Post-investment management information system and post-investment management method

Through post-investment management information system and AI technology, the problem of low efficiency of enterprise data management and post-investment management is solved, intelligent data analysis and risk assessment are realized, and investment decision-making support capabilities are improved.

CN117893326BActive Publication Date: 2024-12-03GUIZHOU IND INVESTMENT DEV CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410076977.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-18
Publication Date
2024-12-03
Estimated Expiration
2044-01-18

AI Technical Summary

Technical Problem

In the investment process, enterprises face problems such as irregular data, lack of unified management and long-term data accumulation, resulting in low post-investment management efficiency and insufficient intelligence, and unable to effectively evaluate the company's operating risks and investment potential.

Method used

Adopt post-investment management information system, combined with cloud-edge computing-terminal architecture, multi-source heterogeneous data standardization and large-capacity data storage technology, and use deep learning models and AI technology to collect, analyze and evaluate enterprise business data to generate and operate. Analyze reports, tap potential problems and evaluate investment potential.

Benefits of technology

It realizes standardized management and intelligent analysis of enterprise data, improves the efficiency and accuracy of post-investment management, can perceive business risks in advance, evaluate investment potential, and support corporate decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117893326B_ABST
    Figure CN117893326B_ABST
Patent Text Reader

Abstract

The present invention relates to artificial intelligence technology, and provides a post-investment management information system and a post-investment management method, including: constructing a post-investment management information system; the post-investment management information system collects enterprise operating data; using a first deep learning model to analyze the collected enterprise operating data; according to the analysis results, mining the potential operating problems of the enterprise; and evaluating the potential of the enterprise at different operating stages.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a post-investment management information system and a post-investment management method. Background Art

[0002] During the investment process, many companies face the problem of irregular, complicated and unorganized management of operating data. In addition, due to the lack of long-term data accumulation and management capabilities, the data required for post-investment management is usually temporarily pieced together. For example, different companies usually have image-based, text-based and pure data-based business data, which are often difficult to manage and standardized, resulting in inefficient data management and inaccurate post-investment management. In addition, post-investment management also lacks intelligent management methods, and cannot perceive the potential operating risks of the company in advance through in-depth analysis of the company's operating data, nor can it conduct intelligent evaluation of its subsequent investment potential. Summary of the invention

[0003] The present application provides a post-investment management information system and a post-investment management method to solve the problems of low efficiency of data management and low intelligence level of post-investment management in the prior art.

[0004] In view of the above problems, the present application provides a post-investment management information system and a post-investment management method.

[0005] The present application embodiment provides a post-investment management method, including:

[0006] Build a post-investment management information system;

[0007] The post-investment management information system collects enterprise operating data;

[0008] Analyzing the collected business operation data using a first deep learning model;

[0009] Based on the analysis results, potential operational problems of the enterprise are explored;

[0010] An assessment is made of the potential of the enterprise at different stages of operation.

[0011] Optionally, analyzing the collected business operation data using the first deep learning model includes:

[0012] Preprocessing the business operation data of the enterprise;

[0013] Dividing the pre-processed enterprise operation data into first image type data and first sequence type data;

[0014] Construct convolutional neural network (CNN) model and recurrent neural network (RNN) model;

[0015] The CNN model is trained using historical image data, and the RNN model is trained using historical sequence data;

[0016] Analyze the first image data using the trained CNN model to output a first enterprise operating status and a first decision suggestion;

[0017] Analyze the first sequence data using the trained RNN model to output a second enterprise operating status and a second decision suggestion;

[0018] The first business operating status, the first decision-making suggestion, the second business operating status and the second decision-making suggestion are integrated to generate a business operating analysis report.

[0019] Optionally, using historical image data to train the CNN model includes:

[0020] Set initial values ​​for the model's weights;

[0021] Set the cross entropy loss function;

[0022] Set the Adam optimizer and set a learning rate;

[0023] Set the batch size;

[0024] Divide the historical image class data into the set batch size for forward propagation to obtain the predicted output;

[0025] Use the loss function to compare the predicted output of the CNN model with the true label and calculate the loss;

[0026] Performing back propagation, including: calculating the gradient of each layer using the chain rule, and updating the weights and biases of the CNN model using an optimizer based on the gradient;

[0027] At the end of each training cycle, the performance of the CNN model is evaluated using the validation set;

[0028] The RNN model is trained using historical sequence data, including:

[0029] Initialize the parameters of the RNN model;

[0030] Input a batch of sequence data into the RNN model;

[0031] Calculating each time step of the sequence data, wherein for each time step, the current input and the hidden state of the previous time step are used to generate the hidden state of the current time step;

[0032] In the last time step, a fully connected layer is used to obtain the prediction output;

[0033] Use the loss function to compare the predicted output of the RNN model with the true label and calculate the loss;

[0034] Perform back propagation;

[0035] The performance of the RNN model was evaluated using the validation set.

[0036] Optionally, based on the analysis results, potential operational problems of the enterprise are explored, including:

[0037] Obtain historical business operating reports;

[0038] Preprocessing the historical business operation reports, and dividing the preprocessed historical business operation reports into a second training set and a second test set;

[0039] Building a second deep learning model;

[0040] Training the second deep learning model using the second training set;

[0041] The trained second deep learning model is used to mine potential business problems of the enterprise.

[0042] Optionally, using the trained second deep learning model to mine potential business problems of the enterprise includes:

[0043] Inputting the business data of the enterprise into the trained second deep learning model and outputting a prediction result;

[0044] Comparing the predicted results with data under normal operating modes to identify abnormal data;

[0045] Based on the abnormal data, potential operating problems of the enterprise are determined, where the potential operating problems are a combination of one or more problems in inventory backlog, customer loss, finance, supply chain disruption, and market competition.

[0046] Optionally, the second deep model includes a teacher model and a student model, and constructing a second deep learning model, and training the second deep learning model using the second training set includes:

[0047] Construct teacher model and student model;

[0048] Using the second training set to train the teacher model;

[0049] Using the output of the teacher model as a soft target to train the student model;

[0050] During the process of training the student model, temperature scaling technology is used to adjust the probability output of the teacher model, and the student model is iteratively trained after each adjustment of the probability output of the teacher model, so that the potential operating problems of the enterprise can be mined through the student model.

[0051] Optionally, the deep learning model includes a parallel model of a CNN and a long short-term memory (LSTM) model, and the trained second deep learning model is used to mine potential business problems of the enterprise, including:

[0052] Use the CNN model to analyze image-based business data and identify data features of the image-based business data;

[0053] Use the LSTM model to analyze the sequence-type business data and identify the data features of the sequence-type business data;

[0054] The recognition results of the CNN model and the LSTM model are integrated to analyze the potential operating problems of the enterprise.

[0055] Optionally, an assessment of the potential of the business at different stages of operation, including:

[0056] Collect historical operating data;

[0057] Preprocessing the historical operating data;

[0058] Constructing a machine learning model, and using the preprocessed historical operating data to train the machine learning model;

[0059] The machine learning model is used to evaluate the potential of the enterprise at different stages of operation.

[0060] Optionally, after the post-investment management information system collects the enterprise operation data, the method further includes:

[0061] Build a large-scale storage system and establish a data redundancy backup mechanism in the storage system;

[0062] Monitor the health status of storage nodes;

[0063] When any storage node goes down, determine the data blocks that need to be migrated;

[0064] An optimal data migration path is selected, and a data migration operation is started to migrate the data blocks that need to be migrated.

