Enterprise asset digital integrated management platform based on deep learning

Abstract

The invention relates to the technical field of enterprise asset management, and provides an enterprise asset digital integrated management platform based on deep learning, which comprises a DBN-RL financial analysis decision optimization module, an LSTM-GCN investment high-risk early warning module and an SOM-GAN fund flow cloud monitoring module. According to the enterprise asset digital integrated management platform based on deep learning, all-around improvement is brought to enterprise asset management through cooperative work of all the modules. In the aspect of financial analysis and decision, the DBN-RL module provides the capabilities of deeply mining data and optimizing decision, so that an enterprise can make a more intelligent decision in a complex financial environment, and the financial stability and profitability of the enterprise are enhanced. The LSTM-GCN module plays a key role in investment management, and through accurate trend prediction and risk early warning, the enterprise is helped to reduce the risk in the investment activity, improve the return on investment, and guarantee the safety and appreciation of enterprise investment assets. And the SOM-GAN module effectively monitors the fund flow, so that the abnormal fund condition can be found in time.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise asset management, and particularly to an enterprise asset digital integration management platform based on deep learning. Background Art

[0002] Under the background of current global economic integration, enterprises are facing an increasingly complex business environment and fierce market competition. In order to maintain a competitive edge, enterprises urgently need to improve asset management efficiency and reduce operational risks through digital transformation. As an important tool for digital transformation, the enterprise asset digital integration management platform aims to integrate advanced information technologies to achieve efficient asset management, real-time risk monitoring, and optimization of financial decision-making. However, existing asset management platforms have limitations in financial analysis, risk warning, and monitoring of capital flow, such as lacking deep learning and pattern recognition capabilities, being unable to provide accurate decision support and risk warning, and having the ability to monitor capital flow in real time. Therefore, it is particularly urgent to develop an enterprise asset digital integration management platform integrating advanced algorithms.

[0003] Therefore, this solution specifically proposes an enterprise asset digital integration management platform based on deep learning to solve the above problems. Summary of the Invention

[0004] To overcome the defects of the prior art, the purpose of the present invention is to provide an enterprise asset digital integration management platform based on deep learning.

[0005] To achieve the above object, the technical solution of the present invention is implemented as follows: An enterprise asset digital integration management platform based on deep learning, comprising: A DBN-RL financial analysis and decision optimization module, where the deep belief network (DBN) is used to extract deep features of financial data, map the original data to a high-dimensional feature space through multi-layer non-linear transformation to capture the complex relationships between data, and reinforcement learning (RL) is used to simulate the financial decision-making process, learn the optimal strategy through interaction with the environment, and continuously try and adjust decisions with the goal of maximizing the cumulative reward. The DBN-RL algorithm combines the feature extraction ability of DBN and the decision optimization ability of RL to provide accurate financial analysis based on historical and real-time data for enterprises and optimize financial decisions; An LSTM-GCN investment high-risk warning module, where the long short-term memory network (LSTM) is used to process time-series investment data and capture the trends of investment data changing over time, and the graph convolutional network (GCN) is used to analyze the structural features of the investment portfolio to identify potential risk points. The LSTM-GCN algorithm combines the time-series prediction ability of LSTM and the structural feature analysis ability of GCN to achieve accurate prediction and warning of investment portfolio risks; The SOM-GAN fund flow cloud monitoring module, where the self-organizing map (SOM) is used to map fund flow data to a low-dimensional space to reveal the patterns and laws of fund flow, and clustering analysis of fund flow is achieved by constructing a topological structure to simulate the internal distribution of data. The generative adversarial network (GAN) is used to generate the distribution of fund flow data to assist in detecting abnormal flows. It consists of a generator and a discriminator. Through continuous adversarial training, the generator generates fund flow data similar to real data, and the discriminator distinguishes between real data and generated data. The SOM-GAN algorithm combines the clustering analysis ability of SOM and the abnormal detection ability of GAN to achieve real-time monitoring and abnormal detection of fund flow.

