A full-link intelligent management and control platform for enterprises driven by multimodal large models

Abstract

The present invention relates to the field of data management technology, and discloses a full-link intelligent management and control platform for an enterprise driven by a multimodal large model, including the following steps: collecting multimodal data from different departments within the enterprise; mapping the modal data collected by different departments to a unified feature vector, and sequentially obtaining the modal features of the feature vector through convolution kernel C1, sampling pool S2, and convolution kernel C3; establishing a feature database for each department based on different departments and their corresponding modal features; judging whether the modal features of the data of each department of the enterprise and the modal features of other parts of the data intersect based on whether the modal features fall into the feature databases of multiple departments at the same time. The present invention can effectively capture the potential correlation of data from different departments by utilizing convolution kernels and sampling pools to extract periodicity and trend features, and accurately identify data intersections in combination with cross-validation of feature databases, breaking the traditional data isolation between departments.

Description

Technical Field

[0001] The present invention relates to the field of data management technology, and in particular to an enterprise full-link intelligent management and control platform driven by a multimodal large model. Background Art

[0002] In today's digital age, the management and processing of enterprise data has become crucial. Companies need to be able to efficiently process large amounts of data to make informed decisions, improve productivity, and maintain a competitive advantage. To meet this demand, the research and development of enterprise digital transformation management methods has become crucial.

[0003] This application solves an existing patent for a method for managing enterprise digital transformation (publication number CN118070342A), which involves the field of data management. The method includes: Step 1: obtaining enterprise data and dividing the enterprise data into multiple sub-enterprise data according to the organizational structure within the enterprise, where each sub-enterprise data corresponds to a minimum organizational structure within the enterprise; Step 2: compressing and encapsulating each sub-enterprise data; Step 3: submitting an enterprise data acquisition application to the minimum organizational structure within the enterprise corresponding to the sub-enterprise data corresponding to the encapsulated enterprise data. The following problems may arise during the application:

[0004] 1) It is necessary to ensure that the data coding sets between each department within the enterprise remain unique and have no overlap. If the enterprise organizational structure changes frequently, the maintenance cost of the coding system will increase significantly;

[0005] 2) There is a lack of signatures for the flow of data between departments within the enterprise, which makes it difficult to intuitively trace the origin of frequently changing data. Summary of the Invention

[0006] In order to solve existing technical problems, the present invention provides an enterprise full-link intelligent management and control platform driven by a multimodal large model, which solves the problems in the above-mentioned background technology.

[0007] To solve the above technical problems, according to one aspect of the present invention, more specifically, a multimodal large model-driven enterprise full-link intelligent management and control platform, the following steps are included:

[0008] S1. Collect multimodal data from different departments within the enterprise;

[0009] S2 maps the modal data collected by different departments to a unified feature vector, and obtains the modal features of the feature vector through convolution kernel C1, sampling pool S2, and convolution kernel C3 in sequence;

[0010] S3. Establish a feature database for each department based on different departments and their corresponding modal features;

[0011] S4. Based on whether the modal feature falls into the feature databases of multiple departments at the same time, it is determined whether the modal features of the data of each department of the enterprise have an intersection with the modal features of other parts of the data;

[0012] S5, and fitting and generating a data label judgment model based on the modal features;

[0013] S6. Calculating the difference coefficient between the modal feature and the modal features in the feature database of each department through the data label judgment model;

[0014] S7. Based on the difference coefficient and the set threshold, determine whether the modal feature of the data of each department of the enterprise is unique. If the modal feature is unique, assign the initial department signature and loop feedback.

[0015] Furthermore, the multimodal data includes structured data, text, images, and time series data.

[0016] Furthermore, the specific steps of obtaining the modal features in step S2 are:

[0017] S201, scanning the convolution kernel C1 on the feature vector, and capturing the periodic feature m1 of the feature vector by calculating the weighted sum of the local area;

[0018] S202, after the periodic feature m1 is pooled through the sampling pool S2, the overall trend of the periodic feature m1 is extracted;

[0019] S203, scanning the convolution kernel C3 on the overall trend to obtain the same trend feature m2 in the overall trend;

[0020] S204: Establish based on the periodic feature m1 and the same trend feature m2.

[0021] Furthermore, when the periodic feature m1 and the same trend feature m2 in the acquired modal feature exist at the same time, it means that the feature vector exists.

[0022] Furthermore, based on the ratio of the obtained trend feature m2 in the periodic feature m1 and the ratio of the trend feature m2 to the periodic feature m1 in the feature databases of other departments, it is determined whether the modal feature also falls into the feature databases of other departments.