[0065] An embodiment of the present application also provides a post-investment management information system, in which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0066] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0067] The embodiment of the present application proposes an information system and management method for post-investment management. Through the system and method, the problems of enterprises in the investment process due to non-standard data, lack of unified management and long-term data accumulation are solved. At the same time, the cloud-edge computing-terminal architecture, multi-source heterogeneous data standardization and large-capacity data storage technology, and AI technology are introduced to realize the intelligent management of post-investment enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A flowchart of the post-investment management method provided for this application;

[0069] Figure 2 A detailed flow chart of S103 provided in this application;

[0070] Figure 3 Provide a flowchart for mining potential business problems of the enterprise for this application;

[0071] Figure 4 A flowchart for mining potential business problems of an enterprise using a trained second deep learning model provided in this application;

[0072] Figure 5 A detailed flow chart of S105 provided in this application;

[0073] Figure 6 Schematic diagram of the post-investment management information system structure provided for this application. DETAILED DESCRIPTION

[0074] This application provides a post-investment management information system and post-investment management method, which solves the problems of enterprises in the investment process due to non-standardized data, lack of unified management and long-term data accumulation. At the same time, it introduces cloud-edge computing-terminal architecture, multi-source heterogeneous data standardization and large-capacity data storage technology, and AI technology to achieve intelligent management of post-investment enterprises.

[0075] Embodiment 1

[0076] like Figure 1 As shown, the present application provides a post-investment management method, the method comprising:

[0077] S101. Build a post-investment management information system;

[0078] The post-investment management information system is a system for managing and controlling enterprises after investment. For investment, it is divided into pre-investment, mid-investment and post-investment according to time. For enterprises after investment, it is necessary to obtain their business data and use information tools for effective management and supervision.

[0079] The establishment of the post-investment management information system requires communication with people from different departments and the definition of effective needs. It adopts the cloud-edge computing-terminal architecture to ensure the real-time and efficient processing of data.

[0080] The system selects the appropriate cloud service provider and sets the cloud computing strategy. It also designs edge computing nodes to ensure fast data processing at the edge. Finally, it is necessary to determine the type and configuration of the terminal device to ensure stable data collection.

[0081] In addition, the system can realize the standardization of multi-source heterogeneous data and large-capacity data storage technology. For example, the system needs to design a data standardization process, ensure data consistency, and select and implement appropriate large-capacity data storage technology.

[0082] Data standardization is the process of converting data from different sources or collection methods into a unified format or structure for subsequent processing and analysis. Data standardization is particularly important in post-investment management because the data involved in investment comes from multiple different companies or departments, and their data formats and structures vary.

[0083] The following is a simple data standardization process A1-A5:

[0084] Step A1: Data source identification and classification

[0085] Identify all data sources, such as the company's financial system, production system, sales system, etc.

[0086] Classify these data sources, such as by data type (numeric, text, etc.) and data frequency (day, month, year, etc.).

[0087] Step A2: Define standard data format

[0088] Based on the needs of post-investment management, define a unified data format or structure, which can be a database table structure, an Excel template, etc.

[0089] Ensure that this format can cover all data fields required for post-investment management and has a certain degree of scalability to adapt to future changes in requirements.

[0090] Step A3: Data cleaning and transformation

[0091] Clean the raw data to remove irrelevant, duplicate, and erroneous data.

[0092] Convert the cleaned data into the standard format defined in the previous step. This requires some data conversion tools or scripts.

[0093] Step A4: Data Verification

[0094] Verify whether the converted data is consistent with the original data to ensure the accuracy of data conversion.

[0095] Perform simple statistical analysis on the standardized data to check whether there are outliers or illogical data, such as negative sales.

[0096] Step A5: Data storage and backup

[0097] Store the standardized data in the database of the post-investment information system.

[0098] Back up this data regularly to ensure data security.

[0099] In this way, through the above-mentioned data standardization process, the consistency and accuracy of the data used for post-investment management can be ensured, providing a solid foundation for subsequent data analysis and decision-making.

[0100] Furthermore, with the rapid growth of enterprise data volume, traditional relational databases have been unable to meet the needs of big data storage and processing. Therefore, choosing a suitable large-capacity data storage technology is crucial for the stable operation of the post-investment management information system.

[0101] The following are the specific implementation steps B1-B5 for data storage and backup using mass storage technology:

[0102] Step B1: Assess data volume and growth rate

[0103] Evaluate data storage needs based on current system data volumes and predicted future growth rates.

[0104] Considering factors such as data backup, disaster recovery, and historical data storage, determine the required storage capacity.

[0105] Step B2: Select Data Storage Technology

[0106] Choose the appropriate data storage technology based on the type of data and access frequency. For example, for structured data, you can choose a relational database; for large amounts of log or text data, you can choose a NoSQL database or a distributed file system.

[0107] Evaluate the pros and cons of various data storage technologies, taking into account data security, high availability, and scalability.

[0108] Step B3: Deploy and optimize data storage system

[0109] Deploy the system based on the selected data storage technology, for example, deploy the distributed file system HDFS, the column storage database HBase, etc.

[0110] Optimize the performance of data storage systems to ensure fast reading and writing of data and stable access.

[0111] Step B4: Data Migration and Synchronization

[0112] Migrate data from existing systems to new data storage systems.

[0113] Set up a data synchronization mechanism to ensure data consistency between the new and old systems.

[0114] Step B5: Monitor and maintain data storage systems

[0115] Monitor the data storage system in real time to ensure the stable operation of the system.

[0116] Back up data regularly to prevent data loss.

[0117] Perform regular maintenance on the data storage system, such as clearing expired data and expanding storage space.

[0118] Through the above steps, it can be ensured that the data in the post-investment management information system is stored safely, efficiently and stably, providing solid data support for the company's post-investment management.

[0119] S102. The post-investment management information system collects the enterprise's operating data;

[0120] Among them, the post-investment management information system deploys data collection terminals to collect corporate operating data, and monitors the collection status of this data in real time to ensure data integrity.

[0121] In the embodiments of the present application, intelligent data collection technology can be used to efficiently collect internal enterprise data.

[0122] For enterprises, there are a large number of data resources within the enterprise, such as ERP system, CRM system, production management system, etc. In order to ensure the effectiveness of post-investment management, these data need to be collected efficiently and accurately.

[0123] Step C1: Determine the scope and frequency of data collection

[0124] According to the needs of post-investment management, determine the type of data, data source and data frequency that need to be collected. For example, sales data needs to be collected every day, inventory data needs to be collected every month, etc.

[0125] Considering the timeliness and importance of the data, key data should be collected first.

[0126] Step C2: Select appropriate data collection tools or platforms

[0127] Choose the appropriate data collection tool or platform based on the characteristics of the data source and the interface type. For example, for a relational database, you can choose an ETL tool; for a web page, you can choose a web crawler tool.

[0128] Evaluate the pros and cons of various tools or platforms, taking into account data security and collection efficiency.

[0129] Step C3: Design the data collection process

[0130] Design the data collection process and steps according to the characteristics of the data source. For example, first log in to the system, then query the data, and finally save the data.

[0131] For data that needs to be collected regularly, design automated data collection tasks to ensure timely updating of data.

[0132] Step C4: Implement data collection

[0133] Use the tool or platform of your choice to collect data.

[0134] Monitor the progress and results of data collection to ensure data integrity and accuracy.

[0135] In addition, it should be noted that there are different types of data, including image, text, and sequence data, and different data are stored in different formats.

[0136] Step C5: Data verification and quality control

[0137] Verify the collected data to ensure its consistency and accuracy.

[0138] Use data quality control tools to perform quality checks on data, such as checking data integrity, uniqueness, consistency, etc.

[0139] Step C6: Data integration and storage

[0140] Integrate the collected data with existing data to ensure data consistency.

[0141] The integrated data is stored in the database of the post-investment information system to provide data support for subsequent data analysis and application.

[0142] Through the above steps, it can be ensured that the internal data of the enterprise is collected efficiently and accurately, providing strong data support for post-investment management.

