set to obtain a supply chain trend prediction feature set and a supply chain anomaly detection feature set.

[0018] In some examples, the supply chain data set includes order interaction record data in the form of text and logistics node monitoring data in the form of images. Step 120 includes:

[0019] Step 121: Time-series image feature extraction processing is performed on the logistics node monitoring data in the form of images to extract the cargo stacking density change feature, transportation tool stay duration feature, and environmental light intensity fluctuation feature of each logistics node, and generate a logistics node dynamic feature set.

[0020] For the image form of logistics node monitoring data, image feature extraction algorithm is used to process each logistics node. First, for the extraction of cargo accumulation density change feature, by analyzing the cargo storage state images at different times, according to the distribution of cargo in the image and the image pixel information, the proportion of the image area occupied by the cargo is calculated to represent the cargo accumulation density, and the density values at different times are recorded to form the change sequence of cargo accumulation density with time as the cargo accumulation density change feature. For example, for the logistics node M, from the cargo storage state images at time t1 to time t2, the number of distributed pixel points of the cargo in the image is analyzed, and the total number of image pixels is combined to calculate the cargo accumulation density D1 at time t1 and the cargo accumulation density D2 at time t2. A series of density values are obtained to form the cargo accumulation density change feature.

[0021] For the transportation tool staying time feature, by analyzing the transportation tool state images frame by frame, the time points of the transportation tool entering and leaving the logistics node are identified, and the time interval between them is calculated as the staying time of the transportation tool in the logistics node. For example, in the transportation tool state image sequence of logistics node N, the image recognition technology is used to determine that the transportation tool A enters at time t3 and leaves at time t4, and the staying time of the transportation tool A in the logistics node N is t4-t3. The staying time of multiple transportation tools at different times is recorded to form the transportation tool staying time feature.

[0022] For the environmental light intensity fluctuation feature, the light intensity value of each frame of image is calculated by using the commonly used light intensity calculation method, and the change at different times is analyzed. For example, the commonly used light intensity calculation model is used to process the environment image of logistics node P to obtain the light intensity value L1 at time t5 and the light intensity value L2 at time t6. The change of the light intensity values with time is analyzed to obtain the environmental light intensity fluctuation feature. The extracted cargo accumulation density change feature, transportation tool staying time feature and environmental light intensity fluctuation feature are integrated to generate the logistics node dynamic feature set.

[0023] Step 122: Perform semantic association analysis on the text form of order interaction record data, extract demand keyword distribution feature, supplier response timeliness feature and contract clause conflict semantic feature in the order text, and generate order interaction semantic feature set.

[0024] For text-form order interaction record data, semantic analysis algorithm is used for processing. For demand keyword distribution characteristics, first, the order text is segmented, and the text is decomposed into independent words. Then, through keyword extraction algorithm, demand-related keywords such as "product name" and "quantity" are identified, and the occurrence frequency and position information of these keywords in the order text are counted to construct the demand keyword distribution characteristics. For example, for order text T, a series of words are obtained after segmentation, and the demand keywords "product X" and "quantity Y" are determined through keyword extraction algorithm. The number of times and position of "product X" in the text and the number of times and position of "quantity Y" in the text are counted to form the demand keyword distribution characteristics.

[0025] For supplier response timeliness characteristics, the time difference between the order issuance time and the supplier reply time in the order text is calculated to measure the timeliness of the supplier response. For example, order O is issued at time t7, and the supplier replies at time t8. The supplier response timeliness is t8-t7. The supplier response timeliness information of multiple orders is recorded to form the supplier response timeliness characteristics.

[0026] For contract clause conflict semantic characteristics, semantic analysis is performed on the contract clause-related content in the order text, and a preset contract clause semantic model is used to identify possible conflict semantics in the text. For example, the descriptions of key contract clauses such as price and delivery period in different order texts are compared to analyze whether there is semantic contradiction or inconsistency. If there are different expressions about the price clause in the order text, the semantic model is used to determine whether there is a conflict. The conflict semantic information is sorted to form the contract clause conflict semantic characteristics. The demand keyword distribution characteristics, supplier response timeliness characteristics, and contract clause conflict semantic characteristics are integrated to generate the order interaction semantic feature set.

[0027] Step 123: Call the pre-trained supply chain trend prediction model to perform cross-modal feature fusion processing on the logistics node dynamic feature set and the order interaction semantic feature set to generate the supply chain trend prediction feature set.

[0028] In an optional embodiment, step 123 includes:

[0029] Step 1231: Time alignment processing is performed on the cargo accumulation density change feature and the transportation tool stay time length feature in the logistics node dynamic feature set to generate a first time sequence feature vector. Semantic encoding processing is performed on the demand keyword distribution feature and the supplier response timeliness feature in the order interaction semantic feature set to generate a second time sequence feature vector.

[0030] Optionally, for the cargo accumulation density change feature and the transportation tool stay time length feature in the logistics node dynamic feature set, since they come from different monitoring dimensions, but both have time series characteristics. First, the time stamps of the time series corresponding to the cargo accumulation density change feature and the transportation tool stay time length feature are matched. For example, the time series of the cargo accumulation density change feature is [D1, D2, D3,...], and the corresponding time stamp is [t1, t2, t3,...], the time series of the transportation tool stay time length feature is [L1, L2, L3,...], and the corresponding time stamp is [t1', t2', t3',...], through the comparison and adjustment of the time stamps, the two time series are aligned, so that they have corresponding feature values at the same time point. Then, the aligned cargo accumulation density change feature and the transportation tool stay time length feature are combined in a certain order to form a first time series feature vector. For example, the cargo accumulation density change feature values and the transportation tool stay time length feature values are arranged in turn to obtain the first time series feature vector V1 = [D1, L1, D2, L2, D3, L3,...].

[0031] For the demand keyword distribution feature and the supplier response timeliness feature in the order interaction semantic feature set, a commonly used semantic encoding algorithm is adopted. For the demand keyword distribution feature, each demand keyword is mapped into a high-dimensional semantic space, and the semantic information of the keyword is converted into a vector representation through a commonly used mapping function. For example, the demand keyword "product X" is mapped to vector S1 through mapping function F1, and the demand keyword "quantity Y" is mapped to vector S2 through mapping function F1. For the supplier response timeliness feature, the numerical information is normalized to make it have the same dimension and scale range as the semantic vector. For example, the supplier response timeliness value t is normalized to obtain the normalized value t' through the normalization function G. Then, the demand keyword distribution feature vector after semantic encoding and the normalized supplier response timeliness feature value are combined to form a second time series feature vector V2. For example, V2 = [S1, t', S2, t",...].

[0032] Step 1232: adopting an attention weight distribution mechanism to dynamically adjust the weights of the first time series feature vector and the second time series feature vector, and generating a fused cross-modal time series feature vector; wherein the dynamic weight adjustment process includes assigning different attention weight coefficients according to the matching degree of the time stamps of the logistics node monitoring data and the time stamps of the order interaction records.

[0033] Optionally, an attention weight distribution mechanism is used to fuse the first time sequence feature vector and the second time sequence feature vector. First, the matching degree of the timestamp of the logistics node monitoring data and the timestamp of the order interaction record is calculated. For example, for the feature value of the logistics node monitoring data at time t and the feature value of the order interaction record at time t', the time difference |t-t'| between them is calculated by the time difference calculation function H, and the matching degree is determined according to the preset threshold. If the time difference is less than the threshold, the matching degree is high, otherwise the matching degree is low.

[0034] According to the matching degree, different attention weight coefficients are assigned to the first time sequence feature vector and the second time sequence feature vector. For example, when the matching degree is high, a higher weight coefficient w1 is assigned to the first time sequence feature vector and a lower weight coefficient w2 is assigned to the second time sequence feature vector; when the matching degree is low, the weight coefficients are assigned in the opposite way. Then, the first time sequence feature vector and the second time sequence feature vector are fused by weighted summation. For example, the fused cross-modal time sequence feature vector V = w1*V1+w2*V2, where V1 is the first time sequence feature vector and V2 is the second time sequence feature vector. Through this dynamic weight adjustment process, the fused cross-modal time sequence feature vector can better integrate the information of the two modalities.