[0023] Furthermore, in step S5, the specific steps of fitting and generating the data label judgment model are:

[0024] S501, manually setting a discrete difference coefficient based on the degree of matching between the modal feature and the modal features in multiple other department feature databases;

[0025] S502, establishing a feature relationship x based on the difference coefficient and the ratio of the trend feature m2 in the acquired modal feature to the periodic feature m1 and the ratio of the trend feature m2 to the periodic feature m1 in the feature databases of other departments;

[0026] S503, establishing a feature relationship y based on the difference coefficient and the ratio of the trend feature m2 in the acquired modal feature to the trend feature m2 in the feature database of other departments;

[0027] S504: Based on the feature relationship x and the feature relationship y, a data label judgment model is fitted and established.

[0028] Furthermore, the data label judgment model calculates the difference coefficient between the modal feature and the modal feature in each department feature database, and the calculation formula is:

[0029] ;

[0030] Where g represents the difference coefficient between the modal feature and the modal feature in the feature database of each department; a represents the ratio of the ratio of the trend feature m2 in the obtained modal feature to the periodic feature m1 to the ratio of the trend feature m2 in the feature database of other departments to the periodic feature m1; h represents the ratio of the trend feature m2 in the obtained modal feature to the trend feature m2 in the feature database of other departments.

[0031] Furthermore, the enterprise full-link intelligent management and control platform driven by multimodal large models also includes a dynamic update module, which is used to regularly update the feature database of each department based on newly collected multimodal data, and adjust the data label judgment model through incremental learning algorithms to adapt to the dynamic changes of enterprise data.

[0032] Furthermore, the enterprise full-link intelligent management and control platform driven by a multimodal large model also includes a cross-departmental collaborative optimization module. When it is determined in step S4 that there is an intersection of modal features, a data fusion strategy is generated based on the difference coefficient, and an automated workflow is triggered to coordinate multiple departments to perform joint tasks.

[0033] The multimodal large model-driven enterprise full-link intelligent management and control platform provided by this invention has the following advantages compared with the existing technology:

[0034] 1. By mapping multimodal data to a unified feature vector and using convolution kernels and sampling pools to extract periodicity and trend features, the present invention can effectively capture the potential correlation of data from different departments. In addition, combined with cross-validation of feature databases, it can accurately identify data intersections and break the traditional data isolation between departments.

[0035] 2. The present invention regularly adjusts the feature database and data label judgment model through an incremental learning algorithm, supporting rapid adaptation to newly collected data. This mechanism avoids data drift problems caused by business expansion or process adjustments, ensuring the long-term effectiveness of the model, while reducing the computational cost of full retraining and improving system efficiency and foresight.

[0036] 3. When the intersection of modal features is detected, the present invention will automatically generate a data fusion strategy based on the difference coefficient and trigger a joint task workflow, such as coordinating sales and production departments to optimize inventory management, shorten decision response time, reduce manual intervention errors, and achieve efficient resource allocation and cost savings.

[0037] 4. Through difference coefficient calculation and threshold determination, the present invention can verify the uniqueness of data modal characteristics and assign an initial department signature to the unique existing data. This mechanism not only enhances the transparency of data ownership, but also supports full-link traceability, providing technical support for enterprise data governance and compliance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of the present invention;

[0039] Figure 2 Schematic diagram of the difference coefficient g and the ratio a of the acquired data to the database data in the present invention;

[0040] Figure 3 Schematic diagram of the difference coefficient g and the ratio h of the acquired data to the database data in the present invention;

[0041] Figure 4 Schematic diagram of the ratio a of database data and the ratio h of acquired data to database data in the present invention;

[0042] Figure 5 Schematic diagram of characteristic vectors in the present invention;

[0043] Figure 6 Schematic diagram of the periodic feature m1 in the present invention;

[0044] Figure 7 This is a schematic diagram of trend feature m2 in the present invention. DETAILED DESCRIPTION

[0045] In order to make the technical solution of the present invention clearer, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Example 1

[0047] like Figure 1 、 Figure 5-Figure 7As shown, according to one aspect of the present invention, a full-link intelligent management and control platform for enterprises driven by a multimodal large model is provided, which includes collecting multimodal data from different departments within the enterprise. The multimodal data includes structured data, text, images, and time series data; and mapping the modal data collected by different departments into a unified feature vector. The specific steps for obtaining the unified feature vector are:

[0048] The following is a mapping formula for the text modality, with a detailed explanation of the definition, function, and training method of each parameter. The parameter logic for image and time series modalities is similar to that for text modalities, requiring only the replacement of the input features and independent parameters of the corresponding modality. The mapping formula for the text modality is:

[0049] ;

[0050] Moreover, K represents the number of kernel functions. It is used to control the complexity and expressiveness of the model. The more kernels, the more refined the model. represents the weight of the kth kernel. It is used to adjust the contribution of each kernel in the final feature map and learn the importance of different kernels. t represents the original feature vector of the text modality. It is used as the input of the mapping and reflects the semantic information of the original data.