[0143] Alternatively, node downtime is a common problem in big data storage. In order to ensure data persistence and high availability, when a node goes down, the data migration mechanism needs to be quickly started to migrate the affected data to other healthy nodes. Therefore, if a downtime occurs, after collecting the data, it is necessary to use the data migration mechanism based on big data storage after node downtime to save the data. This is a very important step, as follows:

[0144] M1. Build a large-scale storage system and establish a data redundancy backup mechanism in the storage system;

[0145] Use technologies such as the data replication mechanism in Hadoop's HDFS to ensure that each piece of data is backed up on multiple nodes.

[0146] Set an appropriate replication factor, usually 3, to ensure high availability and fault tolerance of data.

[0147] M2. Monitor the health status of storage nodes;

[0148] Use tools such as Hadoop's YARN or other cluster management tools to monitor the health status of each node in real time.

[0149] M3. When any storage node goes down, determine the data blocks that need to be migrated;

[0150] When a node crash is detected, the data migration mechanism is triggered immediately.

[0151] Analyze the data stored on the downed node and determine the data blocks that need to be migrated.

[0152] Use HDFS metadata to quickly determine the affected data blocks and their replica locations on other nodes.

[0153] When a node goes down, the first task is to quickly determine which data is affected so that subsequent data recovery and migration operations can be performed. The following are the specific steps on how to determine the affected data:

[0154] Step M31: Query metadata

[0155] Big data storage systems, such as HDFS, maintain the mapping of files to data blocks and data blocks to nodes in the NameNode or other metadata services.

[0156] Query these metadata to list all data blocks on the downed node.

[0157] Step M32: Check the copy status of the data block

[0158] For each affected data block, query the number of its replicas on other nodes.

[0159] Mark data blocks that have copies only on down nodes or whose number of copies is below a preset threshold.

[0160] Step M33: Evaluate data recovery priorities

[0161] Set a recovery priority for each affected data block based on data importance, access frequency, and business needs.

[0162] For example, critical business data that is frequently accessed should have a higher recovery priority.

[0163] Step M34: Consider data association

[0164] Consider the relevance of the affected data to other data. For example, if some partitions of a database table are affected, you may need to consider the recovery of the entire table.

[0165] Ensure that data integrity and consistency are maintained during the recovery process.

[0166] Step M35: Generate a data recovery and migration plan

[0167] Combining the above information, a recovery and migration plan is generated for each affected data block.

[0168] The plan should include the target storage location for the data, the migration path, and the expected recovery time.

[0169] Through the above steps, the embodiment of the present invention can quickly and accurately determine the affected data on the downed node and formulate a specific plan for subsequent data recovery and migration operations.

[0170] M4. Select the best data migration path and start the data migration operation to migrate the data blocks that need to be migrated.

[0171] Select the best data migration path based on the size of the data block, network bandwidth, and load of the target node.

[0172] Start the data migration process to ensure that the data has sufficient copies on other healthy nodes.

[0173] Data migration is a key step to ensure data availability and integrity. When a node goes down, the affected data needs to be quickly migrated to other healthy nodes. The following are the specific steps for data migration:

[0174] Step M41: Select target node

[0175] The best target node is selected based on the cluster's load balancing and the node's storage capacity.

[0176] Make sure the target node has sufficient resources to accept the migrated data.

[0177] Step M42: Initialize data transmission

[0178] Start data transfer using built-in tools of big data storage systems such as HDFS or other data migration tools.

[0179] Select the best data transmission strategy based on the data size and network bandwidth, such as parallel transmission or data compression.

[0180] Step M42: Start data transfer using built-in tools of the big data storage system (such as HDFS) or other data migration tools

[0181] In the big data ecosystem, especially HDFS, a series of built-in tools are provided to help data migration. The following are the detailed steps for data transfer using these tools:

[0182] Step M42.1: Identify data sources and targets

[0183] List the data files or directories that need to be migrated on the downed node.

[0184] Select one or more healthy nodes as the target for data migration.

[0185] Step M42.2: Use the `distcp` command

[0186] `distcp` (distributed copy) is a tool provided by Hadoop, specifically for large-scale data copying.

[0187] Construct the `distcp` command, specifying the source and destination paths.

[0188] For example: `hadoop distcp hdfs: / / source-cluster / data hdfs: / / target-cluster / data`

[0189] Step M42.3: Consider data compression

[0190] If the amount of data is huge or the network bandwidth is limited, you can consider compressing the data before transmission.

[0191] Use compression formats such as gzip or snappy to reduce the size of data transfers.

[0192] Step M42.4: Parallel transfer

[0193] `distcp` supports parallel transfers, that is, copying multiple files or data blocks at the same time.

[0194] Set an appropriate parallel level based on network bandwidth and cluster resources. For example, use the -m option to specify the number of parallel tasks.

[0195] Step M42.5: Handle failed transfers

[0196] `distcp` will log any failed data transfers.

[0197] If a failure occurs, you can rerun the `distcp` command and use the `-i` option to ignore failures for individual files.

[0198] Step M42.6: Other data migration tools

[0199] In addition to `distcp`, there are other third-party tools such as `rsync` or migration tools provided by specific cloud services.

[0200] Choose the tool that best suits your environment and data volume for data migration.

[0201] Through the above steps, the embodiment of the present invention can use the built-in tools of HDFS or other big data storage systems to migrate data, ensuring that data is quickly and accurately restored after a node crashes.

[0202] Step M43: Monitor data migration progress

[0203] Monitor the progress of data migration in real time to ensure the integrity and speed of data transfer.

[0204] If you encounter network failures or other problems, take prompt action.

[0205] Step M44: Verify the integrity of the migrated data

[0206] After the data migration is completed, the data on the target node is checked for integrity.

[0207] Use methods such as checksums to ensure that data is not corrupted during the migration process.

[0208] Step M45: Update data copy strategy

[0209] Adjust the data replication strategy based on the new data layout to ensure data persistence and high availability.

[0210] If necessary, you can increase the number of copies of the data or change the distribution of the copies.

[0211] Through the above steps, the embodiment of the present invention can ensure that in the event of a node crash, the affected data is quickly and accurately migrated to other healthy nodes, thereby ensuring the availability and integrity of the data.

[0212] S103. Analyze the collected business operation data using the first deep learning model;

[0213] like Figure 2 As shown, in S103, the collected business operation data is analyzed using the first deep learning model, including the following steps:

[0214] D1. Preprocessing the business operation data of the enterprise;

[0215] Clean, transform, and normalize data as required by deep learning models.

[0216] For unstructured data such as images, sounds, and text, use appropriate methods to convert them into a format that the model can accept.

[0217] D2. Dividing the pre-processed business operation data into first image data and first sequence data;

[0218] In the above preprocessing process, for structural data, it can be defined as the first sequence data, for non-structural data such as images, it can be defined as the first image data, and text data can be converted into sequence data and image data using the data format conversion method. Therefore, text data can be converted into structural data, that is, the first sequence data, according to business needs, or it can be defined as the first image data as an image format.

[0219] D3. Build convolutional neural network (CNN) model and recurrent neural network (RNN) model;

[0220] Deep learning is a subfield of machine learning that can automatically learn features from data and perform efficient analysis. In post-investment management, deep learning can help the embodiments of the present invention to deeply analyze the business data of the enterprise and explore the value behind the data.

[0221] Select the appropriate deep learning model based on the characteristics of the data and the analysis task. Deep learning models can include CNN and RNN models. For example, for image data, you can choose convolutional neural network; for sequence data, you can choose recurrent neural network.

[0222] For the CNN model, you need to design a CNN architecture, including convolutional layers, pooling layers, fully connected layers, etc. Select appropriate activation functions, loss functions, and optimizers.

[0223] For the RNN model, you need to design the RNN architecture. You can choose traditional RNN, LSTM, GRU or other variants. Add the appropriate number of hidden layers, neurons and output layers. Choose the appropriate activation function, loss function and optimizer.

