Digital system for asset management based on AI and data processing method thereof

Abstract

The invention discloses a digital system for asset management based on AI and a data processing method thereof, and relates to the technical field of artificial intelligence, and the system comprises a data collection module which is used for obtaining multi-modal asset data of a target asset based on AI; the data preprocessing module is used for cleaning and processing the acquired multi-modal asset data to obtain preprocessed multi-modal asset data; the data storage module is used for storing structured data by using a preset first database and storing unstructured data by using a preset second database; the data storage architecture module constructs a mixed index with a plurality of retrieval indexes, and adopts a three-stage storage architecture: a memory stores hot data, an SSD stores warm data, and a disk stores cold data; the asset health assessment module is used for performing fault probability prediction, maintenance priority ranking and life cycle curve analysis by adopting a predictive maintenance algorithm to generate an asset health assessment report; the method and the device have the effect of improving the management efficiency of building asset management.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a digital system for asset management based on AI and its data processing method. Background Art

[0002] Currently, as a core component of modern facility management, Building Asset Management needs to synchronously solve multi-dimensional problems such as physical space management, equipment operation and maintenance, energy consumption, and regulatory compliance.

[0003] Traditional fixed asset management systems are usually software applications installed locally and need to run on the owner's server, with its IT department performing daily management and maintenance work. With the increasing development of Internet technology, various asset management systems have emerged in the market, but they usually require users to manually input data and lack a centralized information system for tracking and managing all assets. In addition, traditional relational databases cannot efficiently integrate building asset data including multi-modal data such as BIM models (GBXML format), equipment logs (CSV / JSON), surveillance videos (MP4), lease contracts (PDF), etc., and the heterogeneous data processing ability of the management system is poor.

[0004] In summary, the existing technology fails to effectively solve the problem of data islands in building asset management, especially lacks the ability to dynamically evaluate the value of collateral (such as houses, vehicles) in the pledge scenario, and has the defect of poor management efficiency in building asset management, and urgently needs to be improved. Summary of the Invention

[0005] In order to improve the management efficiency of building asset management, this application provides a digital system for asset management based on AI and its data processing method.

[0006] In the first aspect, the invention object of this application is realized by adopting the following technical solutions: A digital system for asset management based on AI, comprising: A data acquisition module for obtaining multi-modal asset data of target assets based on AI; A data preprocessing module for cleaning and processing the acquired multi-modal asset data, including rule engine filtering, master data matching, outlier correction, and data balancing processing, to obtain preprocessed multi-modal asset data; A data storage module for storing structured data using a preset first database and unstructured data using a preset second database; A data storage architecture module that constructs a hybrid index with several types of retrieval indexes and adopts a three-level storage architecture: in-memory storage for hot data, SSD storage for warm data, and disk storage for cold data; The asset health assessment module uses predictive maintenance algorithms to predict failure probabilities, rank maintenance priorities, and analyze life cycle curves, generating an asset health assessment report.

[0007] By adopting the above technical solutions, several types of retrieval indexes include global indexes, local indexes, full-text indexes, and high-dimensional indexes; through AI, multi-modal data such as the location, value, usage status, and maintenance records of building assets are obtained by means of real-time crawling, streaming collection, etc., which is applicable to the asset management scenarios where the collateral (such as houses, vehicles) in the pledge scenario is prone to value and usage status changes. This application adopts AI technology, and through the AI-driven data collection module, it realizes the unified access of structured data such as the location, usage status, and maintenance records of building assets and unstructured data such as device logs, monitoring videos, and BIM models, and the data collection integrity rate is relatively high; the preprocessing module based on the rule engine and machine learning has a relatively high outlier correction accuracy and data cleaning efficiency, especially realizing dynamic adaptation for device parameters (such as temperature, humidity) with high frequency changes in building assets; the preset first database includes the Hyperbase database, and the preset first database includes HDFS; among them, the data storage module and the data storage architecture module adopt a three-level storage architecture (memory / SSD / disk) in cooperation with the Erasure Code technology, which can not only ensure the reliability of asset management data, but also reduce the data storage cost and shorten the query response time of the full life cycle data of building assets (such as design drawings, maintenance records); then, through the asset health assessment module, building equipment failure prediction, maintenance priority ranking, and life cycle curve analysis are carried out to realize the full life cycle digital management of building assets, greatly improving the management efficiency of building asset management.