[0051] ;

[0052] In the above formula, Represents the center point of the kth kernel in the text feature space. It is used to define the position of the kernel and determine the response area of ​​the kernel in the feature space; Indicates the Gaussian function bandwidth of the kth kernel. It is used to control the width of the kernel. The larger → the narrower the kernel, the smaller the response range, and the more local features are captured. The smaller → the wider the kernel, the larger the response range, and the more global features can be captured; represents the Gaussian kernel function, which is in the form of It is used to nonlinearly map the input features to a high-dimensional space and enhance the separability of features (the text data is processed by the mapping formula of the text modality to generate vector values ​​as feature vectors, such as Figure 5 shown).

[0053] like Figure 5-7 As shown, the modal features of the feature vector are obtained by sequentially passing through the convolution kernel C1, the sampling pool S2, and the convolution kernel C3. The specific steps of obtaining the modal features in this step are:

[0054] S201, scan the convolution kernel C1 on the feature vector, and capture the periodic feature m1 of the feature vector by calculating the weighted sum of the local area (the periodic feature m1 of the feature vector is obtained. Figure 6 As shown, The periodic characteristics and If the periodic features in are approximately the same, then it means that the approximately identical periodic features in this part are periodic features m1).

[0055] S202 : After the periodic feature m1 is pooled through the sampling pool S2 , the overall trend of the periodic feature m1 is extracted.

[0056] S203, the convolution kernel C3 is scanned on the overall trend to obtain the same trend feature m2 in the overall trend (the same trend feature is used to represent Figure 6 The rate of change of the curve in . Figure 7 As shown, and The trend characteristics are similar, as well as Figure 7 If the x and y trend features in are similar, then these similar trend features are all trend features m2).

[0057] S204: Establish based on the periodic feature m1 and the same trend feature m2.

[0058] Among them, when the periodic feature m1 and the same trend feature m2 in the acquired modal feature exist at the same time, it means that the feature vector exists.

[0059] Example 2

[0060] like Figure 1 As shown, a feature database for each department is established based on its corresponding modal features. Based on whether the modal features fall into the feature databases of multiple departments, it is determined whether the modal features of the enterprise department's data intersect with the modal features of other data. If there is an intersection, the signature of the additional department is assigned to the data with the intersection.

[0061] Among them, according to the ratio of the obtained trend feature m2 in the periodic feature m1 and the ratio of the trend feature m2 in the periodic feature m1 in the feature databases of other departments, it is judged whether the modal feature falls into the feature database of other departments at the same time. By proposing to establish an independent feature database based on the modal features of each department, and judging the correlation of cross-departmental data by analyzing whether the modal features fall into multiple databases at the same time. Its core lies in detecting the potential intersection between data from different departments through feature mapping and similarity analysis. For example, by comparing the difference in the proportion of trend feature m2 in the periodic feature m1, the system can identify whether the data is shared by multiple departments or there is redundancy. The advantages of this method are:

[0062] 1) Ability to accurately locate cross-departmental data connections and reduce data silos;

[0063] 2) Improve the structured level of data management through the independence and cross-validation of feature databases;

[0064] 3) Provide a basis for subsequent data label generation and collaborative optimization, and enhance the global interpretability of enterprise data.

[0065] Example 3

[0066] like Figure 1-4 As shown, a data label judgment model is fitted based on the modal features. The specific steps of fitting and generating the data label judgment model in this step are:

[0067] S501: manually set discrete difference coefficients based on the matching degree between the modal features and the modal features in multiple other departmental feature databases.

[0068] The difference coefficient g between the modal feature and the modal feature in the feature database of each department can be expressed by whether the data needs to be reviewed and approved by other departments in the sample data, as well as the total time for review and approval.

[0069] For example, if data, graphics or texts of 100 enterprise departments are collected, if the total time for the data, graphics or text of a certain enterprise department to be reviewed and approved by other departments exceeds the data in the other 50 samples, then it means that the difference coefficient g of the modal feature matching the modal feature in the feature database of each department is 50%.