[0224] D4. Using historical image data to train the CNN model, and using historical sequence data to train the RNN model;

[0225] The CNN model is trained using historical image data, including the following steps E1-E8:

[0226] E1. Set initial values ​​for the model weights;

[0227] Use initialization methods such as Xavier or He to set initial values ​​for the model's weights. This helps ensure that the weights are within a suitable range, thereby speeding up training and improving model convergence.

[0228] E2. Set the cross entropy loss function;

[0229] For multi-classification tasks, the cross entropy loss function is usually chosen.

[0230] For binary classification tasks, the binary cross entropy loss function can be chosen.

[0231] Depending on the task requirements, other loss functions can also be selected, such as mean square error, Hinge loss, etc.

[0232] In the embodiment of the present invention, the cross entropy loss function is preferably selected as the loss function of the model.

[0233] E3. Set the Adam optimizer and set a learning rate;

[0234] Common optimizers include SGD, Adam, RMSprop, etc. Each optimizer has its own advantages and characteristics, and the choice depends on the specific requirements of the task. The embodiment of the present application gives priority to the most commonly used Adam optimizer.

[0235] Set a learning rate. A learning rate that is too large will cause the model to not converge, while a learning rate that is too small will result in very slow training.

[0236] E4. Set the batch size;

[0237] It is planned to divide the training data into several batches for forward and backward propagation. The batch size affects the training speed and memory usage of the model.

[0238] E5. Divide the historical image class data into the set batch size for forward propagation to obtain the predicted output;

[0239] Among them, the embodiments of the present application can also use image enhancement technology to process image data, such as rotation, translation, scaling, flipping, etc., which can increase the generalization ability of the model and reduce the risk of overfitting.

[0240] In addition, the embodiment of the present application divides the data into training, validation and test sets. The common division ratio is 80% for training, 10% for validation and 10% for testing.

[0241] During the forward propagation process, a batch of data is input into the model, and after calculations at each layer, the predicted output is obtained.

[0242] E6. Use the loss function to compare the predicted output of the CNN model with the true label and calculate the loss;

[0243] The loss is calculated by comparing the model's predicted output with the true label using the cross entropy loss function.

[0244] E7. Performing back propagation, including: calculating the gradient of each layer using the chain rule, and updating the weights and biases of the CNN model using an optimizer based on the gradient;

[0245] E8. At the end of each training cycle, the performance of the CNN model is evaluated using the validation set.

[0246] At the end of each training cycle, evaluate the performance of the model using the validation set. This can help monitor overfitting and decide whether to stop training early or adjust the learning rate.

[0247] If the validation loss does not improve significantly over a few epochs, you can consider stopping training early to prevent overfitting.

[0248] In addition, in D4, the RNN model is trained using historical sequence data. For RNN (recurrent neural network) models, the process of training using historical sequence data is similar to training other types of neural networks, but there are some considerations specific to sequence data and RNN. Specifically, steps F1-F7 are included:

[0249] F1. Initialize the parameters of the RNN model;

[0250] Similar to CNN, you need to set parameters such as loss function, optimizer, learning rate, etc. You also need to set the appropriate number of hidden layers, neurons, and output layers.

[0251] F2. Input a batch of sequence data into the RNN model;

[0252] In the embodiment of the present invention, since the input is sequence data, which may be time series data, text data, audio data, etc., these data need to be converted into numerical data, for example, through text encoding, feature extraction, etc.

[0253] In addition, the data needs to be normalized or standardized to be within a suitable range. If necessary, long sequences can be truncated or padded to meet the input requirements of the model.

[0254] Finally, use sliding windows or other strategies to convert the sequence data into training samples. Split the data into training, validation, and test sets.

[0255] After the division, a batch of sequence data (a batch of sequence data in the training set) is input into the RNN model to start the forward propagation process.

[0256] F3. Calculate each time step of the sequence data, wherein, for each time step, the current input and the hidden state of the previous time step are used to generate the hidden state of the current time step;

[0257] During the forward propagation of the RNN unit, calculations are performed at each time step of the sequence. For each time step, the current input and the hidden state of the previous time step are used to generate the hidden state of the current time step.

[0258] F4. Obtain the prediction output through a fully connected layer in the last time step;

[0259] After the last time step, if it is a classification task, the prediction output can be obtained through a fully connected layer.

[0260] F5. Use the loss function to compare the predicted output of the RNN model with the true label and calculate the loss;

[0261] The loss function is used to compare the model's predicted output with the true label and calculate the loss. For sequence labeling tasks, the loss is calculated at each time step and then averaged.

[0262] F6. Perform back propagation;

[0263] In contrast to forward propagation, backpropagation starts from the last time step of the sequence and gradually goes back to the first time step. It uses the chain rule to calculate the gradients for each time step and each layer. Based on these gradients and the selected optimizer, such as SGD, Adam, etc., the weights and biases of the model are updated.

[0264] After back propagation, the learning rate needs to be adjusted. The learning rate is a key hyperparameter. You can choose a larger learning rate at the beginning, and gradually reduce the learning rate as the training progresses.

[0265] You can use a learning rate decay strategy or use an adaptive learning rate optimizer such as Adam.

[0266] F7. Use the validation set to evaluate the performance of the RNN model.

[0267] At the end of each training cycle, the performance of the model is evaluated using the validation set. This not only monitors the training progress of the model, but also helps detect overfitting.

[0268] Based on the verification results, you can adjust the learning rate, add dropout or early stopping strategies.

[0269] D5. Analyze the first image data using the trained CNN model to output the first enterprise operating status and the first decision-making recommendation;

[0270] Apply the trained deep learning model to actual business data for in-depth analysis.

[0271] Based on the output of the model, provide decision-making recommendations to enterprises, such as market trend forecasting, customer behavior analysis, etc.

[0272] For example, the first enterprise's operating conditions:

[0273] -Company A is an electronics manufacturer.

[0274] -Sales have declined recently, mainly because its latest products have not been widely praised by consumers.

[0275] -Good cost management, but low R&D investment.

[0276] First decision suggestion:

[0277] -Increase R&D investment, improve existing products, and develop new products to meet market demand.

[0278] - Strengthen marketing strategies and increase brand awareness.

[0279] -Consider collaborating with other companies to jointly develop new products.

[0280] D6. Use the trained RNN model to analyze the first sequence data and output the second enterprise operating status and second decision-making recommendations;

[0281] Similarly, RNN also outputs business conditions and decision suggestions, for example:

[0282] Second business status:

[0283] -Company B is a company that provides online education services.

[0284] -The number of users has grown steadily, but the user retention rate is low.

[0285] -Although the course content is rich, the user interface is not user-friendly enough, resulting in a poor user experience.

[0286] Second decision suggestion:

[0287] -Optimize the user interface and experience of the online platform.

[0288] - Enhance interaction with users, such as through community forums or online Q&A.

[0289] - Provide more customized learning paths and recommendations to meet the needs of different users.

[0290] D7. Integrate the first business operating status, the first decision-making suggestion, the second business operating status and the second decision-making suggestion to generate a business operating analysis report.

[0291] Among them, the CNN model outputs the first enterprise's operating status and the first decision-making recommendation, and the RNN outputs the second enterprise's operating status and the second decision-making recommendation;

[0292] The first enterprise operating status, the first decision-making suggestion, the second enterprise operating status and the second decision-making suggestion are integrated to generate an enterprise operating analysis report.

[0293] If the CNN model outputs one company and the RNN model outputs another company, then the two can be directly pieced together to get a fused business analysis report. If the CNN model and the RNN model output the operating conditions and decision-making recommendations of the same company, then the same parts of the two are removed, and the different parts of the two are combined and spliced ​​in the report to form a general business analysis report.