[0008] In a preferred example of this application: the first database is the Hyperbase database; the unstructured data is stored based on a preset hierarchical storage strategy, and the preset hierarchical storage strategy includes that for files exceeding the preset file storage threshold, the attribute information of the file is stored in the first database, the body data of the file is stored in the second database, and the number of copies of the body data is reduced to a preset multiple of the body data.

[0009] By adopting the above technical solutions, it is beneficial to optimize the storage efficiency and reduce the overall storage cost of asset management. In this application, by separating the storage of the attributes and content of large files, the load of the main database is reduced and the query performance is improved; at the same time, the Erasure Code technology is used to reduce the number of copies, effectively reducing the storage overhead while maintaining relatively high data reliability; and the hierarchical storage strategy allows the most suitable storage method to be adopted for files of different sizes and types.

[0010] In a preferred example of this application: the cleaning and processing of the collected multi-modal asset data includes: Build a data rule library for data rule editing. The data rule library includes regular expressions, synonym replacement, and null value filling strategies, and supports graphical rule editing; Use a real-time stream computing engine component to filter outliers in real-time stream data of the collected multimodal data; use a master data matching engine to perform cross-system data verification on batch data to complete the dynamic cleaning operation of multimodal asset data; A preset data balancer detects the node load information of each load node in the data distribution of the digital system, and performs node load balancing scheduling operations according to the node load information in combination with the consistent Hash algorithm; Based on a preset graphical monitoring console, display the cleaning progress and error rate of the collected multimodal data.

[0011] By adopting the above technical solutions, the data rule library supports multiple cleaning rules to ensure data consistency and accuracy; the graphical interface simplifies the rule editing process; the real-time stream computing engine (such as Transwarp Stream) can instantly filter outliers and improve data quality; the master data matching engine realizes cross-system data verification and achieves cross-system consistency of asset management data among multiple different management systems; then, the data balancer automatically adjusts the node load in combination with the consistent Hash algorithm to ensure the high availability and stability of the system to achieve dynamic load balancing operations.

[0012] In a preferred example of this application: The predictive maintenance algorithm is used to predict the failure probability, sort the maintenance priorities, and analyze the life cycle curve to generate an asset health assessment report, including: Obtain the device operation parameters, maintenance records, and environmental monitoring data within the asset management scope; and extract the structured fields of the asset device and the unstructured features including the asset health status. The structured fields include the asset model, service life, and maintenance history; the unstructured features include image texture features and text sentiment tendencies to determine the current asset device health data; Based on the current asset device health data, combine the hybrid architecture of a convolutional neural network and a long short-term memory network to predict the failure probability of the asset device to obtain failure probability prediction information; combine the device criticality coefficient and the remaining life prediction to sort the maintenance priorities to obtain maintenance priority information; Based on the current asset device health data, failure probability prediction information, and maintenance priority information, generate a device health attenuation curve; Generate an asset health assessment report based on a preset health monitoring cycle instruction.

[0013] By adopting the above technical solution, a hybrid architecture combining a Convolutional Neural Network (CNN) and a Long Short-Term Memory Network (LSTM) can accurately predict the failure probability of equipment, take preventive measures in advance, and at the same time calculate the maintenance priority based on the criticality and remaining life of the asset equipment, optimize resource allocation, extend the service life of the equipment, and improve the planning accuracy of intelligent maintenance planning; by comprehensively analyzing various operating parameters of the equipment and environmental monitoring data, a detailed health degradation curve is provided, providing more scientific data support for asset management decision-making for managers, and then by automatically generating a health assessment report according to a preset cycle, the work efficiency is improved and the risk of human error is reduced.

[0014] In a preferred example of the present application: the several types of retrieval indexes include a global index based on the B+ tree structure, a local index based on the LSM tree structure, and a full-text index based on the inverted index and the BM25 algorithm.

[0015] By adopting the above technical solution, the global index utilizes the B+ tree structure to provide efficient point query and range query capabilities; the local index optimizes the write performance through the LSM tree and is suitable for data sets with frequent updates; the full-text index combines the inverted index and the BM25 scoring algorithm to achieve fast and accurate search of text content and is applicable to the retrieval requirements of unstructured data such as documents and logs; multiple index types can select the most suitable index strategy according to specific application scenarios.

[0016] In a second aspect, the invention object of the present application is achieved by adopting the following technical solution: A data processing method for asset management based on AI, applied to the data processing method for asset management based on AI as described above, the method includes: Preprocess the asset data set obtained by the data acquisition module through the data preprocessing module to obtain a first asset data set and a second asset data set; Extract asset features related to asset management according to the first asset data set and the second asset data set, wherein the asset features include first asset features and second asset features; Calculate asset management indicators representing asset value and management efficiency according to the asset features, wherein the asset management indicators include first asset management indicators and second asset management indicators; Classify and manage the first asset data set and the second asset data set according to the asset management indicators; Obtain the classification management result, and generate an asset management strategy according to the classification management result.