[0070] S502: Establish a feature relationship x based on the difference coefficient and the ratio of the trend feature m2 in the acquired modal feature to the periodic feature m1 and the ratio of the trend feature m2 in the periodic feature m1 in the feature databases of other departments. Specifically:

[0071] A mathematical model is established for the relationship between the difference coefficient g and the ratio a of the acquired data to the database data (e.g. Figure 2 As shown in the figure, the red dots are the distribution of the 100 samples collected), then:

[0072] (Formula 1);

[0073] In the above formula 1, k represents an empirical constant for adjusting the sensitivity of the above model.

[0074] S503: Establish a feature relationship y based on the difference coefficient and the ratio of the trend feature m2 in the acquired modal feature to the trend feature m2 in the feature database of other departments. Specifically:

[0075] A mathematical model is established for the relationship between the difference coefficient g and the ratio h of the acquired data to the database data (e.g. Figure 3 As shown in the figure, the red dots are the distribution of the 100 samples collected), then:

[0076] (Formula 2);

[0077] In the above formula 2, k represents an empirical constant for adjusting the sensitivity of the above model.

[0078] S504: Based on the feature relationship x and the feature relationship y, a data label judgment model is built. Specifically:

[0079] A mathematical model is established for the relationship between the difference coefficient g and the ratio a of the acquired data to the database data, and the ratio h of the acquired data to the database data (e.g. Figure 4 As shown, the model can be used to know the linear relationship between the ratio a of the acquired data and the database data, and the ratio h of the acquired data and the database data), and combined with the characteristic relationship of the above formula 1 and formula 2, then:

[0080] .

[0081] Example 4

[0082] like Figure 1 、 Figure 5-Figure 7 As shown, the data label judgment model is used to calculate the difference coefficient between the modal feature and the modal feature in each department's feature database. Based on this difference coefficient and the set threshold, it is determined whether the modal feature of the enterprise's department data is unique. If the modal feature is unique, the initial department signature is assigned and feedback is looped. The data label judgment model calculates the difference coefficient between the modal feature and the modal feature in each department's feature database. The calculation formula is:

[0083] .

[0084] Where g represents the difference coefficient between the modal feature and the modal feature in the feature database of each department; a represents the ratio of the ratio of the trend feature m2 in the obtained modal feature to the periodic feature m1 to the ratio of the trend feature m2 in the feature database of other departments to the periodic feature m1; h represents the ratio of the trend feature m2 in the obtained modal feature to the trend feature m2 in the feature database of other departments.

[0085] The explanations are:

[0086] The ratio of the trend feature m2 in the obtained modal feature to the periodic feature m1 and the ratio of the trend feature m2 in the periodic feature m1 in the feature databases of other departments are expressed as follows:

[0087] ;

[0088] And there are:

[0089] ;

[0090] So, .

[0091] The ratio of the trend feature m2 in the obtained modal feature to the trend feature m2 in the feature database of other departments is expressed as follows:

[0092] .

[0093] Examples include:

[0094] like Figure 5-7 As shown, in the periodic feature m1 The trend characteristics m2 (such as Figure 7 The x part in The trend feature m2 in Figure 7 y part in ), then:

[0095] ;

[0096] In the above calculation, the two 0.5s represent the range of k in the x part and the range of k in the y part. The two 2s represent is the range of k and The range of k when When a=0.625. And the comparison between the trend feature m2 in the modal feature obtained and the trend feature m2 in the feature database of other departments is , then we have:

[0097] ;

[0098] According to the calculation of the above formula, we can know that the difference coefficient between the modal feature and the modal feature in the feature database of each department is . And compared with multiple groups of implementation data:

[0099] Table 1: Judgment on whether some implementation data and acquired data are unique

[0100]

[0101] Based on the data in Table 1 above, we can know that when the implementation data tends to be infinite, the difference coefficient g will be used as the dividing line to determine whether the acquired data is unique. For example, when , it can indicate that the acquired data is the only one that exists.

[0102] Example 5

[0103] like Figure 1As shown in the figure, the enterprise full-link intelligent management and control platform driven by a multimodal large model also includes a dynamic update module, which is used to regularly update the feature database of each department based on newly collected multimodal data and adjust the data label judgment model through an incremental learning algorithm to adapt to the dynamic changes in enterprise data. By introducing the dynamic update module, the feature database is regularly updated and the data label judgment model is optimized through an incremental learning algorithm. This module automatically adjusts feature parameters based on newly collected multimodal data, such as the weights of the periodic feature m1 and the trend feature m2, to adapt to dynamic changes in enterprise data (such as business expansion or process adjustments). Its advantages include:

[0104] 1) Realize real-time synchronization between models and data to avoid model failure due to data drift;

[0105] 2) Using incremental learning to reduce the computational overhead of retraining and improve system efficiency;

[0106] 3) Support the continuous mining of long-term data value to ensure the foresight and adaptability of the management and control platform.