[0294] For example:

[0295] Business analysis report 1:

[0296] Business analysis of Enterprise A:

[0297] Company A, an electronic product manufacturer, has recently faced the challenge of declining sales. Despite its good cost management, product innovation and market feedback show that further improvements are needed. In order to reverse this trend, Company A is recommended to increase R&D investment, not only to improve existing products, but also to develop new products to meet the changing market needs. In addition, by strengthening marketing and cooperating with other companies, Company A is expected to increase its market share and competitiveness.

[0298] Business analysis report 2:

[0299] Business analysis of Company B:

[0300] Online education service provider Company B has been successful in attracting new users, but faces a low user retention rate. After in-depth analysis, it was found that although it provided rich course content, the user interface of the online platform was unsatisfactory. In order to improve user retention and satisfaction, Company B needs to comprehensively optimize its online platform and strengthen interaction with users. By providing a more personalized learning experience, Company B is expected to further consolidate its position in the online education market.

[0301] Business analysis report 3:

[0302] Reporting time: second quarter of 2023

[0303] 1. Sales

[0304] Total sales: RMB 50 million

[0305] Year-over-year growth: 10%

[0306] Month-over-month growth: 5%

[0307] 2. Customer Analysis

[0308] Number of new customers: 1,000

[0309] Number of lost customers: 200

[0310] Major customer industries: manufacturing, service, retail

[0311] 3. Product status

[0312] Hot-selling products: Product A, Product B, Product C

[0313] Inventory backlog products: Product X, Product Y

[0314] 4. Financial Status

[0315] Total assets: RMB 100 million

[0316] Debt ratio: 50%

[0317] Net profit: RMB 5 million

[0318] 5. Market trends

[0319] Market share: 15%

[0320] Competitor situation: Competitor 1 increased by 5%, Competitor 2 decreased by 2%

[0321] Market forecast: Market demand is expected to grow by 10% in the next quarter

[0322] 6. Business suggestions

[0323] Strengthen the production and sales of Product A, Product B, and Product C to increase market share.

[0324] Carry out promotions or adjust production strategies for product X and product Y that have inventory backlogs.

[0325] Strengthen cooperation with manufacturing, service and retail industries to expand customer base.

[0326] This report is based on the operating data of a certain industrial manufacturing company, and is analyzed through a deep learning method to provide decision support for the company. The embodiment of the present invention can provide the report to relevant departments for decision support. In addition, in terms of report verification, the accuracy of the information can be verified by comparing it with the on-site data. The system can also be adjusted based on the on-site verification results.

[0327] S104. Based on the analysis results, potential operating problems of the enterprise are explored;

[0328] There are many potential problems in the business operation process, such as low production efficiency, inventory backlog, poor capital turnover, etc. Using AI technology, these potential problems can be mined from a large amount of business data to provide decision support for enterprises.

[0329] like Figure 3 As shown, according to the analysis results, the potential operating problems of the enterprise are mined, including steps G1-G5:

[0330] G1. Obtain historical business operation reports;

[0331] For a certain enterprise, its operating reports in different historical periods are different. Therefore, it is necessary to obtain different enterprise operating reports from different historical periods. These enterprise operating reports from different historical periods can help the AI ​​model better learn and discover potential enterprise operating problems.

[0332] G2. preprocessing the historical business operation report, and dividing the preprocessed historical business operation report into a second training set and a second test set;

[0333] Clean, transform and standardize data according to the needs of AI models.

[0334] The data set is divided into a second training set and a second test set to facilitate model training and verification.

[0335] G3. Build the second deep learning model;

[0336] Choose the appropriate AI model based on the characteristics of the business problem. For example, for inventory forecasting problems, you can choose a time series forecasting model; for customer churn forecasting, you can choose a classification model.

[0337] G4. Training the second deep learning model using the second training set;

[0338] Use the second training set data to train the AI ​​model.

[0339] Adjust model parameters, such as learning rate, regularization parameters, etc., to optimize model performance.

[0340] Let’s take a neural network as an example:

[0341] Neural network is a core model in the field of AI, and it performs particularly well in complex data analysis and prediction tasks. The following is the training process using a neural network as an example:

[0342] Step G41: Neural network structure design

[0343] According to the characteristics of the data and task requirements, determine the structure of the neural network, such as the number of layers and the number of neurons in each layer.

[0344] Choose a suitable activation function, such as ReLU, sigmoid, tanh, etc.

[0345] If the task is classification, determine the number of neurons in the output layer and use softmax as the activation function.

[0346] Step G42: Initialize neural network parameters

[0347] Initialize the weights of the network using small random numbers or specific initialization techniques (such as Xavier initialization, He initialization).

[0348] Initialize the bias term to 0 or a small positive value.

[0349] Step G43: Select loss function and optimizer

[0350] Choose a loss function based on the task type, such as mean squared error (for regression tasks) or cross entropy loss (for classification tasks).

[0351] Select an optimizer such as Gradient Descent, Adam, RMSprop, etc. and set the learning rate and other parameters.

[0352] Step G44: Neural network training

[0353] Feed the neural network with training data.

[0354] In each epoch, forward propagation is performed to calculate the predicted value, and then the error is calculated through the loss function.

[0355] The network weights are updated using the back-propagation algorithm.

[0356] Repeat the above steps until the preset number of epochs is reached or the loss function value is lower than a certain threshold.

[0357] Step G45: Verification and Adjustment

[0358] The performance of the neural network was evaluated using the validation dataset.

[0359] Based on the validation results, adjust the network structure, parameters, or optimizer settings, and then retrain.

[0360] To prevent overfitting, regularization methods, dropout and other techniques can be used.

[0361] Step G46: Model saving and deployment

[0362] After training is completed, save the structure and weight parameters of the neural network.

[0363] According to actual needs, the model is deployed to the corresponding application environment to provide support for subsequent data analysis and decision-making.

[0364] Using neural networks for model training can effectively capture nonlinear relationships and complex patterns in data, providing a powerful tool for business data analysis.

[0365] In addition, test set data is needed to verify the AI ​​model.

[0366] Based on the validation results, the accuracy, robustness, and generalization ability of the model are evaluated.

[0367] G5. Use the trained second deep learning model to explore potential operational problems of the enterprise.

[0368] like Figure 4 As shown, in G4, the trained second deep learning model is used to mine the potential business problems of the enterprise, including:

[0369] G41. Input the business data of the enterprise into the trained second deep learning model and output the prediction result;

[0370] The prediction results depend on the objectives of the model, such as "sales volume", "number of orders", "debt ratio" and other specific objectives.

[0371] G42. Compare the forecast results with the data under normal operating mode to identify abnormal data;

[0372] Analyze the output of the model to find data points that are significantly different from the normal operating model. You can use the difference comparison method. If the difference exceeds the preset threshold, it is defined as an outlier.

[0373] G43. Based on the abnormal data, determine the potential operating problems of the enterprise, where the potential operating problems are a combination of one or more problems in inventory backlog, customer churn, finance, supply chain disruption, and market competition.

[0374] Based on the model's prediction results and data analysis, the following potential business problems were discovered:

[0375] 1. Inventory backlog problem: The inventory level of some products is much higher than normal due to overproduction or decreased market demand.

[0376] 2. Customer churn problem: The rate at which customers decrease accelerates over a period of time because of product quality issues or changes in competitor strategies.

[0377] 3. Financial risk: The company’s debt ratio suddenly increases and faces financial risks.

[0378] 4. Supply chain disruption: The reduction in the purchase volume of certain key raw materials is due to problems with suppliers or supply chain disruptions.

[0379] 5. Intensified market competition: The company’s market share suddenly drops because its competitors have launched more competitive products.

[0380] Based on the business issues discovered, detailed reports are written to provide decision makers with clear and accurate business information.

[0381] Provide suggestions and improvement measures in the report to help companies solve problems.

[0382] Using AI models to analyze actual operating data can not only uncover potential operating problems, but also provide strong decision-making support for enterprises.