[0035] Step 1233: input the cross-modal time sequence feature vector into the supply chain trend prediction model, extract trend change patterns of different time granularities through a multi-layer time convolution network, and generate the supply chain trend prediction feature set.

[0036] Optionally, the fused cross-modal time sequence feature vector is input into a pre-trained supply chain trend prediction model, which includes a multi-layer time convolution network, and each layer of the time convolution network has a commonly used convolution kernel size and step. First, the input cross-modal time sequence feature vector enters the first layer of the time convolution network, and the convolution kernel performs sliding convolution operation on the feature vector in the time dimension. For example, the convolution kernel size is k and the step is s, and the convolution kernel starts from the starting position of the feature vector and moves s positions each time to perform convolution calculation on the feature vector segment with length k. Through convolution calculation, local features and trend change information at this time granularity are extracted.

[0037] An exemplary calculation process is that, for each position i in the feature vector V, the convolution calculation result C(i) = ∑(j=0tok-1) w(j) * V(i+j) is calculated, where w(j) is the weight coefficient of the convolution kernel. The result output by the first layer convolution network is taken as the input of the second layer time convolution network, and the second layer time convolution network repeats the above convolution operation, but the convolution kernel size and step length can be different, further extracting more complex and more macroscopic trend change patterns at different time granularities. Through layer-by-layer processing of the multi-layer time convolution network, different time granularity trend change patterns are finally extracted, which constitute the supply chain trend prediction feature set.

[0038] Step 124: performing local anomaly pattern recognition processing on the logistics node dynamic feature set by using an anomaly detection convolution network to generate a supply chain anomaly detection feature set; the supply chain anomaly detection feature set includes cargo retention risk features, transportation route deviation features, and order matching anomaly features.

[0039] In an optional embodiment, step 124 includes:

[0040] Step 1241: performing light abnormality detection processing on the environmental light intensity fluctuation feature in the logistics node dynamic feature set, extracting the occurrence time and duration of the light intensity mutation event, and generating a light abnormality event feature; performing density gradient analysis processing on the cargo stacking density change feature, identifying abnormal stacking areas with a density change rate exceeding a preset threshold, and generating a cargo retention risk feature; performing stay pattern clustering processing on the transportation tool stay duration feature, marking transportation tools deviating from the preset stay duration as route deviation candidates, and generating a transportation route deviation feature.

[0041] It can be understood that, for the environmental light intensity fluctuation feature, the light abnormality detection algorithm is used for processing. First, the change rate of the light intensity is calculated, by comparing the light intensity values at adjacent time points, such as the light intensity at time t L(t) and the light intensity at time t+1 L(t+1). The light intensity change rate R = |L(t+1)-L(t)| is calculated. When the light intensity change rate R exceeds a preset light intensity change threshold T1, it is determined that a light intensity mutation event occurs. The occurrence time t and the duration of the mutation event are recorded, and the duration is obtained by calculating the time interval until the light intensity change rate returns to below the threshold, thereby generating a light abnormality event feature.

[0042] For the characteristics of the change of the cargo stacking density, a density gradient analysis is performed. The gradient change of the cargo stacking density in space and time is calculated, for example, the change of the cargo stacking density at different positions and different times of the logistics nodes. For the cargo stacking density D(x, t) at position x and time t, the density gradient is obtained by calculating the density difference of the adjacent positions and adjacent times. When the density change rate exceeds the preset density change threshold T2, the region is determined as an abnormal stacking region, and the region and its related time information are marked to generate the cargo retention risk feature.

[0043] For the transportation tool stay duration feature, a stay mode clustering algorithm is used. First, according to historical data or a preset standard, a preset stay duration range is determined. Then, the current transportation tool stay duration is analyzed and compared with the preset stay duration range. For the transportation tools with stay duration deviating from the preset range, they are marked as route deviation candidates, and the identification, stay duration and related logistics node information of these transportation tools are recorded to generate the transportation route deviation feature.

[0044] Step 1242: The light abnormal event feature, the cargo retention risk feature and the transportation route deviation feature are associated and matched with the contract clause conflict semantic feature in the order interaction semantic feature set to generate the supply chain anomaly detection feature set.

[0045] Optionally, the light abnormal event feature, the cargo retention risk feature and the transportation route deviation feature are associated and matched with the contract clause conflict semantic feature in the order interaction semantic feature set. By establishing an association rule, for example, when an abnormal stacking area in the cargo retention risk feature is associated with the contract clause of the cargo delivery of a certain order in the order interaction semantic feature set, they are matched. For the light abnormal event feature, if its occurrence time overlaps with the transportation time of a certain order, and the order interaction semantic feature set has a contract clause conflict semantic feature for the order, they are associated. For the transportation route deviation feature, if the deviated transportation tool transports the cargo related to a certain order, and the order has a contract clause conflict semantic feature, they are also associated and matched. These features that have been associated and matched are integrated to generate the supply chain anomaly detection feature set.

[0046] In an alternative embodiment, after the supply chain anomaly detection feature set is generated, it further includes:

[0047] Step 125: input the goods retention risk features and transportation route deviation features in the supply chain anomaly detection feature set into the reinforcement learning decision optimization model, trigger the anomaly feedback learning mechanism of the reinforcement learning decision optimization model; dynamically adjust the reward and punishment function weights in the anomaly feedback learning mechanism according to the historical processing records of the goods retention risk features and the severity of the current transportation route deviation features; based on the adjusted reward and punishment function weights, incrementally train the policy network of the reinforcement learning decision optimization model to generate an updated reinforcement learning decision optimization model; apply the updated reinforcement learning decision optimization model to the processing of subsequent supply chain anomaly detection feature sets to realize dynamic adaptive adjustment of the multi-stage decision optimization path.

[0048] In the embodiment of the application, the goods retention risk features and transportation route deviation features in the supply chain anomaly detection feature set are input into the reinforcement learning decision optimization model. After the model receives these features, the anomaly feedback learning mechanism is triggered. First, the historical processing records of the goods retention risk features are queried, which contain the decisions and processing results taken in the past for similar goods retention risk situations. At the same time, the severity of the current transportation route deviation features is evaluated, for example, by comprehensively evaluating the distance, time, and other factors of the deviation.

[0049] According to the historical processing records and the current severity, the reward and punishment function weights in the anomaly feedback learning mechanism are dynamically adjusted. If the historical processing records show that a certain decision-making approach has achieved good results, and the current transportation route deviation severity is high, the weight of the corresponding reward function is increased; on the contrary, if the historical decision-making effect is poor, and the current situation is relatively light, the reward function weight is reduced or the punishment function weight is increased.

[0050] Based on the adjusted reward and punishment function weights, the policy network of the reinforcement learning decision optimization model is incrementally trained. During the training process, the model adjusts the parameters of the policy network according to the current input features and the adjusted reward and punishment function weights. For example, through the gradient descent algorithm, the error between the output of the policy network and the expected output is calculated, and the weight parameters of the policy network are updated according to the error backpropagation, so that the model can better cope with the current abnormal situation. After multiple incremental training, an updated reinforcement learning decision optimization model is generated. The updated model is applied to the processing of subsequent supply chain anomaly detection feature sets, and the model will make more reasonable decisions based on new anomaly features, realizing dynamic adaptive adjustment of the multi-stage decision optimization path.

[0051] Step 130: based on the preset reinforcement learning decision optimization model, perform dynamic decision path generation processing on the supply chain trend prediction feature set and the supply chain anomaly detection feature set to generate a multi-stage decision optimization path.