[0006] Preferably, in the DBN-RL financial analysis decision optimization module, the multi-layer non-linear transformation of the deep belief network (DBN) specifically includes the stacking of at least three restricted Boltzmann machines (RBMs).

[0007] Preferably, in the LSTM-GCN investment high-risk warning module, the input data of the long short-term memory network (LSTM) includes, but is not limited to, historical time series data of stock prices and market indices.

[0008] Preferably, in the SOM-GAN fund flow cloud monitoring module, the topological structure of the self-organizing map (SOM) is a hexagonal topological structure.

[0009] Preferably, in the DBN-RL financial analysis decision optimization module, the environment of reinforcement learning (RL) includes the internal financial environment and the external market environment of the enterprise.

[0010] Preferably, in the LSTM-GCN investment high-risk warning module, in the graph structure of the investment portfolio constructed by the graph convolutional network (GCN), the nodes represent investment assets, and the edges represent the correlation relationships between the assets.

[0011] Preferably, in the SOM-GAN fund flow cloud monitoring module, the discriminator of the generative adversarial network (GAN) adopts a multi-layer perceptron (MLP) structure.

[0012] Preferably, the platform further includes a data interface module for connecting to the enterprise's internal financial system, investment management system, and fund management system to obtain financial data, investment data, and fund flow data.

[0013] Preferably, the data interface module uses Secure Sockets Layer (SSL) encryption technology to ensure the security of data transmission.

[0014] Preferably, the platform further includes a visual display module for displaying the financial analysis results, investment risk warning information, and fund flow monitoring results in the form of charts.

[0015] The beneficial effects of the present invention are reflected in: (1) DBN-RL Financial Analysis and Decision Optimization Module Accurate Financial Analysis and Decision Optimization Through the stacking of multi-layer restricted Boltzmann machines (RBMs) of deep belief networks (DBNs) (specifically three layers with clearly set numbers of neurons in each layer) and strict data preprocessing (filling missing values with the mean, handling outliers using the 3-sigma principle, and normalizing in the [0,1] interval), the complex non-linear relationships in financial data can be deeply explored, and deep features can be accurately extracted. This helps enterprises comprehensively and meticulously understand their financial status and provides a richer and more accurate information basis for decision-making.

[0016] Reinforcement learning (RL) can simulate the actual financial decision-making process and learn the optimal strategy by clearly constructing an environment that includes internal financial and external market factors and reasonably defining a reward function linked to the enterprise's financial goals (such as setting specific reward values according to profit growth and risk levels). This enables enterprises to make more scientific and reasonable decisions in investment, financing, and cost control, effectively improving the quality of decisions and achieving the goals of maximizing profits and minimizing risks.

[0017] (2) LSTM-GCN High-Risk Investment Warning Module Effective Investment Risk Warning Based on historical time series data such as stock prices and market indices with a long time span (5 years) and high frequency (daily), the long short-term memory network (LSTM) constructs a network with a specific structure (two layers with 128 and 64 neurons respectively) and is trained using the Adam optimization algorithm (learning rate 0.001), which can accurately capture the changing trends of investment data over time. This provides forward-looking information for investment decisions and helps enterprises anticipate market trends in advance.

[0018] The graph convolutional network (GCN) can deeply analyze the structural characteristics of the investment portfolio and accurately identify potential risk points by reasonably constructing the investment portfolio graph structure (determining asset correlation relationships based on a Pearson correlation coefficient of 0.5) and carefully designing node and edge features (such as market value, price-to-earnings ratio, and correlation coefficient to quantify the correlation strength, etc.). Combining with the trend prediction of LSTM, when the risk value exceeds the preset threshold of 0.3, it issues a timely warning, effectively reducing investment risks and protecting the safety of enterprise assets.