[0383] In another embodiment, the second deep model includes a teacher model and a student model, and constructing a second deep learning model, and training the second deep learning model using the second training set includes:

[0384] H1. Construct teacher model and student model;

[0385] In many applications, the large model (teacher model) is too large and computationally intensive, making it difficult to deploy in resource-constrained environments. By using a teacher-student architecture, embodiments of the present invention can transfer the knowledge of the large model to a smaller, more lightweight model (student model) without sacrificing too much accuracy.

[0386] H2. Use the second training set to train the teacher model;

[0387] Train a deep and complex model on a large amount of data to achieve high accuracy.

[0388] This model will serve as a source of knowledge and provide guidance for the student models.

[0389] Ensure that the training data set is representative and can cover various scenarios for exploring potential business problems of the enterprise.

[0390] The teacher model can be a deep convolutional neural network or a deep recurrent neural network.

[0391] The network is trained until satisfactory accuracy is achieved. For the specific training process, please refer to the description of the above embodiment.

[0392] H3. Use the output of the teacher model as a soft target to train the student model;

[0393] A lightweight network architecture is selected as the student model, and the output of the teacher model is used as a "soft target" to train the student model. For each student model, the training process uses not only the original labels but also the predictions of the teacher model as additional soft targets.

[0394] Lightweight network architectures usually have fewer parameters and simpler structures, which makes them more efficient in computing and storage and suitable for deployment in resource-constrained environments. The following are some common examples of lightweight network architectures:

[0395] 1. MobileNet

[0396] MobileNet uses depth-wise separable convolutions to reduce computation and model size.

[0397] It is particularly suitable for mobile and embedded vision applications.

[0398] 2. SqueezeNet

[0399] SqueezeNet uses 1x1 convolution (or point convolution) to reduce the number of parameters.

[0400] By using smaller convolution kernels and deeper network structures, it significantly reduces the model size while maintaining the same accuracy.

[0401] 3.ShuffleNet

[0402] ShuffleNet introduces point convolution and channel shuffle operations to optimize the computational efficiency of the model.

[0403] Channel shuffling can increase the diversity of features and thus improve the representation ability of the model.

[0404] 4. EfficientNet

[0405] EfficientNet achieves better performance by uniformly expanding the depth, width, and resolution of the network.

[0406] It uses a compound scaling approach to obtain efficient models under different resource constraints.

[0407] 5. TinyML

[0408] TinyML is a lightweight deep learning framework designed for microcontrollers and edge devices.

[0409] It provides a range of optimization tools to further reduce model size and computational effort.

[0410] When choosing a suitable lightweight network architecture as a student model, you need to consider the complexity of the model, the computational and storage requirements, and the characteristics of the target task. For example, if the goal is to perform image classification on an embedded device, MobileNet or SqueezeNet is a good choice.

[0411] H4. In the process of training the student model, the temperature scaling technology is used to adjust the probability output of the teacher model, and the student model is iteratively trained after each adjustment of the probability output of the teacher model, so that the potential operating problems of the enterprise can be mined through the student model.

[0412] Adjusting the probability distribution of the teacher model output to make it softer helps the training of the student model.

[0413] In the process of teacher-student knowledge distillation, temperature scaling technology is a commonly used method to make the output probability distribution of the teacher model more "soft". Such a "soft" probability distribution contains more information and is helpful for the training of the student model.

[0414] The following are the specific implementation steps of temperature scaling technology:

[0415] Step H41: Calculate the original probability distribution

[0416] Perform a forward pass on the training data using the teacher model to obtain the original output of the model, usually unnormalized logits.

[0417] Step H42: Apply temperature value

[0418] Divide the original logits by a temperature value (T). The temperature value (T) is usually greater than 1, which can make the probability distribution more "soft".

[0419] Step H43: Calculate the probability distribution of "soft"

[0420] Use the softmax function to convert the adjusted logits into a probability distribution.

[0421] Step H44: Training with "soft" probability as target

[0422] When training the student model, the above "softened" probability distribution is used as the target instead of the original hard labels.

[0423] Typically, the loss function is the cross entropy between the probability distribution of the student model and the "softened" probability distribution of the teacher model.

[0424] Using temperature scaling technology, the student model not only learns the original label information of the data, but also learns the knowledge of the teacher model, which helps to improve the generalization ability and accuracy of the student model.

[0425] Deploy lightweight student models in target environments, such as mobile devices, embedded systems, etc., and use student models to explore potential business problems of enterprises.

[0426] Due to its lightweight nature, the student model is particularly suitable for real-time analysis and prediction in resource-constrained environments. The following are the specific steps on how to use the student model to discover potential business problems of an enterprise:

[0427] Step J1: Data preprocessing

[0428] Export the business data that needs to be analyzed from the enterprise management system, such as historical business analysis reports.

[0429] Clean and preprocess the data, such as dealing with missing values, outliers, and data normalization, so that it meets the input requirements of the student model.

[0430] Step J2: Model prediction

[0431] Input the preprocessed operating data into the student model.

[0432] Record the output of the model, which are the predicted values ​​of business indicators, anomaly detection scores, etc.

[0433] Step J3: Analyze the prediction results

[0434] Analyze the student model's predictions to identify data points or areas that differ significantly from normal operating patterns.

[0435] Based on these differences, identify operational issues.

[0436] Step J3: Analyze the prediction results

[0437] Analyzing the prediction results of the student model is a key step in mining potential business problems of the enterprise. In this step, the embodiment of the present invention will analyze the output of the model in detail to determine the business problems.

[0438] Step J3.1: Collect forecast data

[0439] Compare the student model's predictions with actual operating data.

[0440] Generate a difference value for each prediction result, such as the difference between the predicted value and the actual value, the deviation between the predicted trend and the actual trend, etc.

[0441] Step J3.2: Set threshold

[0442] Based on historical data and business experience, thresholds are set for each operating indicator. When the difference value exceeds these thresholds, it indicates that there is an operating problem.

[0443] Step J3.3: Identify outliers

[0444] Use statistical or machine learning methods to identify anomalies in forecast data.

[0445] The outliers are caused by inaccurate model predictions and problems in actual operations.

[0446] Step J3.4: Analyze the cause of the abnormality

[0447] For each data point identified as an anomaly, deep analysis is performed to understand the reasons behind it.

[0448] For example, if sales suddenly drop, it may be because of an unsuccessful new product launch, increased market competition, or a wrong marketing strategy.

[0449] Step J3.5: Summarize the business problem

[0450] Based on the above analysis, summarize the operational problems faced by the company.

[0451] For example, improper inventory management, product quality issues, poor marketing strategies, etc.

[0452] Through the above steps, the embodiment of the present invention can conduct a detailed analysis of the prediction results of the student model, dig out the business problems, and provide valuable references for the next decision.

[0453] In another embodiment, assuming that the deep learning model includes a parallel model of a CNN and a long short-term memory (LSTM) model, the trained second deep learning model is used to mine potential business problems of the enterprise, including steps K1-K3:

[0454] K1. Use the CNN model to analyze image-based business data and identify data features of image-based business data;

[0455] The specific process is described in E1-E8. Use the trained CNN model to forward propagate the business data and extract the output of the middle layer as the data features of image-based business data, such as financial data, human resources data, etc.

[0456] K2. Use the LSTM model to analyze the sequence-type business data and identify the data characteristics of the sequence-type business data;

[0457] LSTM (Long Short-Term Memory) is a variant of RNN, specially designed to handle long-term dependency problems. LSTM can better capture long-term patterns in time series data through its unique gate mechanism.

[0458] The following are the specific steps for using the LSTM model to process business data:

[0459] Step K21: Data integration and preprocessing

[0460] Obtain serial business data and integrate all collected data.

[0461] Standardize and normalize the LSTM model to make it within a suitable value range.

[0462] Step K22: Build LSTM model

[0463] Design a basic LSTM architecture, including an input layer, multiple LSTM layers, and an output layer.