[0052] In another example, the multi-stage decision optimization path is used to indicate the priority sequence of supply chain resource configuration and the execution order of abnormal handling operations, step 130, including:

[0053] Step 131: Decision priority sorting processing is performed on the target goods demand distribution features and the logistics efficiency bottleneck prediction features in the supply chain trend prediction feature set, to generate a resource configuration priority sequence.

[0054] In embodiments of the present application, the target goods demand distribution features and the logistics efficiency bottleneck prediction features in the supply chain trend prediction feature set are sorted by decision priority. For the target goods demand distribution features, factors such as the demand frequency and demand urgency of different goods are analyzed. For example, for goods A and goods B, the demand frequency of goods A and goods B in historical order data, and the delivery deadline of goods A and goods B in the current order, etc. are counted, and the demand urgency of goods A and goods B is comprehensively evaluated. For the logistics efficiency bottleneck prediction features, the bottleneck position and impact degree that may occur in the logistics link are analyzed, such as the congestion possibility of the transportation route, the storage capacity limitation of the warehouse, etc.

[0055] According to these analysis results, a priority sorting algorithm is used, for example, sorting according to the high and low of the demand urgency and the bottleneck impact degree. If the demand urgency of goods A is higher than that of goods B, and the bottleneck impact degree of the logistics link involved by goods A is relatively low, the priority of goods A is higher than that of goods B. The sorted target goods demand distribution features and logistics efficiency bottleneck prediction features are integrated to generate a resource configuration priority sequence, which clearly defines the priority order of different goods and logistics links in supply chain resource configuration.

[0056] Step 132: Abnormal impact degree evaluation processing is performed on the goods retention risk features and the transportation route deviation features in the supply chain abnormal detection feature set, to generate an abnormal handling operation sequence.

[0057] Optionally, the abnormal impact degree of the goods retention risk features and the transportation route deviation features in the supply chain abnormal detection feature set is evaluated. For the goods retention risk features, the impact degree of goods retention on production progress, delivery deadline, etc. is evaluated. For example, the criticality of the retained goods in the production process is analyzed. If the retained goods are a key raw material for producing a product, and the delivery deadline of the product is approaching, the impact degree of goods retention is high. For the transportation route deviation features, the impact of deviation on transportation cost and transportation time is evaluated, such as deviation leading to a substantial increase in transportation distance and a prolongation of transportation time, etc.

[0058] According to the evaluation results, the cargo retention risk features and the transportation route deviation features are ranked from high to low according to the degree of influence. The abnormal handling operations corresponding to the ranked features are sorted to form an abnormal handling operation sequence, and the execution order of different operations in handling the supply chain abnormal situation is determined.

[0059] Step 133: The resource configuration priority sequence and the abnormal handling operation sequence are dynamically path combined and optimized by using a reinforcement learning decision optimization model to generate an initial multi-stage decision optimization path containing multiple decision sub-paths; wherein each decision sub-path corresponds to an optimized operation combination of a supply chain link.

[0060] The reinforcement learning decision optimization model dynamically combines and optimizes the resource configuration priority sequence and the abnormal handling operation sequence. The model first comprehensively analyzes each resource requirement in the resource configuration priority sequence and each abnormal situation in the abnormal handling operation sequence. For example, for the demand of different goods in the resource configuration priority sequence, combined with the potential problems pointed out by the logistics efficiency bottleneck prediction features, and the abnormal situations such as cargo retention risk and transportation route deviation in the abnormal handling operation sequence, the mutual relationship and influence between them are considered.

[0061] The model generates multiple decision sub-paths by continuously trying different combinations and using its internal strategy network. Taking a supply chain link as an example, the link involves goods transportation and warehouse storage. When generating decision sub-paths, the model will consider the resource configuration priority. If a high-priority cargo needs to be quickly transported to a specific warehouse, and the logistics node where the warehouse is located has an abnormal situation of transportation route deviation, the model will try different combinations of transportation schemes and abnormal handling measures according to the strategy of reinforcement learning.

[0062] For example, one combination may be to prioritize the allocation of a specific transportation tool to transport the cargo, and at the same time, to solve the transportation route deviation problem by temporarily opening an alternative route to ensure that the cargo arrives at the warehouse on time. These operation combinations that consider resource configuration and abnormal handling for different supply chain links constitute a decision sub-path. These decision sub-paths are integrated to generate an initial multi-stage decision optimization path containing multiple decision sub-paths.

[0063] Step 134: According to a pre-set supply chain link weight distribution table, the execution effect of each decision sub-path is simulated and predicted, and the decision sub-path combination with the best prediction effect is selected as the multi-stage decision optimization path.

[0064] Optionally, the preset supply chain link weight distribution table indicates the importance of different supply chain links in the overall supply chain operation. For example, the production link weight is w1, the transportation link weight is w2, the warehousing link weight is w3, and the like. According to this weight distribution table, the execution effect of each decision sub-path is simulated and predicted.

[0065] For each decision sub-path, the simulation and prediction process considers the supply chain links involved in the sub-path and the operation of the links on the overall supply chain target. Taking a decision sub-path as an example, the sub-path involves the transportation of goods from the production place to the warehouse by the transportation link, and the storage arrangement of the goods by the warehousing link. During simulation and prediction, a comprehensive score is calculated according to the weight w2 of the transportation link and the weight w3 of the warehousing link, combined with the transportation cost and transportation time of the transportation scheme in the decision sub-path, and the storage cost and goods loss of the storage scheme.

[0066] For example, through a commonly used evaluation function F, the indicators of the transportation link such as transportation cost C1 and transportation time T1, and the indicators of the warehousing link such as storage cost C2 and goods loss rate L are calculated with the weights of the corresponding links: score S = w2*F1(C1, T1) + w3*F2(C2, L), where F1 and F2 are specific evaluation functions for different indicators of the transportation link and the warehousing link.

[0067] After all the decision sub-paths are calculated with such simulation and prediction, the scores of the decision sub-paths are compared, and the decision sub-path combination with the highest score, i.e. the optimal prediction effect, is selected as the multi-stage decision optimization path. This multi-stage decision optimization path can optimize the overall operation of the supply chain to the greatest extent based on the importance of different supply chain links, and realize the rational allocation of resources and effective handling of abnormal situations.

[0068] Step 140: generating a visualized graphic analysis report according to the multi-stage decision optimization path and the supply chain data set, and feeding back the visualized graphic analysis report to the enterprise management system; wherein the visualized graphic analysis report is used to instruct the enterprise management system to perform supply chain dynamic adjustment operation.

[0069] In a preferred embodiment, step 140 comprises:

[0070] Step 141: converting the resource configuration priority sequence in the multi-stage decision optimization path into a first visualized decision tree structure, and labeling the demand satisfaction rate and resource consumption rate of each resource configuration node in the first visualized decision tree structure.

[0071] First, the resource allocation priority sequence in the multi-stage decision optimization path is structured to transform into a first visual decision tree structure. Starting from the highest priority resource in the resource allocation priority sequence, the resource is taken as the root node of the decision tree. For example, if resource A is ranked first in the resource allocation priority sequence, resource A is the root node.

[0072] For each resource node, the subsequent resource allocation decisions and dependencies are generated to generate child nodes. For example, the allocation of resource A may depend on transportation resource B and warehouse resource C, and transportation resource B and warehouse resource C can be child nodes of resource A. In this way, a complete decision tree structure is gradually constructed.

[0073] Next, the demand satisfaction rate and resource consumption rate of each resource allocation node are calculated and labeled. For the demand satisfaction rate, the ratio of the actual allocation quantity to the expected demand quantity is calculated. For example, for resource D, the expected demand quantity is N1, and the actual allocation quantity is N2, then the demand satisfaction rate = N2 / N1. For the resource consumption rate, the ratio of the actual consumption quantity to the total available quantity in the allocation and use process is calculated. For example, the total available quantity of resource E is M1, and the actual consumption quantity is M2, then the resource consumption rate = M2 / M1. The calculated demand satisfaction rate and resource consumption rate are labeled on the corresponding resource allocation node to form the first visual decision tree structure, which intuitively displays the priority and related indicators of resource allocation.