[0017] By adopting the above technical solutions, automated asset classification management is carried out, that is, it is divided into high-efficiency assets / low-efficiency assets, high-risk assets / low-risk assets, which is applicable to the dynamic allocation requirements in the whole life cycle management of building assets; the asset management strategies generated based on the classification results (such as preferentially maintaining high-risk equipment and optimizing the utilization rate of high-efficiency assets) improve the comprehensive utilization rate of building assets; and it can support the dynamic assessment of asset value in financial scenarios such as REITs; integrating structured data (equipment parameters) and unstructured data (images, contract texts), and realizing cross-modal correlation analysis through hybrid indexing technology can not only improve the multi-source data fusion ability, but also improve the accuracy of fault root cause location.

[0018] In a preferred example of the present application: the data preprocessing module preprocesses the asset data set obtained by the data acquisition module to obtain a first asset data set and a second asset data set, including: The data preprocessing module collects asset information from different data sources of the data acquisition module to form an initial asset data set; Perform data cleaning and formatting on the initial asset data set to obtain a preprocessed asset data set; According to the asset type and business requirements, divide the preprocessed asset data set into a first asset data set and a second asset data set.

[0019] By adopting the above technical solutions, the quality of data cleaning is controllable, which is conducive to meeting the requirements of high-frequency operation and maintenance scenarios of building assets; at the same time, based on the node load balancing scheduling of the consistent Hash algorithm, the data distribution deviation between nodes is <5%, and the overall throughput of the cluster is improved, which is applicable to the scenario of concurrent data acquisition of multiple devices on a construction site.

[0020] In a preferred example of the present application: calculating asset management indicators representing asset value and management efficiency according to the asset characteristics, including: Define preset management correlation factors, and screen the first asset characteristics according to the preset management correlation factors to obtain first key asset characteristics, where the first key asset characteristics include asset quality characteristics, asset utilization rate characteristics, and asset depreciation characteristics; Calculate the first asset management indicator according to the first key asset characteristics; According to the preset management correlation factors, screen the second asset characteristics to obtain second key asset characteristics, where the second key asset characteristics include asset income characteristics, asset risk characteristics, and asset liquidity characteristics; Calculate the second asset management indicator according to the second key asset characteristics.

[0021] By adopting the above technical solutions, through the evaluation of the dual-dimensional asset management indicators combining asset value and management efficiency, the accuracy of asset valuation is improved, providing a reliable basis for mortgage loans and asset securitization; through the joint modeling of risk characteristics (such as the probability of equipment failure) and liquidity characteristics (such as asset turnover rate), the visualization of asset risk management is carried out, which is beneficial to improving the risk warning response and the accuracy of high-risk asset identification; then, based on the dynamic analysis of the equipment technical advancement characteristics (such as energy efficiency grade) and market value characteristics, the matching degree of asset disposal strategies is improved.

[0022] In a preferred example of the present application: the classification management of the first asset dataset and the second asset dataset according to the asset management indicators includes: According to the first asset management indicator, identify the high-efficiency assets and low-efficiency assets in the first asset dataset, and classify and manage the high-efficiency assets and low-efficiency assets; According to the second asset management indicator, identify the high-risk assets and low-risk assets in the second asset dataset, and classify and manage the high-risk assets and low-risk assets; and / or The extraction of asset characteristics related to asset management according to the first asset dataset and the second asset dataset includes: According to the first asset dataset, extract the first asset characteristics, where the first asset characteristics include asset quality characteristics, asset historical income characteristics, and asset maintenance cost characteristics; According to the second asset dataset, extract the second asset characteristics, where the second asset characteristics include asset market value characteristics, asset liquidity characteristics, asset risk level characteristics, and asset technical advancement characteristics.

[0023] By adopting the above technical solutions, the resource allocation is optimized to dynamically classify and manage assets; through the classification of high-efficiency assets / low-efficiency assets, the efficiency of operation and maintenance resource allocation is improved; the high-risk asset priority processing mechanism reduces the number of unplanned outages; combined with the building asset usage scenarios (such as commercial complexes, industrial parks), the first asset characteristics and the second asset characteristics of the assets are dynamically analyzed; to improve the adaptability and refinement degree of the management strategies for asset management.