[0464] Choose the appropriate number of LSTM units, activation function, and other parameters.

[0465] Step K23: Train the LSTM model

[0466] Use the integrated business data to train the LSTM model.

[0467] Use time series cross-validation or other methods to optimize model parameters to ensure that the model does not overfit.

[0468] Step K24: Model evaluation

[0469] Evaluate the performance of the LSTM model on the test set or validation set.

[0470] Use mean squared error (MSE), mean absolute error (MAE), or other appropriate evaluation metrics.

[0471] Adjust the model structure or parameters based on the evaluation results.

[0472] Step K25: Extract time series patterns

[0473] Use the trained LSTM model to predict the integrated business data.

[0474] Analyze model output to identify long-term trends and patterns in operating data.

[0475] These patterns can help companies understand the historical performance and future trends of their business activities.

[0476] Through the above steps, the LSTM model can not only help enterprises capture long-term patterns in business data, but also provide decision makers with valuable insights into future business trends.

[0477] K3. Integrate the recognition results of the CNN model and the LSTM model to analyze the potential operating problems of the enterprise.

[0478] The fusion process can be weighted fusion or other forms of fusion. Exemplarily, the prediction results of CNN and LSTM are integrated to obtain a comprehensive business data forecast. Considering that CNN mainly captures short-term changes, while LSTM captures long-term trends, the merging result should fully consider information from these two aspects. Therefore, when the embodiment of the present invention combines the prediction results of RNN and LSTM models, the embodiment of the present invention can obtain an in-depth understanding of the business data of the enterprise and identify possible business problems from the short-term and long-term trends.

[0479] The following are the specific steps for analyzing business problems:

[0480] Step K31: Merge model outputs

[0481] Integrate the prediction results of RNN and LSTM to obtain a comprehensive business data forecast.

[0482] Considering that RNN mainly captures short-term changes, while LSTM captures long-term trends, the combined result should fully consider the information from both aspects.

[0483] Step K32: Set the warning threshold

[0484] Set warning thresholds for each operating indicator based on historical operating data and business experience.

[0485] When forecasts fall outside these thresholds, it could be a sign of an operational problem.

[0486] Step K33: Identify anomalies and deviations

[0487] Compare forecasts with actual operating data and identify any significant deviations.

[0488] Analyze the reasons for these deviations to determine whether they are due to model errors or actual operational problems.

[0489] Step K34: Summarize the business problem

[0490] Based on the above analysis, summarize the business problems that the company may face, such as:

[0491] 1. Sales volume is lower than expected: The forecasted sales volume is much higher than the actual sales volume, which may be due to the failure of new product launch, increased market competition or improper marketing strategy.

[0492] 2. Inventory backlog: Forecasted inventory levels are consistently lower than actual levels, which may lead to excessive inventory and high inventory costs.

[0493] 3. Decreased production efficiency: The long-term production volume predicted by the LSTM model is lower than the actual production volume, which may be due to production line problems or decreased employee efficiency.

[0494] Step K35: Provide improvement suggestions

[0495] Provide corresponding suggestions and improvement measures for each identified operational problem.

[0496] For example, to increase sales, you can recommend conducting market research to understand consumer needs, optimize products, or adjust marketing strategies.

[0497] Through the above steps, the embodiment of the present invention can effectively utilize the prediction results of the CNN and LSTM models, deeply analyze and explore potential business problems of the enterprise, and provide specific suggestions for decision makers.

[0498] S105. Evaluate the potential of the enterprise at different stages of operation.

[0499] Assessing the potential of an enterprise is an important part of post-investment management. Using machine learning algorithms, we can learn the laws of enterprise growth from historical data and predict the potential of enterprises at different stages.

[0500] like Figure 5 As shown, S105 specifically includes L1-L4:

[0501] L1: Collect historical operating data;

[0502] Similar to S104, it is necessary to first collect a large amount of historical operating data, such as sales, number of customers, market share, etc.

[0503] L2: pre-processing the historical operating data;

[0504] Data cleaning and preprocessing are performed based on the quality and completeness of the data. The preprocessing includes:

[0505] Extract meaningful features from raw data, such as sales growth rate, customer growth rate, etc.

[0506] Normalize or standardize the features to ensure the stability and accuracy of the model.

[0507] L3: constructing a machine learning model, and using the preprocessed historical operating data to train the machine learning model;

[0508] Select the appropriate machine learning model based on the characteristics of the evaluation task. For example, for regression tasks, you can choose models such as linear regression and decision tree regression; for classification tasks, you can choose models such as logistic regression and support vector machine.

[0509] Considering the amount of data and feature dimensionality, choose an appropriate model complexity.

[0510] Use the training set data to train the machine learning model.

[0511] Use the validation set data to validate the model and evaluate the accuracy and generalization ability of the model.

[0512] Taking the decision tree as an example, the decision tree is a commonly used machine learning model that segments and classifies data through a tree structure. In the scenario of evaluating the potential of an enterprise, the decision tree can help the embodiment of the present invention predict the potential of an enterprise at different stages based on different features, such as sales, number of customers, etc.

[0513] Step L31: Select decision tree algorithm

[0514] Select the appropriate decision tree algorithm based on the characteristics of the evaluation task. For example, for classification tasks, you can choose CART, ID3, C4.5 and other algorithms; for regression tasks, you can choose regression tree.

[0515] Step L32: Set decision tree parameters

[0516] Set parameters such as the depth of the decision tree, the minimum number of samples for leaf nodes, and feature selection criteria to control the complexity of the model.

[0517] Use the cross-validation method to find the optimal parameter combination to achieve the best model performance.

[0518] Step L33: Training decision tree model

[0519] Use the training set data to train the decision tree model.

[0520] Monitor the model training process, such as changes in the loss function, tree depth, etc., to ensure stable convergence of the model.

[0521] Step L34: Verify the decision tree model

[0522] Use the validation set data to validate the decision tree model.

[0523] Evaluate the model's accuracy, recall, F1 score and other indicators to ensure the model's generalization ability.

[0524] Step L35: Model interpretation and visualization

[0525] The tree structure of the decision tree is used to visualize the model to help decision makers understand the decision logic of the model.

[0526] Use model interpretation tools, such as SHAP, LIME, etc., to further explain the model's prediction results.

[0527] Through the above steps, using the decision tree model, the potential of an enterprise at different stages can be accurately evaluated and strong decision-making support can be provided for the enterprise.

[0528] L4: Using the machine learning model to evaluate the potential of the enterprise at different operating stages.

[0529] Apply the trained model to real data to predict the potential of enterprises at different stages.

[0530] Based on the output of the model, provide enterprises with growth strategies and decision-making recommendations.

[0531] Use model interpretation tools, such as SHAP and LIME, to explain the model's prediction results and help decision makers understand the model's decision logic.

[0532] Taking a Guizhou technology company as an example, this embodiment of the present invention will demonstrate how to use a trained AI model to predict the potential of a company at different stages.

[0533] Guizhou XX Technology Co., Ltd.

[0534] Guizhou XX Technology Co., Ltd. is a technology company located in Guizhou, mainly engaged in the research and development of smart hardware and AI technology.

[0535] Step L41: Data acquisition and preprocessing

[0536] Obtain business data from the ERP system, CRM system and other internal systems of Guizhou XX Technology Co., Ltd.

[0537] Preprocess the data, such as normalizing and filling missing values, to ensure that the data quality is consistent with the input requirements of the model.

[0538] Step L42: Model prediction

[0539] The processed data is input into the trained AI model.

[0540] Get the model's prediction results, which represent the potential scores of the enterprise at different stages.

[0541] Step L43: Potential Assessment and Analysis

[0542] According to the prediction results of the model, the potential of Guizhou X Technology Co., Ltd. in the following stages is analyzed:

[0543] 1. Start-up stage: The score is 80 / 100. At this stage, the company has high innovation ability and market response speed, but it still needs to strengthen team building and accumulate management experience.