[0074] Step 142: transforming the abnormal handling operation sequence in the multi-stage decision optimization path into a second visual decision tree structure, and labeling the risk reduction rate and operation execution cost of each abnormal handling node in the second visual decision tree structure.

[0075] Optionally, the abnormal handling operation sequence in the multi-stage decision optimization path is processed to transform into a second visual decision tree structure. The starting point of the abnormal handling is taken as the root node, for example, the handling operation for the risk of goods retention is taken as the root node. According to the flow and subsequent operations of the abnormal handling, the child nodes are generated in turn. If the handling of the risk of goods retention needs to perform inventory checking operation first, and then perform re-allocation of transportation tools operation, the inventory checking operation and the re-allocation of transportation tools operation can be taken as the child nodes of the root node of the risk of goods retention handling.

[0076] For each exception handling node, calculate and label the risk reduction rate and operation execution cost. The risk reduction rate is calculated by the change of risk assessment value before and after processing. For example, before processing the transportation route deviation exception, the risk value caused by the exception to the supply chain is R1, and after processing, the risk value is reduced to R2, then the risk reduction rate = (R1-R2) / R1. The operation execution cost includes the sum of all costs related to the operation execution, such as labor cost, material cost, etc. For example, the labor cost of executing a certain exception handling operation is C3, the material procurement cost is C4, etc., then the operation execution cost = C3+C4. Label the risk reduction rate and operation execution cost on the corresponding exception handling node to form a second visual decision tree structure, clearly presenting the flow and effect related information of exception handling.

[0077] Step 143: Key frame extraction processing is performed on the image form logistics node monitoring data in the supply chain data set, a logistics state change animation is generated, and the logistics state change animation is spatio-temporally aligned with the first and second visual decision tree structures.

[0078] For example, a key frame extraction algorithm is used, which determines key frames according to the degree of change in image content. For example, by comparing the pixel differences between adjacent image frames, object motion states, and other factors. If there is a significant change in the state of goods accumulation between adjacent image frames, or the position of the transportation tool moves greatly, then the frame may be selected as a key frame.

[0079] After extracting the key frames from the image sequence of the logistics node monitoring data, arrange these key frames in chronological order and add appropriate transition effects to generate a logistics state change animation, which can visually display the state change of the logistics node in a period of time, such as the entry and exit of goods, the stay and departure of transportation tools, etc.

[0080] Next, the logistics state change animation is spatio-temporally aligned with the first and second visual decision tree structures. According to the time information, match the time points in the animation with the time-related information involved in the decision tree structure. For example, the execution time of a certain resource allocation operation in the first visual decision tree structure corresponds to the time of goods transportation to the corresponding warehouse in the logistics state change animation; the time of a certain exception handling operation in the second visual decision tree structure is associated with the time of the occurrence of the exception in the logistics state change animation. Through this spatio-temporal alignment, the logistics state change animation and the two visual decision tree structures are closely linked in time and space, facilitating subsequent comprehensive display and analysis.

[0081] Step 144: interactive layout matching processing is performed on the aligned visual decision tree structure and the logistics state change dynamic diagram to generate the visual graphic analysis report containing dynamic data correlation.

[0082] Optionally, the first visual decision tree structure, the second visual decision tree structure and the logistics state change dynamic diagram are subjected to interactive layout matching processing. First, a layout framework is determined, and the two visual decision tree structures and the logistics state change dynamic diagram are arranged reasonably in one display interface. For example, the first visual decision tree structure can be placed on the left side of the interface, the second visual decision tree structure can be placed on the right side of the interface, and the logistics state change dynamic diagram can be placed in the middle position.

[0083] Then, an interactive correlation is established. When a user hovers the mouse over a certain resource configuration node of the first visual decision tree structure, the corresponding logistics operation in the logistics state change dynamic diagram is highlighted, and the abnormal handling node related to the resource configuration in the second visual decision tree structure is also prompted. For example, when the user hovers the mouse over the configuration node of resource A, the transportation process of resource A is displayed in the logistics state change dynamic diagram, and the handling node if an abnormality occurs in the transportation of resource A is displayed in the second visual decision tree structure.

[0084] Conversely, when a user operates a certain abnormal handling node of the second visual decision tree structure, the resource configuration node affected by the abnormality in the first visual decision tree structure and the abnormal occurrence time and the handling process in the logistics state change dynamic diagram are also correlated and displayed. Through this interactive layout matching processing, the visual graphic analysis report containing dynamic data correlation is generated, which provides an intuitive and comprehensive supply chain information display for enterprise managers, facilitating them to make decisions and perform supply chain dynamic adjustment operations.

[0085] In another alternative embodiment, after the visual graphic analysis report is generated, the method further includes:

[0086] Step 145: Monitor the response operation data of the enterprise management system to the visual graphic analysis report, extract resource configuration adjustment data and abnormal handling execution data from the response operation data; compare the resource configuration adjustment data with the resource configuration priority sequence in the multi-stage decision optimization path, generate resource configuration strategy deviation features; evaluate the effect of the abnormal handling execution data and the abnormal handling operation sequence in the multi-stage decision optimization path, generate abnormal handling effect lag features; feed the resource configuration strategy deviation features and abnormal handling effect lag features back to the reinforcement learning decision optimization model, trigger the online strategy adjustment mechanism of the reinforcement learning decision optimization model; update the multi-stage decision optimization path according to the output result of the online strategy adjustment mechanism, and regenerate the visual graphic optimization report containing the updated decision path.

[0087] Optionally, the response operation data of the enterprise management system to the visual graphic analysis report is monitored in real time through commonly used monitoring tools. From these response operation data, resource configuration adjustment data and abnormal handling execution data are accurately extracted. For example, resource configuration adjustment data may include new resource allocation plans, changes in resource priority, etc.; abnormal handling execution data may include actual abnormal handling operations, operation execution time, etc.

[0088] The resource configuration adjustment data is compared with the resource configuration priority sequence in the multi-stage decision optimization path. The difference between the actual resource configuration and the planned resource configuration priority is calculated, for example, the planned resource A priority is higher than resource B, but the actual resource allocation of resource B exceeds resource A. By calculating the degree and direction of this difference, resource configuration strategy deviation features are generated.

[0089] For abnormal handling execution data, the effect is evaluated with the abnormal handling operation sequence in the multi-stage decision optimization path. The risk reduction effect achieved by the actual abnormal handling operation is compared with the expected effect. If the actual risk reduction rate is lower than the expected risk reduction rate, it indicates that there is an abnormal handling effect lag. By quantifying the lag degree, abnormal handling effect lag features are generated.

[0090] The resource configuration strategy deviation features and abnormal handling effect lag features are fed back to the reinforcement learning decision optimization model. After the model receives these feedback information, the online strategy adjustment mechanism is triggered. This mechanism adjusts the strategy network of the model according to the feedback features. For example, if the resource configuration strategy deviation is large, the model will increase the reward weight related to correct resource configuration; if the abnormal handling effect lags, the model will adjust the parameters of the abnormal handling strategy to improve the effect of future abnormal handling.

[0091] According to the output result of the online strategy adjustment mechanism, the multi-stage decision optimization path is updated. For example, the priority order of resource allocation is adjusted, the execution order of exception handling operation is modified, etc. Then, based on the updated multi-stage decision optimization path, a visual text optimization report containing the updated decision path is regenerated, providing more accurate and effective decision support information for the enterprise management system.

[0092] In an implementation manner, the training process of the preset reinforcement learning decision optimization model comprises:

[0093] Step 210: Obtain a historical supply chain data set and corresponding historical decision path execution effect data.