[0024] In a preferred example of the present application: the first asset management indicator includes a first asset efficiency indicator and a first asset value indicator, and the calculation of the first asset management indicator according to the first key asset characteristics specifically includes: Calculate the first asset efficiency indicator according to the asset quality characteristics and the asset utilization rate characteristics; calculate the first asset value indicator according to the asset historical income characteristics and the asset depreciation characteristics; The second asset management metric includes a second asset return metric and a second asset risk metric. Calculating the second asset management metric based on the second key asset characteristics specifically includes: Calculating the second asset return metric based on the asset return characteristics and the asset liquidity characteristics; calculating the second asset risk metric based on the asset risk characteristics and the asset technological advancement characteristics.

[0025] By adopting the above technical solution, a comprehensive assessment of asset value and risk is provided based on a combination of metrics from different dimensions, supporting more refined management and decision-making; by refining metric decisions, the asset management metrics become more operable and interpretable, facilitating the generation of quantifiable and traceable asset management strategies and supporting the maximization of the full life cycle value of building assets.

[0026] In summary, the present application includes at least one of the following beneficial technical effects: 1. By adopting AI technology and through an AI-driven data acquisition module, unified access to structured data such as the location, usage status, and maintenance records of building assets and unstructured data such as device logs, monitoring videos, and BIM models is achieved; through building equipment fault prediction, maintenance priority ranking, and life cycle curve analysis, full life cycle digital management of building assets is realized, greatly improving the management efficiency of building asset management; 2. Integrating structured data (equipment parameters) and unstructured data (images, contract texts), cross-modal correlation analysis is achieved through hybrid indexing technology, which can not only improve the multi-source data fusion ability but also enhance the accuracy of fault root cause location; 3. A comprehensive assessment of asset value and risk is provided based on a combination of metrics from different dimensions, supporting more refined management and decision-making; by refining metric decisions, the asset management metrics become more operable and interpretable. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a system framework diagram of a digital system for asset management based on AI in an embodiment of the present application; Figure 2 is a flowchart of a data processing method for asset management based on AI in an embodiment of the present application; Figure 3 is a flowchart of step S3 in the data processing method for asset management based on AI in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] The present application will be further described in detail below with reference to the accompanying drawings.

[0029] In one embodiment, as Figure 1As shown, the present application discloses a digital system for asset management based on AI. The digital system for asset management based on AI includes a data acquisition module, a data preprocessing module, a data storage module, a data storage architecture module, and an asset health assessment module.

[0030] The data acquisition module is used to obtain multimodal asset data of target assets based on AI. In this embodiment, based on AI technology, the asset location information is collected through an RFID reader, the device temperature status is detected using an infrared sensor, and the device operation video is collected by a high-definition camera. At the same time, it supports multi-source data access such as manual filling, API interface import, and web crawler. Structured data (such as device models, maintenance records) and unstructured data (such as images, log files) are uniformly accessed into the data lake. The preliminary cleaning and standardization of cross-system data are realized through the Transwarp DataHub platform.

[0031] The data preprocessing module cleans and processes the collected multimodal asset data, including rule engine filtering, master data matching, outlier correction, and data balancing processing, to obtain preprocessed multimodal asset data. Among them, rule engine filtering includes filtering invalid characters (such as HTML tags) using regular expressions, synonym replacement (such as "system downtime" and "shutdown"), and null value filling (such as filling with the mean or historical data). It supports dynamic editing of the graphical rule library (such as pre-defining more than 200 preset rules). Real-time stream data processing includes real-time outlier filtering of sensor data through the Transwarp Stream component and detecting data mutations (such as sudden temperature rise) using the sliding window algorithm. The outlier filtering of real-time stream data of the collected multimodal data can be performed through the real-time stream computing engine component. Batch data verification includes verifying the cross-system data consistency through the master data matching engine (MDM) (such as comparing ERP and IoT data). The preset data balancer detects the node load information of each load node in the data distribution of the digital system, and performs node load balancing scheduling operations according to the node load information combined with the consistent Hash algorithm (such as the node load threshold ≤ 10%). The second-level rebalancing is realized through the consistent Hash algorithm in seconds. Based on the preset graphical monitoring console (i.e., the intelligent monitoring terminal), the cleaning progress and error rate of the collected multimodal data are displayed.