[0544] 2. Growth stage: The score is 85 / 100. The company has established a certain brand awareness in the market and has a stable customer base, but needs to increase R&D investment to maintain technological leadership.

[0545] 3. Mature stage: The score is 75 / 100. The enterprise has stable income and profits, but faces the challenges of market saturation and intensified competition and needs to find new growth points.

[0546] 4. Decline stage: The score is 60 / 100. The company's growth slows down and needs to transform or find a new business model.

[0547] Step L44: Provide decision-making recommendations

[0548] Based on the results of the potential assessment, provide decision-making recommendations for Guizhou XX Technology Co., Ltd.

[0549] For example, increase R&D investment, strengthen cooperation with universities and research institutions, and explore overseas markets.

[0550] Through the above steps, the embodiment of the present invention can not only evaluate the potential of Guizhou XX Technology Co., Ltd. at different stages, but also provide it with targeted decision-making suggestions to help it continue to grow and develop.

[0551] An embodiment of the present application also provides a post-investment management information system, in which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0552] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0553] The embodiment of the present application proposes an information system and management method for post-investment management. Through the system and method, the problems of enterprises in the investment process due to non-standard data, lack of unified management and long-term data accumulation are solved. At the same time, the cloud-edge computing-terminal architecture, multi-source heterogeneous data standardization and large-capacity data storage technology, and AI technology are introduced to realize the intelligent management of post-investment enterprises.

[0554] Embodiment 2

[0555] Based on the same inventive concept as the post-investment management method in the aforementioned embodiment, the present application also provides a computer-readable storage medium and / or system, on which a computer program is stored, and when the computer program is executed by a processor, the method in embodiment one is implemented.

[0556] Embodiment 3

[0557] The embodiment of the present invention also provides a post-investment management information system 6000, such as Figure 6 As shown, it includes a memory 64 and a processor 61, the memory stores computer executable instructions, and the processor implements the above method when running the computer executable instructions on the memory. In practical applications, the system can also include other necessary components, including but not limited to any number of input systems 62, output systems 63, processors 61, controllers, memories 64, etc., and all methods that can implement the embodiments of the present application are within the scope of protection of the present application.

[0558] The memory includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or compact disc read-only memory (CD-ROM), which is used for related instructions and data.

[0559] The input system 62 is used to input data and / or signals, and the output system 63 is used to output data and / or signals. The output system 63 and the input system 62 may be independent devices or an integrated device.

[0560] The processor may include one or more processors, for example, one or more central processing units (CPUs). When the processor is a CPU, the CPU may be a single-core CPU or a multi-core CPU. The processor may also include one or more dedicated processors, which may include GPUs, FPGAs, etc., for accelerated processing.

[0561] The memory is used to store program codes and data of the network device.

[0562] The processor is used to call the program code and data in the memory to execute the steps in the above method embodiment. For details, please refer to the description in the method embodiment, which will not be repeated here.

[0563] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the division of the unit is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling, direct coupling, or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the system or unit can be electrical, mechanical or other forms.

[0564] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0565] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media integrations. The available medium may be a read-only memory (ROM), or a random access memory (RAM), or a magnetic medium, such as a floppy disk, a hard disk, a tape, a disk, or an optical medium, such as a digital versatile disc (DVD), or a semiconductor medium, such as a solid state disk (SSD), etc.

[0566] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A post-investment management method, It is characterized in that include: Build a post-investment management information system; The post-investment management information system collects enterprise operating data; Analyzing the collected business operation data using a first deep learning model; Based on the analysis results, potential operational problems of the enterprise are explored; Assess the potential of the enterprise at different stages of operation; The first deep learning model is used to analyze the collected business operation data, including: Preprocessing the business operation data of the enterprise; Dividing the pre-processed enterprise operation data into first image type data and first sequence type data; Construct convolutional neural network (CNN) model and recurrent neural network (RNN) model; The CNN model is trained using historical image data, and the RNN model is trained using historical sequence data; Analyze the first image data using the trained CNN model to output a first enterprise operating status and a first decision suggestion; Analyze the first sequence data using the trained RNN model to output a second enterprise operating status and a second decision suggestion; Integrate the first enterprise operation status, the first decision suggestion, the second enterprise operation status and the second decision suggestion to generate an enterprise operation analysis report; Among them, according to the analysis results, the potential operating problems of the enterprise are explored, including: Obtain historical business operating reports; Preprocessing the historical business operation reports, and dividing the preprocessed historical business operation reports into a second training set and a second test set; Building a second deep learning model; Training the second deep learning model using the second training set; Using the trained second deep learning model to mine potential business problems of the enterprise; Wherein, using the trained second deep learning model to mine potential business problems of the enterprise includes: Inputting the business data of the enterprise into the trained second deep learning model and outputting a prediction result; Comparing the predicted results with data under normal operating modes to identify abnormal data; Based on the abnormal data, determine potential operating problems of the enterprise, where the potential operating problems are a combination of one or more problems in inventory backlog, customer loss, finance, supply chain disruption, and market competition; The second deep learning model includes a teacher model and a student model, and then constructing the second deep learning model, and training the second deep learning model using the second training set, including: Construct teacher model and student model; Using the second training set to train the teacher model; Using the output of the teacher model as a soft target to train the student model; In the process of training the student model, a temperature scaling technique is used to adjust the probability output of the teacher model, and the student model is iteratively trained after each adjustment of the probability output of the teacher model, so that the potential operating problems of the enterprise are mined through the student model; Alternatively, the second deep learning model includes a parallel model of a CNN and a long short-term memory (LSTM) model, and the trained second deep learning model is used to mine potential business problems of the enterprise, including: Use the CNN model to analyze image-based business data and identify data features of the image-based business data; Use the LSTM model to analyze the sequence-type business data and identify the data features of the sequence-type business data; Fusion the recognition results of the CNN model and the LSTM model to analyze potential operating problems of the enterprise; Among other things, the potential of the enterprise at different stages of operation is assessed, including: Collect historical operating data; Preprocessing the historical operating data; Constructing a machine learning model, and using the preprocessed historical operating data to train the machine learning model; The machine learning model is used to evaluate the potential of the enterprise at different stages of operation.

2. The method according to claim 1, It is characterized in that The CNN model is trained using historical image data, including: Set initial values ​​for the model's weights; Set the cross entropy loss function; Set the Adam optimizer and set a learning rate; Set the batch size; Divide the historical image class data into the set batch size for forward propagation to obtain the predicted output; Use the loss function to compare the predicted output of the CNN model with the true label and calculate the loss; Performing back propagation, including: calculating the gradient of each layer using the chain rule, and updating the weights and biases of the CNN model using an optimizer based on the gradient; At the end of each training cycle, the performance of the CNN model is evaluated using the validation set; The RNN model is trained using historical sequence data, including: Initializing parameters of the RNN model; Input a batch of sequence data into the RNN model; Calculating each time step of the sequence data, wherein for each time step, the current input and the hidden state of the previous time step are used to generate the hidden state of the current time step; In the last time step, a fully connected layer is used to obtain the prediction output; Use the loss function to compare the predicted output of the RNN model with the true label and calculate the loss; Perform back propagation; The performance of the RNN model was evaluated using the validation set.

3. The method according to claim 1, It is characterized in that After the post-investment management information system collects the enterprise operation data, the method further includes: Build a large-scale storage system and establish a data redundancy backup mechanism in the storage system; Monitor the health status of storage nodes; When any storage node goes down, determine the data blocks that need to be migrated; An optimal data migration path is selected, and a data migration operation is started to migrate the data blocks that need to be migrated.

4. A post-investment management information system, It is characterized in that A computer program is stored in the system, and when the computer program is executed by a processor, the steps of any method of claims 1-3 are implemented.