[0094] For example, the historical supply chain data set is obtained from the historical data storage system of the enterprise, which contains various supply chain related information such as order interaction record data, logistics node monitoring data, etc. in the past period of time. At the same time, the historical decision path execution effect data corresponding to these historical data is collected, which records the decision path made in the past for the supply chain situation and the effect evaluation after the actual execution of the decision path. For example, it records whether the resource allocation under a certain decision path is reasonable, whether the abnormal situation is effectively handled, etc. These historical data and execution effect data provide rich samples and reference basis for subsequent model training.

[0095] Step 220: Perform feature enhancement processing on the historical supply chain data set to generate a simulated supply chain trend prediction feature set and a simulated supply chain anomaly detection feature set.

[0096] Optionally, the historical supply chain data set is processed for feature enhancement to enrich the feature information of the data. For order interaction record data, in addition to extracting regular demand keyword distribution features, supplier response timeliness features, etc., potential market trend features, customer preference features, etc. are extracted from order texts through text mining technology. For example, potential customer preferences are mined by analyzing customer descriptions of product characteristics in order texts.

[0097] For logistics node monitoring data, in addition to original cargo accumulation density change features, transportation tool stay time features, etc., image processing technology is used to enhance the detail information of the image, and more detailed logistics operation features are extracted, such as cargo handling mode features, transportation tool loading and unloading efficiency features, etc.

[0098] The enhanced order interaction record data and the logistics node monitoring data are integrated and classified to generate a simulated supply chain trend prediction feature set and a simulated supply chain anomaly detection feature set. The simulated supply chain trend prediction feature set contains various features that can be used to predict future trends of the supply chain, and the simulated supply chain anomaly detection feature set contains various features that can be used to detect abnormal situations of the supply chain. These feature sets provide more comprehensive and valuable data support for model training.

[0099] Step 230: Construct an initial reinforcement learning decision optimization model and set a composite reward function containing resource allocation efficiency rewards and abnormal handling effect rewards.

[0100] The model adopts a network structure suitable for reinforcement learning, such as a deep Q network (DQN) or its variants. The input of the model is the simulated supply chain trend prediction feature set and the simulated supply chain anomaly detection feature set, and the output is the action selection for supply chain decision-making.

[0101] A composite reward function containing resource allocation efficiency rewards and abnormal handling effect rewards is set. For resource allocation efficiency rewards, the rationality and effectiveness of resource allocation are measured. For example, if the resource allocation decision made by the model improves the demand satisfaction rate of resources while reducing resource consumption, a higher reward is given; on the contrary, if the resource allocation is unreasonable, leading to resource waste or unsatisfied demand, a lower reward or punishment is given.

[0102] For abnormal handling effect rewards, the actual effect of abnormal handling is evaluated. If the abnormal handling operation successfully reduces the risk and the operation execution cost is within a reasonable range, the corresponding reward is given; if the abnormal handling fails to effectively reduce the risk or the cost is too high, negative feedback is given. The composite reward function integrates the rewards of these two aspects, so that the model can consider the effects of resource allocation and abnormal handling during the training process to learn a more optimized decision strategy.

[0103] Step 240: Input the simulated supply chain trend prediction feature set and the simulated supply chain anomaly detection feature set into the initial reinforcement learning decision optimization model, and generate the reinforcement learning decision optimization model that meets the preset decision effect through multiple rounds of alternating training of strategy exploration and strategy utilization. The strategy exploration stage includes random attempts on decision sub-paths that have not appeared in historical decision path execution effect data, and the strategy utilization stage includes selecting the historical optimal decision sub-path for reinforcement learning according to the feedback value of the composite reward function.

[0104] For example, the simulated supply chain trend prediction feature set and the simulated supply chain anomaly detection feature set are input into an initial reinforcement learning decision optimization model for training. During the training process, multiple rounds of alternating training of strategy exploration and strategy utilization are performed.

[0105] In the strategy exploration phase, the model randomly tries some decision sub-paths that have not appeared in the historical decision path execution effect data. For example, for resource allocation decisions, a completely new resource allocation combination method can be tried; for anomaly handling decisions, a different anomaly handling process can be explored. Through such random attempts, the model can discover new possible effective decision methods and expand the decision space.

[0106] In the strategy utilization phase, the model selects the historical optimal decision sub-path for reinforcement learning according to the feedback value of the composite reward function. If a certain decision sub-path has obtained a higher reward in the past execution, the model will increase the probability of selecting this decision sub-path, further strengthening this successful decision strategy.

[0107] Through multiple rounds of such alternating training, the model continuously adjusts its strategy network parameters to adapt to different supply chain situations and gradually learns more optimized decision strategies. When the decision effect of the model meets the preset standards, such as the resource allocation efficiency reaching a certain level, the anomaly handling success rate reaching a certain proportion, etc., a reinforcement learning decision optimization model that meets the preset decision effect is generated.

[0108] In an independent embodiment, after the visual graphic analysis report is fed back to the enterprise management system, it further includes: performing key area enhancement processing on the logistics state change animation in the visual graphic analysis report, extracting frame images with cargo accumulation density exceeding the threshold value in the animation, and generating a highlighted marked area; performing spatial coordinate mapping processing on the highlighted marked area and the resource allocation node in the first visual decision tree structure, generating an enhanced decision tree layer with a heat map superimposed; performing dynamic transparency adjustment processing on the enhanced decision tree layer, dynamically adjusting the transparency parameter of the corresponding heat map according to the priority of the resource allocation node; performing frame synchronization fusion processing on the adjusted enhanced decision tree layer and the logistics state change animation, generating a dynamic interactive enhanced report and updating it to the enterprise management system.

[0109] In this embodiment, the key area enhancement processing on the logistics state change animation in the visual graphic analysis report first sets a threshold value for cargo accumulation density, and calculates the cargo accumulation density by analyzing each frame image of the animation. For example, the cargo accumulation density is represented by the ratio of the number of pixels occupied by the cargo in the image to the total number of pixels in the image using image pixel analysis technology. When the cargo accumulation density of a certain frame image exceeds the set threshold value, the frame image is extracted, and these images are collected to generate a highlighted marked area.

[0110] The highlighted area is mapped to the resource configuration node in the first visual decision tree structure in spatial coordinates. First, determine the spatial correspondence between the logistics scene in the moving picture and the resource configuration scene represented by the first visual decision tree structure. For example, the logistics warehouse in the moving picture has a commonly used location identifier in the actual space, and the resource configuration node in the first visual decision tree structure also has a logical association with the warehouse.

[0111] By establishing this spatial correspondence, each pixel position in the highlighted area is mapped to the spatial position of the resource configuration node. For example, if a certain position in the highlighted area represents a specific area of goods accumulation in the warehouse, map this position to the resource configuration node related to the warehouse in the first visual decision tree structure, determine the coordinate correspondence between them, and generate an enhanced decision tree layer with heat map superimposition.

[0112] Next, the enhanced decision tree layer is processed for dynamic transparency adjustment. The transparency parameter of the corresponding heat map is dynamically adjusted according to the priority of the resource configuration node. For example, for a resource configuration node with high priority, set its corresponding heat map transparency to be higher, so that it is more prominently displayed in the layer; for a resource configuration node with low priority, reduce the transparency of its heat map accordingly. The specific implementation can be realized through a transparency adjustment function, which takes the priority of the resource configuration node as the input parameter and outputs the corresponding transparency value. For example, set the priority range to [1, 10], the higher the priority value, the greater the value, and the transparency adjustment function is: transparency = priority / 10 (value range between 0 and 1), so that the transparency of the heat map can be dynamically adjusted according to different priorities.