[0032] The data storage module uses a preset first database to store structured data and a preset second database to store unstructured data. The first database is a Hyperbase database. The unstructured data is stored based on a preset hierarchical storage strategy. The preset hierarchical storage strategy includes that for files exceeding the preset file storage threshold (such as 10MB), the attribute information of the file is stored in the first database, the ontology data of the file is stored in the second database, and the copy number of the ontology data is reduced to a preset multiple of the ontology data.

[0033] The data storage architecture module constructs a hybrid index with several types of retrieval indexes and adopts a three - level storage architecture: in - memory storage for hot data, SSD storage for warm data, and disk storage for cold data; the data life - cycle management strategy is: set the data cooling time (such as 120 days) and trigger the automatic reduction of the number of replicas; archive cold data to the tape library to reduce storage costs. The several types of retrieval indexes include a global index based on the B + tree structure, a local index based on the LSM tree structure, and a full - text index based on the inverted index and the BM25 algorithm.

[0034] The asset health assessment module uses a predictive maintenance algorithm for fault probability prediction, maintenance priority ranking, and life - cycle curve analysis to generate an asset health assessment report; specifically, obtain the device operation parameters, maintenance records, and environmental monitoring data within the asset management scope; and extract the structured fields of the asset devices and the unstructured features including the asset health status. The structured fields include the asset model, service life, and maintenance history; the unstructured features include image texture features and text sentiment tendencies. For example, use a pre - trained CNN model (such as ResNet50) to extract the texture features (such as edge density, color histogram) of the device surface image, and detect local anomalies (such as oil stains, cracks) through a sliding window; use maintenance logs for word segmentation (Chinese word - segmentation tools: HanLP / Jieba); calculate the sentiment polarity (positive / negative / neutral), and combine TF - IDF to extract the keyword sentiment tendency to determine the current asset device health data; based on the current asset device health data, combine a hybrid architecture of a convolutional neural network (CNN) and a long - short - term memory network (LSTM) to predict the fault probability of the asset device and obtain the fault probability prediction information; combine the device criticality coefficient and the remaining life prediction to conduct maintenance priority ranking and obtain the maintenance priority information. The maintenance priority ranking is scored based on the fault probability, device criticality coefficient, and the remaining days of the remaining life associated with corresponding weight coefficients and then sorted according to the score. Among them, the remaining life prediction can be calculated based on the Weibull distribution model; based on the current asset device health data, fault probability prediction information, and maintenance priority information, generate a device health decay curve, and use a non - linear regression model (such as Gaussian process regression) to fit the decay curve of the health degree changing with time; predict the health degree threshold for the next 30 days (such as a warning is triggered when the health degree < 80%); generate an asset health assessment report based on the preset health monitoring cycle instruction; the structured content of the asset health assessment report includes device basic information, fault prediction results, and maintenance suggestions; the visualization charts include the health degree trend chart, the fault probability distribution heat map, and the priority ranking matrix.

[0035] The implementation principle of the digital system for asset management based on AI in the embodiments of this application is: At the data acquisition layer, structured and unstructured data throughout the asset's entire life cycle are perceived and integrated through multi-modal data, and dynamic data cleaning rules of rule engines, real-time stream processing, and batch verification are used. Combining the extraction and optimization of structured and unstructured features for data preprocessing, and then adopting a data storage method of a hybrid storage strategy plus a hybrid index system. At the analysis and decision-making layer, predictive maintenance and intelligent decision-making engines for the entire life cycle of asset management are carried out to achieve AI-driven health assessment management of the asset management process. Through the management logic of "data-driven + AI model + intelligent decision-making", a digital management closed-loop covering the entire life cycle of the asset is constructed; the purpose of improving the management efficiency of building asset management is achieved.

[0036] In one embodiment, as Figure 2 shown, a data processing method for asset management based on AI is provided. This data processing method for asset management based on AI is applied to a digital system for asset management based on AI. The data processing method for asset management based on AI specifically includes the following steps: S1: The data preprocessing module preprocesses the asset data set obtained by the data acquisition module to obtain a first asset data set and a second asset data set.

[0037] In this embodiment, the first asset data set refers to a high-quality data set for structured analysis after cleaning (such as equipment parameters, maintenance records, contract texts), which is stored in a relational database (such as Hyperbase); the second asset data set refers to the original data set for unstructured analysis (such as images, log files, surveillance videos, BIM models), which is stored in a distributed file system (such as HDFS).

[0038] Specifically, multi-source data is obtained from the data acquisition module: for example, structured data sources include ERP systems (asset ledgers), IoT sensors (temperature, vibration); for example, unstructured data sources include equipment maintenance logs (PDF), 4K camera videos (H.265 encoding), infrared thermal imaging pictures (JPEG).