[0107] Example 6

[0108] like Figure 1 As shown in the figure, the enterprise full-link intelligent management and control platform driven by a multimodal large model also includes a cross-departmental collaborative optimization module. When it is determined whether the modal characteristics of the data of each department of the enterprise have an intersection with the modal characteristics of other parts of the data, a data fusion strategy is generated based on the difference coefficient, and an automated workflow is triggered to coordinate multiple departments to perform joint tasks. By designing a cross-departmental collaborative optimization module, when the intersection of modal features is detected, a data fusion strategy is automatically generated and a joint task workflow is triggered. For example, if the data of the sales and production departments have an intersection of trend features m2, the system can generate an inventory optimization strategy and coordinate the two departments to synchronously execute procurement and production scheduling plans. Its advantages are:

[0109] 1) Reduce response time for cross-departmental collaboration through data-driven workflow automation;

[0110] 2) Generate precise strategies based on the coefficient of variation to reduce the risk of errors in manual decision-making;

[0111] 3) Optimize resource allocation (such as sharing data to reduce duplicate collection) and improve the overall operational efficiency and cost-effectiveness of the enterprise.

[0112] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. The enterprise full-link intelligent management and control platform driven by multimodal large models is characterized by: The following steps are involved: S1. Collect multimodal data from different departments within the enterprise; S2 maps the modal data collected by different departments to a unified feature vector, and obtains the modal features of the feature vector through convolution kernel C1, sampling pool S2, and convolution kernel C3 in sequence; S3. Establish a feature database for each department based on different departments and their corresponding modal features; S4. Based on whether the modal feature falls into the feature databases of multiple departments at the same time, it is determined whether the modal features of the data of each department of the enterprise have an intersection with the modal features of other parts of the data; S5, and fitting and generating a data label judgment model based on the modal features; S6. Calculating the difference coefficient between the newly acquired modal features and the modal features in the feature database of each department through the data label judgment model; S7. Based on the difference coefficient and the set threshold, determine whether the modal feature of the data of each department of the enterprise is unique. If the modal feature is unique, assign the initial department signature and loop feedback.

2. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 1 is characterized by: The multimodal data includes structured data, text, images, and time series data.

3. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 1 is characterized by: The specific steps of obtaining the modal features in step S2 are: S201, scanning the convolution kernel C1 on the feature vector, and capturing the periodic feature m1 of the feature vector by calculating the weighted sum of the local area; S202, after the periodic feature m1 is pooled through the sampling pool S2, the overall trend of the periodic feature m1 is extracted; S203, scanning the convolution kernel C3 on the overall trend to obtain the same trend feature m2 in the overall trend; S204: Establish based on the periodic feature m1 and the same trend feature m2.

4. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 3 is characterized by: When the periodic feature m1 and the same trend feature m2 in the acquired modal feature exist at the same time, it means that the feature vector exists.

5. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 3 is characterized by: According to the ratio of the obtained trend feature m2 in the periodic feature m1 and the ratio of the trend feature m2 to the periodic feature m1 in the feature databases of other departments, it is determined whether the modal feature also falls into the feature databases of other departments.

6. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 1 is characterized by: In step S5, the specific steps of fitting and generating the data label judgment model are: S501, manually setting a discrete difference coefficient based on the degree of matching between the modal feature and the modal features in multiple other department feature databases; S502, establishing a feature relationship x based on the difference coefficient and the ratio of the trend feature m2 in the acquired modal feature to the periodic feature m1 and the ratio of the trend feature m2 to the periodic feature m1 in the feature databases of other departments; S503, establishing a feature relationship y based on the difference coefficient and the ratio of the trend feature m2 in the acquired modal feature to the trend feature m2 in the feature database of other departments; S504: Based on the feature relationship x and the feature relationship y, a data label judgment model is fitted and established.

7. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 3 is characterized by: The data label judgment model calculates the difference coefficient between the modal feature and the modal feature in the feature database of each department. The calculation formula is: ； Where, The difference coefficient indicating the matching between the modal feature and the modal feature in each department’s feature database; It represents the ratio of the trend feature m2 in the periodic feature m1 in the obtained modal feature to the ratio of the trend feature m2 in the periodic feature m1 in the feature database of other departments; It represents the ratio of the trend feature m2 in the acquired modal features to the trend feature m2 in the feature database of other departments.

8. The enterprise full-link intelligent management and control platform driven by a multimodal large model according to claim 1 is characterized in that: It also includes a dynamic update module for regularly updating the feature database of each department based on newly collected multimodal data, and adjusting the data label judgment model through an incremental learning algorithm to adapt to the dynamic changes of enterprise data.

Images