[0113] Finally, the adjusted enhanced decision tree layer is fused with the logistics state change moving picture in frame synchronization. In the fusion process, ensure that the time information of the enhanced decision tree layer and the time information of the logistics state change moving picture are accurately matched. For example, when the logistics state change moving picture plays to a certain time, the corresponding enhanced decision tree layer also displays the resource configuration and heat map information related to that time. Through this frame synchronization fusion processing, a dynamic interactive enhanced report is generated and updated to the enterprise management system. When viewing the visual text analysis report, enterprise management personnel can more intuitively understand the relationship between the goods accumulation density abnormal area and resource configuration, as well as the dynamic information changing over time, so as to make more accurate decisions.

[0114] In an independent embodiment, after the visual graphic analysis report is fed back to the enterprise management system, it further includes: identifying supply chain anomaly detection feature keywords in the text description section of the visual graphic analysis report, generating anomaly location annotation boxes associated with each supply chain anomaly detection feature keyword; performing semantic matching processing on the anomaly location annotation boxes and the anomaly handling nodes of the second visual decision tree structure, generating an anomaly handling topology graph with multiple annotation boxes; performing layout conflict detection processing on the multiple annotation boxes, automatically adjusting annotation hierarchy distribution and folding and unfolding trigger conditions according to annotation box overlapping areas, to obtain an adjusted anomaly handling topology graph; performing cross-layer association processing on the adjusted anomaly handling topology graph and the order interaction semantic feature set, generating an interactive report supporting click-to-display semantic association links and returning for update.

[0115] First, the text description section in the visual graphic analysis report is subjected to keyword identification. Through natural language processing technology, morphological analysis, syntactic analysis and other methods, keywords related to supply chain anomaly detection features are identified. For example, keywords such as "goods retention", "transportation route deviation", "contract clause conflict" are extracted from the text. For each keyword, an anomaly location annotation box associated with it is generated. The content of the annotation box can include a brief description of the anomaly detection feature represented by the keyword and the location information where the anomaly may occur.

[0116] Then, the anomaly location annotation boxes are subjected to semantic matching processing with the anomaly handling nodes of the second visual decision tree structure. Through semantic analysis algorithms, the content in the annotation boxes is compared with the semantic information of the anomaly handling nodes to determine their relevance. For example, if an anomaly location annotation box mentions "goods retention risk in warehouse A", and there is an anomaly handling node in the second visual decision tree structure that addresses the goods retention problem in warehouse A, the two are matched. Through this matching, an anomaly handling topology graph with multiple annotation boxes is generated, making the anomaly handling process and related anomaly location information more clear and intuitive.

[0117] Next, the multiple annotation boxes are subjected to layout conflict detection processing. By calculating the positional relationship and overlapping area between the annotation boxes, it is determined whether there is a layout conflict. For example, if the positions of two annotation boxes overlap, causing the information to be displayed unclearly, it is considered that there is a layout conflict. According to the detected conflict situation, the annotation hierarchy distribution and folding and unfolding trigger conditions are automatically adjusted. For example, if a certain annotation box is covered by other annotation boxes, it is adjusted to a higher level of display; at the same time, folding and unfolding trigger conditions are set, when a user's mouse hovers over a certain annotation box, the detailed information related to it is unfolded. Through these adjustments, an adjusted anomaly handling topology graph is obtained, improving the readability and display effect of the annotation information.

[0118] Finally, the adjusted abnormal handling topology graph is cross-layer associated with the order interaction semantic feature set. By establishing semantic association rules, the abnormal information in the abnormal handling topology graph is associated with the related semantic information in the order interaction semantic feature set. For example, if a certain abnormal handling node in the abnormal handling topology graph involves "contract clause conflict", the relevant contract clause conflict semantic feature information in the order interaction semantic feature set is searched and linked. An interactive report supporting click-to-display semantic association links is generated. When a user clicks on a certain node in the abnormal handling topology graph, the relevant order interaction semantic information can be viewed to further understand the root cause and impact of the abnormality. This interactive report is updated back to the enterprise management system, providing enterprise management personnel with more comprehensive and in-depth supply chain abnormality analysis information, which helps them better develop response strategies.

[0119] It should be noted that, when implementing the above technical solutions, a person skilled in the art can perform dynamic alignment processing on the heterogeneous timestamps of the logistics node monitoring data and the order data based on the time series interpolation algorithm (such as cubic spline interpolation) in the prior art, thereby solving the time synchronization problem of data with different sampling frequencies.

[0120] For the attention weight distribution mechanism, the multi-head attention (Multi-head Attention) technology can be used to jointly model the time matching degree and feature importance. A feature correlation matrix is constructed by calculating the mutual information amount of the logistics dynamic features and the order semantic features, thereby optimizing the dynamic weight distribution strategy of cross-modal features.

[0121] For the abnormality detection threshold setting, an adaptive threshold calculation method based on historical data (such as the 3σ principle) can be introduced. The sliding window is used to calculate the average and variance of the cargo accumulation density of each logistics node. By dynamically adjusting the threshold boundary, differentiated abnormality determination for nodes of different scales is achieved.

[0122] In terms of multi-model interface dimension matching, feature embedding technology (such as the BERT pre-training model) can be used to vectorize and encode text features. Convolutional neural networks (CNN) are used to reduce the spatial dimension of image features. The feature vector dimensions are unified through a fully connected layer to achieve the compatibility of cross-modal data input.

[0123] For the dimensional conversion problem of cargo bulk density, a mapping relationship table of logistics monitoring image pixel distribution and actual volume of storage space can be established based on the FastDTW algorithm, the pixel ratio-cubic volume conversion coefficient of a typical cargo stacking scene is measured through calibration experiment, and the quantitative conversion from pixel feature to physical density is realized; for time unit standardization, the millisecond level synchronization calibration of the clock source of the order system and the logistics equipment can be realized by using the UNIX timestamp mechanism, and the dimensional difference of the time record accuracy of different business systems can be eliminated by Z-score standardization; in terms of cost calculation unit unification, the manual labor hours involved in abnormal processing operation can be converted into monetary units in real time according to the post salary standard by integrating the human cost accounting module of the enterprise ERP system, and the normalization processing of multi-dimensional cost indicators can be realized by linear weighting.

[0124] For the quantitative improvement of light intensity, the V channel brightness value of the monitoring image in the HSV color space can be extracted based on the OpenCV image processing library, a brightness-illumination regression model can be constructed by combining the photometric sensor calibration data, and the accurate mapping of pixel brightness value and physical illumination unit (Lux) can be realized.

[0125] In the dynamic decision path generation link, the reinforcement curriculum learning (Curriculum Reinforcement Learning) technology can be introduced, the training strategy of gradually increasing the complexity of the environment can be used to improve the exploration efficiency of the model for multi-stage decision combination, and the Monte Carlo tree search (MCTS) algorithm can be combined to perform game deduction on the decision sub-path, so as to optimize the cooperative strategy of resource allocation and abnormal processing.

[0126] Through the combination application of the above existing technologies, the logic and dimension of the embodiments of the present application can be systematically optimized, and the whole process technical closed loop from the collection, feature extraction to the decision generation of the supply chain data can be ensured.

[0127] In summary, the embodiments of the present application can generate visual text analysis reports based on comprehensive and diversified supply chain data by integrating dynamic decision logic, which can improve the comprehensiveness, dynamics and decision guidance of the generated text, thereby effectively guiding the supply chain dynamic adjustment of the enterprise management system.

[0128] In detail, the embodiment of the present application provides a comprehensive and diverse data cornerstone for the generation of graphs and texts by collecting supply chain data sets containing texts and image forms; the supply chain trend prediction feature set and the supply chain anomaly detection feature set obtained through feature extraction inject depth and pertinence into the graph and text content; the multi-stage decision optimization path generated based on the reinforcement learning decision optimization model further provides a clear logical framework for the generation of graphs and texts, ensuring that the generated visual graph analysis report is not a simple data list, but a deep analysis result with logical coherence and decision-oriented. The visual graph analysis report not only intuitively presents the supply chain status, but also indicates the enterprise management system to perform supply chain dynamic adjustment operations according to the multi-stage decision path in the form of clear and understandable graphs and texts, thereby improving the visualization and intelligent level of enterprise decision-making.