[0019] (3) SOM-GAN Capital Flow Cloud Monitoring Module Efficient Capital Flow Monitoring and Anomaly Detection The self-organizing map (SOM) uses a hexagonal topology (10×10 dimension, initial learning rate decreasing from 0.5) to process the fund flow data that has been denoised and normalized to [0,1], which can clearly reveal the patterns and rules of fund flow and accurately identify different types of fund flow patterns through cluster analysis. At the same time, by setting a threshold (0.2) for monitoring the change of the cluster center, abnormal signs of fund flow can be detected in a timely manner.

[0020] The generative adversarial network (GAN) conducts 500 rounds of adversarial training through a generator with a specific structure (a four-layer fully connected neural network with a decreasing number of neurons and a ReLU activation function) and a discriminator (a three-layer MLP structure with a LeakyReLU activation function), and can generate a distribution similar to the real fund flow data. Using the generated data to assist in detection, when the mean squared error between the actual fund flow and the generated data distribution exceeds 0.1, it is determined as abnormal, which improves the accuracy and efficiency of fund flow anomaly detection and ensures the normal turnover of enterprise funds.

[0021] (4) Data interface module Safe and reliable data interaction The RESTful API interface protocol is used to connect various internal systems of the enterprise, and the data is converted into JSON format, ensuring the compatibility and stability of data transmission between systems and realizing seamless data interaction.

[0022] The Secure Sockets Layer (SSL) encryption technology (2048-bit RSA key exchange, AES-256 data encryption, two-way identity authentication, SHA-256 integrity verification) ensures the confidentiality, integrity and legality of data during transmission, effectively preventing data leakage, tampering and illegal access, and protecting the security of the enterprise's core asset data.

[0023] (5) Visualization display module Intuitive decision support According to different data types (financial analysis results, investment risk warnings, fund flow monitoring), select appropriate chart types (line charts, bar charts, pie charts) for display and provide interactive functions, enabling enterprise managers to intuitively and quickly understand complex asset data information. This helps to improve decision-making efficiency, enabling managers to make timely and accurate decisions based on clear data presentations and optimize enterprise asset management.

[0024] In summary, the enterprise asset digital integration management platform based on deep learning brings comprehensive improvements to enterprise asset management through the collaborative work of each module. In terms of financial analysis and decision-making, the DBN-RL module provides the ability to deeply mine data and optimize decisions, enabling enterprises to make more informed decisions in complex financial environments, enhancing the financial stability and profitability of enterprises. The LSTM-GCN module plays a key role in investment management. By accurately predicting trends and warning of risks, it helps enterprises reduce risks in investment activities, improve the return on investment, and ensure the safety and appreciation of enterprise investment assets. The effective monitoring of fund flows by the SOM-GAN module can promptly detect abnormal fund situations, ensure the stability of the enterprise's fund chain, and avoid business crises caused by fund risks. The secure data transmission channel constructed by the data interface module protects the security of enterprise data assets and provides a solid foundation for the stable operation of the entire platform. The visualization display module converts complex data into intuitive charts, not only improving decision-making efficiency but also promoting information sharing and collaborative work among different departments within the enterprise. In summary, the platform integrates the advantages of deep learning technology in all aspects of enterprise asset management, comprehensively improving the efficiency, security, and scientific nature of enterprise asset management, helping enterprises maintain their advantages in fierce market competition and achieve sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In the drawings: Figure 1 is a schematic structural diagram of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The present invention will be further described in detail below with reference to the drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of the invention, rather than all of the embodiments. Without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the invention without creative efforts shall fall within the scope of protection of the invention.

[0027] Please refer to the attached Figure 1 specification. The present invention provides an enterprise asset digital integration management platform based on deep learning: I. Overall Architecture of the Platform The enterprise asset digital integration management platform aims to integrate various asset-related data of the enterprise, such as finance, investment, and fund management, and achieve comprehensive and accurate management of enterprise assets through deep learning algorithms. The platform mainly includes a DBN-RL financial analysis and decision-making optimization module, an LSTM-GCN investment high-risk warning module, a SOM-GAN fund flow cloud monitoring module, a data interface module, and a visualization display module.