[0039] Furthermore, step S1 includes: S11: The data preprocessing module collects asset information from different data sources of the data acquisition module to form an initial asset data set.

[0040] Specifically, data source access includes: ① In the movable property pledge scenario: Access building equipment sensor data (such as tower crane operating status, concrete pump truck fuel consumption); upload equipment maintenance work orders (CSV format), equipment purchase contracts (PDF).

[0041] ② Right pledge scenario: Connect to the project revenue right transaction data (such as rent income, parking fee income); upload the proof of accounts receivable creditor's rights (electronic contract, blockchain deposit document).

[0042] S12: Clean and format the initial asset data set to obtain a preprocessed asset data set.

[0043] In this embodiment, the asset data set after data cleaning includes structured data and unstructured data; the structured data such as house lease contract (start and end dates, rent), vehicle mileage, insurance expiration date; the unstructured data such as house inspection photos (JPEG), vehicle repair videos (MP4), lease contract scans (PDF).

[0044] S13: Divide the preprocessed asset data set into a first asset data set and a second asset data set according to the asset type and business requirements.

[0045] Specifically, the asset types include immovable property and movable property, and the business requirements include risk monitoring, compliance auditing, and value assessment; the asset types are divided according to the asset attributes and functions, and the asset types include movable property and immovable property. For example, for house pledge assets, structured data such as property right certificates, lease contracts, and regular inspection reports need to be stored, as well as unstructured data such as actual photos of houses and VR panoramic images; for vehicle pledge assets, structured data such as driving license information, GPS positioning data, and maintenance records need to be stored, as well as unstructured data such as vehicle accident site videos and vehicle condition detection images. The business requirement priorities include real-time monitoring, compliance auditing, and risk warning.

[0046] S2: Extract asset features related to asset management according to the first asset data set and the second asset data set, where the asset features include first asset features and second asset features.

[0047] In this embodiment, structured feature extraction includes direct extraction and calculation of derivative features. Direct extraction such as reading device model, purchase date, and accumulated depreciation value from Hyperbase; calculation of derivative features such as device utilization rate = running duration / total available duration × 100%; contract default risk score = number of defaults × 0.6 + overdue days × 0.4.

[0048] S3: Calculate asset management indicators representing asset value and management efficiency according to the asset features, where the asset management indicators include first asset management indicators and second asset management indicators.

[0049] Specifically, as Figure 3 shown, step S3 includes: S31: Define preset management correlation factors. According to the preset management correlation factors, screen the first asset features to obtain the first key asset features, where the first key asset features include asset quality features, asset utilization features, and asset depreciation features.

[0050] In this embodiment, the preset management correlation factors refer to the key elements that affect asset management preset according to business objectives. For example, in house pledge management, the stability of the collateral value, tenant credit risk, and regional economic fluctuations; in vehicle pledge management, the vehicle residual value depreciation rate, accident history records, and GPS positioning anomaly frequency. The first key asset features refer to the fields directly related to asset quality, utilization, and depreciation extracted from structured data. For example, asset quality features include the safety grade of the house structure, the engine condition score of the vehicle; asset utilization features include the house rental rate, the average daily driving mileage of the vehicle; asset depreciation features include the service life of the house, the total driving mileage of the vehicle.

[0051] Specifically, define the core management objectives according to the pledge scenario: when it comes to risk control, screen the features strongly related to the asset depreciation risk (such as the house crack detection result); when optimizing the income, screen the features that affect the asset income (such as the average daily income of the vehicle). In the house pledge scenario, if it is necessary to evaluate the long-term mortgage risk, it is necessary to focus on "house maintenance records" and "surrounding supporting facilities"; in the vehicle pledge scenario, if it is necessary to quickly dispose of the assets, it is necessary to give priority to extracting "market used car price fluctuation data" S32: Calculate the first asset management indicator according to the first key asset features.

[0052] In this embodiment, the first asset management indicator refers to the indicator that quantifies the asset value and management efficiency based on structured data. Among them, asset efficiency indicators such as house rental rate, vehicle utilization rate; asset value indicators such as house present value assessment, vehicle residual value prediction, contract income prediction; the calculation of asset efficiency can adopt the formula of "actual usage / theoretical maximum value × 100%". For example, the house rental rate = actual number of rented households / total number of rooms × 100%; the vehicle utilization rate = average daily driving mileage / designed maximum mileage × 100%. The asset value indicator is calculated according to the depreciation formula: "asset present value = original value × (1 - depreciation rate) ^ service life". For example, the house depreciation rate is calculated by the straight-line method (assuming a residual value rate of 5% and a service life of 50 years); the vehicle is calculated by the declining balance method (depreciation rate of 30% per year in the first 3 years).