[0129] Based on the same inventive concept, the embodiment of the present application also provides a digital enterprise management data analysis system. Referring to Figure 2 , which is a possible structure of a digital enterprise management data analysis system provided by the embodiment of the present application, Figure 2 , the digital enterprise management data analysis system 200 comprises a processor 210 and a memory 220. The memory 220 stores a computer program executable by the processor 210, and the processor 210 can execute the steps of the above-mentioned artificial intelligence-based digital enterprise management data analysis method by executing the instructions stored in the memory 220.

[0130] Based on the same inventive concept, the embodiment of the present application provides a computer readable storage medium comprising a computer program, when the computer program runs on the digital enterprise management data analysis system, the computer program is used to make the digital enterprise management data analysis system execute the steps of the above-mentioned artificial intelligence-based digital enterprise management data analysis method. In some possible embodiments, various aspects of the artificial intelligence-based digital enterprise management data analysis method provided by the present application can also be implemented in the form of a program product, which comprises a computer program, when the program product runs on the digital enterprise management data analysis system, the computer program is used to make the digital enterprise management data analysis system execute the steps of the above-mentioned artificial intelligence-based digital enterprise management data analysis method, for example, the digital enterprise management data analysis system can execute the steps as shown in Figure 1 .

[0131] In the technical solutions related to the above embodiments of the present application, whether it is a multi-dimensional feature comparison calculation or a composite parameter construction, if there are problems caused by significant differences in the number of dimensions, units of dimensions, and semantic meanings of different features, those skilled in the art can fully understand that these differences need to be properly handled based on their professional knowledge and past practical experience, so that the calculation results are accurate and comparable, and logical confusion, unclear mathematical meaning, and other conditions are avoided.

[0132] In detail, when facing features with different numbers of dimensions, in order to accurately calculate the similarity, matching degree, or feature distance between different features, those skilled in the art can use various strategies.

[0133] Feature selection is a commonly used method. For a high-dimensional feature set, a feature subset that is most representative and matches the number of low-dimensional features can be selected from the high-dimensional features according to indicators such as feature importance and correlation. By using methods such as chi-square test and information gain for feature selection, the most valuable features for the technical solution are screened out, thereby reducing the high-dimensional features to a dimension comparable to the low-dimensional features, and then calculating the similarity or distance.

[0134] Feature extraction is also an effective means. By constructing a suitable feature extraction model, features of different dimensions can be mapped to a common low-dimensional feature space. Principal component analysis (PCA) can not only be used to handle dimensional differences, but also project high-dimensional features to a low-dimensional space composed of principal components, so that features of different dimensions are comparable in this low-dimensional space. In addition, deep learning models such as autoencoders can also be used for feature extraction, which can automatically learn the latent representation of input features and convert features of different dimensions into feature vectors of the same dimension for subsequent similarity, matching degree, or feature distance calculation.

[0135] In addition, kernel methods can also be used. Kernel functions can calculate the similarity between features in high-dimensional space without explicitly mapping features to high-dimensional space. For features with different numbers of dimensions, appropriate kernel functions such as Gaussian kernel functions and polynomial kernel functions can be selected to directly calculate their similarity. This method avoids the difficulty of direct calculation caused by different feature dimensions and can effectively measure the relationship between features in the original feature space or an implicit high-dimensional space.

[0136] When dealing with multi-dimensional feature comparison, in order to achieve the comparability alignment of the feature space, those skilled in the art can use various existing general technical means.

[0137] Standardization preprocessing is a widely used and effective method that transforms the original feature data into a standard normal distribution with a mean of 0 and a standard deviation of 1 through a specific linear transformation. This processing method can essentially eliminate the influence of the dimensions of different features, allowing all features to be compared on the same scale. For example, in a dataset containing features with different dimensions, after standardization preprocessing, these features can be calculated for similarity or distance on the same scale, avoiding calculation bias caused by different dimensions.

[0138] Mapping transformation is also an effective way to solve the problem of dimension difference. It can map the original features to a new space according to the specific properties of the features and actual business needs. In this new space, different dimensions of features can have better comparability. For features with nonlinear relationships, those skilled in the art can use logarithmic transformation, power transformation, etc. to convert them to linear relationships, making it easier to calculate similarity or distance. For example, when dealing with features with exponential growth trends, logarithmic transformation can convert them to linear relationships, making subsequent calculations more accurate and convenient.

[0139] Space projection is also an important technical means, which projects high-dimensional feature space into low-dimensional space while preserving important information between features as much as possible. By carefully selecting the projection direction and projection dimension, those skilled in the art can reduce the influence of dimension difference on the calculation results while reducing the data dimension. Common space projection methods include principal component analysis (PCA), linear discriminant analysis (LDA), etc. Taking principal component analysis as an example, it projects high-dimensional data into a low-dimensional space composed of principal components, simplifying the data structure while reducing the interference of dimension difference on feature comparison.

[0140] In the construction process of composite parameters (such as loss function values), different parameter items often have different dimensions. Those skilled in the art can use normalization processing or adaptive weight distribution mechanism based on distribution characteristics.

[0141] Normalization processing is to unify the value range of different parameter items to a fixed interval, such as [0, 1]. This processing method can eliminate the influence of dimension difference and ensure that each parameter item has the same importance when weighted fusion. Common normalization methods include min-max normalization, Z-score normalization, etc. Taking min-max normalization as an example, it performs linear transformation on the parameter items to scale their value range to the [0, 1] interval, so that parameter items with different dimensions can be weighted and fused on the same standard.

[0142] The adaptive weight distribution mechanism based on distribution characteristics dynamically adjusts the weights of different parameter items according to their distribution characteristics. For parameter items with larger variances, the weight of the parameter item can be appropriately reduced by those skilled in the art; for parameter items with smaller variances, the weight of the parameter item can be appropriately increased by those skilled in the art. In this way, the composite loss function can pay more attention to the parameter items with smaller variances, thereby improving the stability and generalization ability of the model. For example, in a composite loss function containing multiple parameter items, if the variance of a certain parameter item is large, it means that the fluctuation of the parameter item is large, which may adversely affect the stability of the model. At this time, reducing the weight of the parameter item can reduce the adverse effect; for parameter items with smaller variances, increasing the weight of the parameter item can make the model pay more attention to the information reflected by the parameter item, thereby improving the overall performance of the model.

[0143] The above-mentioned general technical means for solving the problems of feature matching and loss balancing belong to the common knowledge in the art. These technical means have been fully verified and widely used in a large number of practical applications. When facing similar dimensional difference problems, those skilled in the art can skillfully and flexibly use these methods to solve the problems.

[0144] The formulas and calculation processes involved in the embodiments of the present application, whether for multi-dimensional feature comparison or composite loss function construction, strictly follow the dimensional correspondence principle. The variables in each formula have a clear and explicit physical meaning, and the operation logic is completely consistent with the basic mathematical and physical logic. The operation result is necessarily a reasonable result expected by the present application. Those skilled in the art can effectively solve various problems caused by the number of dimensions, dimensional differences, etc. in the multi-dimensional feature comparison calculation and the composite loss function construction according to the specific data situation and business requirements by comprehensively using the above-mentioned general technical means, and ensure the accuracy, reliability and implementability of the technical solutions of the present application.