[0028] II. Specific Implementation of Each Module DBN-RL Financial Analysis and Decision Optimization Module Deep Belief Network (DBN) Stacking of Restricted Boltzmann Machines (RBMs): Construct three layers of Restricted Boltzmann Machines (RBMs). The number of visible layer neurons in the first layer of RBM is determined to be 500 according to the feature dimension of the input financial data, and the number of hidden layer neurons is set to 250. The number of visible layer neurons in the second layer of RBM is 250, and the number of hidden layer neurons is 125. The number of visible layer neurons in the third layer of RBM is 125, and the number of hidden layer neurons is 60. The learning rate is set to 0.01.

[0029] Data Preprocessing: Before inputting the financial data into the DBN, preprocess the data. Use the method of mean filling to handle missing values, and identify and handle outliers according to the 3-sigma principle. Then perform standardization processing on the data to map the feature values of the data to the interval [0, 1].[[]]

[0030] Feature Extraction Process: Input the preprocessed financial data into the stacked RBMs in sequence. Through multi-layer non-linear transformation, map the original financial data to a high-dimensional feature space, and automatically capture the complex non-linear relationships between financial data.

[0031] Reinforcement Learning (RL) Environment Construction: Define the environment of reinforcement learning. The internal financial environment of the enterprise includes the enterprise's financial statement data (balance sheet, income statement, cash flow statement, etc.), financial indicators (solvency indicators, profitability indicators, operating capacity indicators, etc.), and relevant information such as financial budgets and cost controls. The external market environment includes macroeconomic data (GDP growth rate, inflation rate, interest rate, etc.), industry competition situation, market price fluctuations and other factors.

[0032] State Representation: Integrate the financial data features extracted by the DBN and other relevant information in the environment to construct the state representation of reinforcement learning. The state vector includes financial data features and key indicators of the current market environment, such as GDP growth rate, interest rate, etc.

[0033] Action Definition and Policy Learning: Define actions in reinforcement learning, such as investment decisions (investment amount, investment direction, etc.), financing decisions (financing method, financing amount, etc.), and cost control decisions (cost-cutting projects, amplitude, etc.). RL interacts with the environment, tries different actions according to the current state, and learns the optimal policy based on the reward signal feedback from the environment. The design of the reward function comprehensively considers the financial goals of the enterprise, such as profit maximization, risk minimization, etc. For example, when the profit growth brought by the investment decision is 10% and the risk is within 5%, the reward value is 8; when the investment decision results in a 5% financial loss or the risk exceeds 8%, the reward value is -6. Through continuous attempts and adjustments, RL finally learns the policy of choosing the optimal action in different states, thus optimizing the financial decisions of the enterprise.

[0034] LSTM-GCN High-Risk Early Warning Module for Investment Long Short-Term Memory Network (LSTM) Data Collection and Arrangement: Collect historical time series data such as stock prices and market indices. The time span of the data is the past 5 years, and the data frequency is daily. Arrange the collected data in chronological order to form a continuous time series dataset.

[0035] Network Structure Construction: Build a two-layer LSTM network. The first layer contains 128 neurons, and the second layer contains 64 neurons. The activation functions of the input gate, forget gate, and output gate are the sigmoid function and the tanh function respectively.

[0036] Training Process: Divide the arranged historical time series data into a training set, a validation set, and a test set according to the ratio of 7:2:1. Use the training set to train the LSTM network, adopt the Adam optimization algorithm with a learning rate of 0.001, and adjust the weight parameters of the network. During the training process, monitor the performance of the model through the validation set. When the loss on the validation set no longer decreases for 3 consecutive epochs, stop the training and use the test set to evaluate the final performance of the model.

[0037] Graph Convolutional Network (GCN) Construction of Investment Portfolio Graph Structure: Take investment assets as nodes in the graph structure, and determine the association relationship between assets to construct edges according to the correlation analysis of assets (calculate the Pearson correlation coefficient between stocks, and the threshold is set to 0.5).