[0053] The contract income prediction formula can adopt the following formula: For example, the monthly rent of a commercial lease contract is 1 million yuan, with an annual increase of 3%, a discount rate of 5%, and the present value of the expected income for 3 years is 2.768 million yuan.

[0054] Further, the first asset management metric includes a first asset efficiency metric and a first asset value metric. According to the first key asset characteristics, the first asset management metric is calculated, specifically including: S321: Calculate the first asset efficiency metric based on the asset quality characteristics and asset utilization rate characteristics; calculate the first asset value metric based on the asset historical return characteristics and asset depreciation characteristics.

[0055] In this embodiment, the asset quality characteristics are in the form of an asset quality score, where the asset quality score = remaining asset life / total life × 100%; the asset utilization rate calculation formula is asset utilization rate = actual usage duration / theoretical maximum duration. Correlation weights are assigned to the asset quality characteristics and asset utilization rate respectively, and then the first asset efficiency metric is synthesized. The first asset efficiency metric = α × asset quality score + β × asset utilization rate. For example, the quality score weight is 0.6 and the utilization rate weight is 0.4.

[0056] S33: Screen the second asset characteristics according to the preset management correlation factors to obtain the second key asset characteristics, where the second key asset characteristics include asset return characteristics, asset risk characteristics, and asset liquidity characteristics.

[0057] In this embodiment, the second key asset characteristics refer to the characteristics related to return, risk, and liquidity extracted from unstructured data; for example, the asset return characteristics include the probability of lease contract renewal and the activity of used car transactions; the asset risk characteristics include the detection results of roof leaks in houses and the sentiment tendency of vehicle accident record texts; the asset liquidity characteristics include the transaction cycle of surrounding similar assets and the frequency of vehicle transfer.

[0058] Specifically, when parsing unstructured data, a pre-trained model (such as YOLOv8) is used to detect house cracks and vehicle scratches; for example, the area of water seepage on the roof of a house is identified through image semantic segmentation; keywords are extracted from the lease contract (such as "preferential renewal clause"); sentiment analysis is performed on the maintenance log (such as marking "severe failure" as high risk); then asset characteristics are screened based on a preset feature rule library. For example, "if 'cracks' are found in the image detection results, the risk score will increase by 0.3 points; in addition, if the sentiment analysis of the log text shows negative sentiment, the risk score will increase by another 0.2 points", and then a random forest classifier is used to predict the probability of lease contract renewal (input features: tenant credit score, historical renewal times).

[0059] S34: Calculate the second asset management metric according to the second key asset characteristics.

[0060] In this embodiment, the second asset management metric refers to an evaluation metric that quantifies the potential return and risk of assets based on unstructured data, such as return metrics and risk metrics. Return metrics include potential rental income of a house and valuation premium rate of a used car; risk metrics include probability of housing default risk, historical severity of vehicle accidents, and liquidity risk.

[0061] Specifically, the return metric is calculated as follows: Use linear regression to predict the growth rate of house rent (input features: surrounding housing price index, degree of improvement of supporting facilities); for example, rent growth rate = 0.3 × housing price index + 0.2 × subway opening progress; the risk metric is calculated by setting a scoring system, such as binning unstructured features (e.g., crack area is divided into "no crack", "≤5cm 2 ", ">5cm 2 "); for example, housing risk score = bin value of crack area × 0.4 + bin value of water seepage level × 0.6.

[0062] Furthermore, the second asset management metric includes a second asset return metric and a second asset risk metric. According to the second key asset feature, the second asset management metric is calculated, specifically including: S341: Calculate the second asset return metric according to the asset return feature and the asset liquidity feature.

[0063] In this embodiment, the asset return feature refers to the predicted data of the future return ability of the asset (such as rent growth rate, dividend ratio); the asset liquidity feature refers to the indicator of the asset's liquidity (such as transaction cycle, market demand heat); the second asset return metric refers to the comprehensive value that quantifies the potential return and liquidity of the asset (such as expected annualized return rate).

[0064] The formula for predicting the return potential of the asset return feature is: expected return = current return × (1 + growth rate) ^ prediction period, and the asset liquidity score is: liquidity score = 1 / average transaction cycle (days) × 100%. The second asset return metric = ∈ × expected return + μ × liquidity score, where ∈ and μ are weight coefficients.