Claims

1. An artificial intelligence-based digital enterprise management data analysis method, characterized in that, The method comprises the following steps: Collecting a supply chain data set of a target enterprise; The supply chain data set includes order interaction record data in text form and logistics node monitoring data in image form; Feature extraction is performed on the supply chain data set to obtain a supply chain trend prediction feature set and a supply chain anomaly detection feature set; Based on a preset reinforcement learning decision optimization model, the supply chain trend prediction feature set and the supply chain anomaly detection feature set are subjected to dynamic decision path generation processing to generate a multi-stage decision optimization path: the target goods demand distribution features and logistics efficiency bottleneck prediction features in the supply chain trend prediction feature set are subjected to decision priority sorting processing to generate a resource configuration priority sequence; The goods stagnation risk features and transportation route deviation features in the supply chain anomaly detection feature set are subjected to anomaly influence degree evaluation processing to generate an anomaly handling operation sequence; the reinforcement learning decision optimization model is used to perform dynamic path combination optimization processing on the resource configuration priority sequence and the anomaly handling operation sequence to generate an initial multi-stage decision optimization path comprising multiple decision sub-paths; each decision sub-path corresponds to an optimization operation combination of a supply chain link; the execution effect of each decision sub-path is simulated and predicted according to a preset supply chain link weight distribution table, and the decision sub-path combination with the optimal prediction effect is selected as the multi-stage decision optimization path; A visual graphic analysis report is generated based on the multi-stage decision optimization path and the supply chain data set, and the visual graphic analysis report is fed back to an enterprise management system; the visual graphic analysis report is used to instruct the enterprise management system to perform supply chain dynamic adjustment operations.

2. The method of claim 1, wherein, The feature extraction on the supply chain data set to obtain the supply chain trend prediction feature set and the supply chain anomaly detection feature set comprises: Temporal image feature extraction processing is performed on the image form logistics node monitoring data to extract the goods accumulation density change features, transportation tool stay time features and environmental light intensity fluctuation features of each logistics node to generate a logistics node dynamic feature set; Semantic association analysis processing is performed on the text form order interaction record data to extract the demand keyword distribution features, supplier response timeliness features and contract clause conflict semantic features in the order text to generate an order interaction semantic feature set; A pre-trained supply chain trend prediction model is called to perform cross-modal feature fusion processing on the logistics node dynamic feature set and the order interaction semantic feature set to generate the supply chain trend prediction feature set; An anomaly detection convolutional network is used to perform local anomaly pattern recognition processing on the logistics node dynamic feature set to generate the supply chain anomaly detection feature set; the supply chain anomaly detection feature set includes goods stagnation risk features, transportation route deviation features and order matching anomaly features.

3. The method of claim 2, wherein, The calling of the pre-trained supply chain trend prediction model, the cross-modal feature fusion processing on the logistics node dynamic feature set and the order interaction semantic feature set, and the generation of the supply chain trend prediction feature set comprise: aligning the cargo accumulation density change feature and the transportation tool stay duration feature in the logistics node dynamic feature set in time to generate a first time sequence feature vector; performing semantic coding processing on the demand keyword distribution feature and the supplier response timeliness feature in the order interaction semantic feature set to generate a second time sequence feature vector; performing dynamic weight adjustment processing on the first time sequence feature vector and the second time sequence feature vector by using an attention weight distribution mechanism to generate a fused cross-modal time sequence feature vector; wherein the dynamic weight adjustment processing includes assigning different attention weight coefficients according to the matching degree of the time stamp of the logistics node monitoring data and the time stamp of the order interaction record; inputting the cross-modal time sequence feature vector into the supply chain trend prediction model, extracting trend change patterns of different time granularities by a multi-layer time convolution network to generate the supply chain trend prediction feature set.

4. The method of claim 2, wherein, The local anomaly pattern recognition processing of the logistics node dynamic feature set by the anomaly detection convolution network to generate the supply chain anomaly detection feature set includes: performing illumination anomaly detection processing on the environmental illumination intensity fluctuation feature in the logistics node dynamic feature set to extract the occurrence time and duration of the illumination intensity mutation event to generate an illumination anomaly event feature; performing density gradient analysis processing on the cargo accumulation density change feature to identify abnormal accumulation areas with a density change rate exceeding a preset threshold to generate a cargo retention risk feature; performing stay mode clustering processing on the transportation tool stay duration feature to mark transportation tools deviating from the preset stay duration as route deviation candidates to generate a transportation route deviation feature; performing association matching processing on the illumination anomaly event feature, the cargo retention risk feature, and the transportation route deviation feature with the contract clause conflict semantic feature in the order interaction semantic feature set to generate the supply chain anomaly detection feature set.

5. The method of claim 4, wherein, After generating the supply chain anomaly detection feature set, the method further includes: inputting the cargo retention risk feature and the transportation route deviation feature in the supply chain anomaly detection feature set into the reinforcement learning decision optimization model to trigger the anomaly feedback learning mechanism of the reinforcement learning decision optimization model; dynamically adjusting the reward and punishment function weight in the anomaly feedback learning mechanism according to the historical processing records of the cargo retention risk feature and the severity of the current transportation route deviation feature; incrementally training the policy network of the reinforcement learning decision optimization model based on the adjusted reward and punishment function weight to generate an updated reinforcement learning decision optimization model; applying the updated reinforcement learning decision optimization model to the processing of subsequent supply chain anomaly detection feature sets to realize dynamic adaptability adjustment of the multi-stage decision optimization path.

6. The method of claim 1, wherein, The method of generating a visual text analysis report according to the multi-stage decision optimization path and the supply chain data set includes: converting the resource configuration priority sequence in the multi-stage decision optimization path into a first visual decision tree structure, and labeling the demand satisfaction rate and resource consumption rate of each resource configuration node in the first visual decision tree structure; transforming the sequence of abnormal handling operations in the multi-stage decision optimization path into a second visual decision tree structure, and marking the risk reduction rate and operation execution cost of each abnormal handling node in the second visual decision tree structure; performing key frame extraction processing on the image-form logistics node monitoring data in the supply chain dataset to generate a logistics state change animation, and performing spatio-temporal alignment processing on the logistics state change animation, the first visual decision tree structure, and the second visual decision tree structure; performing interactive layout matching processing on the aligned visual decision tree structure and the logistics state change animation to generate the visual graphic analysis report containing dynamic data correlation.

7. The method of claim 6, wherein, After the visual graphic analysis report is generated, the method further includes: monitoring response operation data of the enterprise management system to the visual graphic analysis report, and extracting resource configuration adjustment data and abnormal handling execution data from the response operation data; performing difference comparison processing on the resource configuration adjustment data and the resource configuration priority sequence in the multi-stage decision optimization path to generate resource configuration strategy deviation features; performing effect evaluation processing on the abnormal handling execution data and the sequence of abnormal handling operations in the multi-stage decision optimization path to generate abnormal handling effect lag features; feeding the resource configuration strategy deviation features and the abnormal handling effect lag features back to the reinforcement learning decision optimization model to trigger an online strategy adjustment mechanism of the reinforcement learning decision optimization model; updating the multi-stage decision optimization path according to the output result of the online strategy adjustment mechanism, and regenerating a visual graphic optimization report containing the updated decision path.

8. The method of claim 1, wherein, The training process of the preset reinforcement learning decision optimization model includes: obtaining a historical supply chain dataset and corresponding historical decision path execution effect data; performing feature enhancement processing on the historical supply chain dataset to generate a simulated supply chain trend prediction feature set and a simulated supply chain anomaly detection feature set; constructing an initial reinforcement learning decision optimization model, and setting a composite reward function containing resource configuration efficiency rewards and abnormal handling effect rewards; inputting the simulated supply chain trend prediction feature set and the simulated supply chain anomaly detection feature set into the initial reinforcement learning decision optimization model, and generating the reinforcement learning decision optimization model meeting the preset decision effect through multi-round strategy exploration and strategy utilization alternation training; wherein, the strategy exploration stage includes randomly trying a decision sub-path that does not appear in the historical decision path execution effect data, and the strategy utilization stage includes selecting a historical optimal decision sub-path for reinforcement learning according to the feedback value of the composite reward function.

9. A digital enterprise management data analysis system, characterized by, It includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method of any one of claims 1-8. It includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method of any one of claims 1-8.

Images