[0038] Feature Engineering: Define features for the nodes in the graph, including the market value, price-earnings ratio, historical return rate, etc. of the assets. The feature of the edge represents the strength of the association between assets and is quantified by the correlation coefficient.

[0039] Risk Identification and Early Warning: Combine the investment data trend information predicted by LSTM with the portfolio structure characteristics analyzed by GCN. When the risk value of the portfolio exceeds the preset threshold of 0.3, an early warning signal is issued.

[0040] SOM-GAN Fund Flow Cloud Monitoring Module Self-Organizing Map (SOM) Data Preprocessing and Mapping: Obtain the fund flow data of the enterprise, remove noise from the data, and perform normalization processing to make its value between [0, 1]. Then input the processed fund flow data into the SOM network with a hexagonal topology. The dimension of the SOM network is 10×10, the initial learning rate is 0.5, and the learning rate gradually decreases as the training progresses.

[0041] Cluster Analysis: According to the mapping results of the SOM network, perform cluster analysis on the fund flow data. By analyzing the differences between different clusters and the data characteristics within the clusters, different types of fund flow patterns are identified. When the change in the cluster center exceeds the preset threshold of 0.2, it may indicate an abnormal fund flow situation.

[0042] Generative Adversarial Network (GAN) Network Construction: Construct a GAN network. The generator uses a four-layer fully connected neural network with the number of neurons being 128, 64, 32, and 16 respectively, and the activation function is the ReLU function. The discriminator uses a three-layer multi-layer perceptron (MLP) structure with the number of neurons being 64, 32, and 1 respectively, and the activation function is the LeakyReLU function.

[0043] Adversarial Training Process: During the training process, the generator and the discriminator perform adversarial training. The generator attempts to generate as realistic fund flow data as possible to deceive the discriminator; the discriminator tries to improve its ability to distinguish between real data and generated data. The number of training rounds is set to 500. In each round of training, the generator and the discriminator are alternately optimized. When the training reaches a certain level, the fund flow data distribution generated by the generator is used to assist in detecting abnormal flows. If the mean square error between the actual fund flow data and the normal data distribution generated by the generator exceeds 0.1, there may be an abnormal fund flow situation.

[0044] Data Interface Module System Connection: Use RESTful API as the data interface protocol to connect to the enterprise's internal financial system, investment management system, and fund management system. Ensure the compatibility and stability of the interface to achieve seamless data transmission. For the differences in data formats of different systems, they are processed through a data conversion and adaptation mechanism to convert various format data into the JSON format that the platform can handle.

[0045] Data Encryption Transmission: During data transmission, Secure Sockets Layer (SSL) encryption technology is adopted. The SSL certificate uses the 2048-bit RSA algorithm for key exchange and the AES-256 algorithm for data encryption. When establishing a connection, two-way authentication is performed to ensure the identity legality of both communication parties. At the same time, integrity verification is carried out on the data during the transmission process, and the SHA-256 algorithm is used to generate a message digest to prevent data tampering.

[0046] Visualization Display Module Chart Design and Selection: According to different display contents, select appropriate chart types. For financial analysis results, such as the change trend of financial indicators, a line chart is used for display; for investment risk warning information, such as the comparison of risk levels of different investment portfolios, a bar chart is used for display; for the monitoring results of fund flows, such as the clustering of fund flows, a pie chart is used for display.

[0047] Data Visualization Presentation: Visualize the result data obtained from each module. Present the financial analysis results, investment risk warning information, and fund flow monitoring results in an intuitive chart form on the user interface of the platform. At the same time, provide interactive functions, and users can click on the charts, filter data, etc. to view specific data details in depth.

[0048] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0049] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

Claims

1. An enterprise asset digital integrated management platform based on deep learning, characterized in that, Including: DBN-RL financial analysis and decision optimization module, where the Deep Belief Network (DBN) is used to extract deep features of financial data, mapping the original data to a high-dimensional feature space through multi-layer non-linear transformation to capture complex relationships between data. Reinforcement Learning (RL) is used to simulate the financial decision-making process, learning the optimal strategy through interaction with the environment, and continuously trying and adjusting decisions with the goal of maximizing cumulative rewards. The DBN-RL algorithm combines the feature extraction ability of DBN and the decision optimization ability of RL to provide accurate financial analysis based on historical and real-time data for enterprises and optimize financial decisions; LSTM-GCN investment high-risk warning module, where the Long Short-Term Memory Network (LSTM) is used to process time-series investment data, capturing the trend of investment data changing over time. The Graph Convolutional Network (GCN) is used to analyze the structural features of the investment portfolio to identify potential risk points. The LSTM-GCN algorithm combines the time-series prediction ability of LSTM and the structural feature analysis ability of GCN to achieve accurate prediction and warning of investment portfolio risks; SOM-GAN fund flow cloud monitoring module, where Self-Organizing Map (SOM) is used to map fund flow data to a low-dimensional space to reveal the patterns and rules of fund flow, achieving clustering analysis of fund flow by constructing a topological structure to simulate the internal distribution of data. Generative Adversarial Network (GAN) is used to generate the distribution of fund flow data to assist in detecting abnormal flows, consisting of a generator and a discriminator. Through continuous adversarial training, the generator generates fund flow data similar to real data, and the discriminator distinguishes between real data and generated data. The SOM-GAN algorithm combines the clustering analysis ability of SOM and the abnormal detection ability of GAN to achieve real-time monitoring and abnormal detection of fund flow.

2. The enterprise asset digital integration management platform based on deep learning according to claim 1, characterized in that In the DBN-RL financial analysis and decision optimization module, the multi-layer non-linear transformation of the Deep Belief Network (DBN) specifically includes the stacking of at least three Restricted Boltzmann Machines (RBM).

3. The enterprise asset digital integration management platform based on deep learning according to claim 1, wherein In the LSTM-GCN investment high-risk warning module, the input data of the Long Short-Term Memory Network (LSTM) includes but is not limited to historical time-series data of stock prices and market indices.

4. A digital integrated management platform for enterprise assets based on deep learning according to claim 1, characterized in that, In the SOM-GAN fund flow cloud monitoring module, the topological structure of the Self-Organizing Map (SOM) is a hexagonal topological structure.

5. A digital integrated management platform for enterprise assets based on deep learning according to claim 1, characterized in that In the DBN-RL financial analysis and decision optimization module, the environment of Reinforcement Learning (RL) includes the internal financial environment and the external market environment of the enterprise.

6. The enterprise asset digital integration management platform based on deep learning according to claim 1, characterized in that In the LSTM-GCN investment high-risk warning module, in the graph structure of the investment portfolio constructed by the Graph Convolutional Network (GCN), nodes represent investment assets, and edges represent the association relationships between assets.

7. An enterprise asset digital integration management platform based on deep learning according to claim 1, characterized in that, In the SOM-GAN fund flow cloud monitoring module, the discriminator of the Generative Adversarial Network (GAN) adopts a Multi-Layer Perceptron (MLP) structure.

8. A digital integrated management platform for enterprise assets based on deep learning according to claim 1, characterized in that, The platform also includes a data interface module for connecting to the enterprise's internal financial system, investment management system, and fund management system to obtain financial data, investment data, and fund flow data.

9. The enterprise asset digital integration management platform based on deep learning according to claim 8, characterized in that, The data interface module adopts the Secure Sockets Layer (SSL) encryption technology to ensure the security of data transmission.

10. A digital integrated management platform for enterprise assets based on deep learning according to claim 1, characterized in that, The platform further includes a visualization display module for displaying the financial analysis results, investment risk warning information, and fund flow monitoring results in the form of charts.

Images