[0065] S342: Calculate the second asset risk metric according to the asset risk feature and the asset technological advancement feature.

[0066] Specifically, the risk scoring formula for the asset risk probability is: where i is the feature identifier related to the asset risk; ω i is the weight, and the weight represents the importance of the i-th feature in the overall risk score. The larger the weight value, the more significant the influence of the feature on the final score; x i is the feature value of the risk score, representing the specific value of the i-th feature, and n represents the total number of features participating in the scoring.

[0067] For the part of evaluating technological advancement, the expert scoring method is adopted. For example, the level of intelligence (0 - 10 points): 8 points; the energy efficiency grade (0 - 10 points): 9 points; the comprehensive score: (8 + 9) / 2 = 8.5 points.

[0068] The calculation formula for the second asset risk indicator is as follows: The second asset risk indicator = γ1 × risk score + γ2 × technology score, where γ1 and γ2 are predefined weight coefficients.

[0069] S4: Classify and manage the first asset dataset and the second asset dataset according to the asset management indicators.

[0070] In this embodiment, step S4 includes: S41: Identify the high - efficiency assets and low - efficiency assets in the first asset dataset according to the first asset management indicator, and classify and manage the high - efficiency assets and low - efficiency assets.

[0071] Specifically, high - efficiency assets refer to those with asset utilization rate ≥ 80%, residual value ≥ 50%, and risk score ≤ 20%; low - efficiency assets refer to those with asset utilization rate < 50%, residual value < 30%, and risk score > 40%.

[0072] S42: Identify the high - risk assets and low - risk assets in the second asset dataset according to the second asset management indicator, and classify and manage the high - risk assets and low - risk assets.

[0073] Specifically, high - risk assets and low - risk assets are divided according to the asset risk score value; classification management measures such as equipment pledge in chattel mortgage: high - efficiency assets (such as a normally operating excavator) are included in the core asset pool; low - efficiency assets (such as a mixer that has been idle for 6 months) trigger the asset disposal process. Contract pledge in chattel mortgage: for high - efficiency assets (such as a commercial property with stable rent collection), the pledge rate is increased to 70%; for high - risk assets (such as a lease contract with multiple defaults), the pledge rate is restricted to 40%.

[0074] For accounts receivable pledge in right pledge classification: for high - efficiency assets (such as medical insurance receivables from a top - tier hospital), the repayment period is shortened to quarterly interest payment; for low - efficiency assets (such as project payment in arrears for 3 months), legal collection procedures are initiated. Intellectual property pledge: for high - efficiency assets (such as an invention patent authorization), the value - added potential is evaluated; for high - risk assets (such as a trademark right in an infringement dispute), the pledge operation is frozen.

[0075] S5: Obtain the classification management results, and generate an asset management strategy according to the classification management results.

[0076] In this embodiment, the asset management strategy includes a management strategy specified based on the dynamic pledge rate and risk mitigation measures; the dynamic pledge rate refers to the pledge ratio adjusted in real time according to the asset health (e.g., when the health > 90%, the pledge rate is 80%, and when the health < 70%, the pledge rate ≤ 40%); the risk mitigation measures include increasing margin, purchasing insurance, introducing third-party guarantees, etc.; furthermore, it is manifested as: the asset management strategy includes a movable property pledge strategy and a right pledge strategy.

[0077] Among them, the movable property pledge strategy includes equipment pledge and contract pledge: For equipment pledge, such as generating a "List of Equipment Maintenance Priorities" (high-risk equipment is repaired first); pushing a "Suggestion for Pledge Rate Adjustment" to the risk control department (e.g., reducing the pledge rate when the equipment usage rate decreases). For contract pledge, such as automatically renewing low-risk contracts (such as shop leases with good performance); terminating high-risk contracts (such as rent arrears for 3 consecutive months).

[0078] Among them, the right pledge strategy includes accounts receivable pledge and intellectual property pledge; for accounts receivable pledge, such as issuing ABS products for high-quality accounts receivable (such as government project payments); making bad debt provisions for low-quality accounts receivable (such as small and medium-sized enterprise debts). For intellectual property pledge, such as accelerating the examination of high-value patent applications; initiating an invalidation procedure for trademarks at risk of infringement.

[0079] In one embodiment, in step S2, according to the first asset dataset and the second asset dataset, asset features related to asset management are extracted, including: S21: According to the first asset dataset, the first asset features are extracted, where the first asset features include asset quality features, asset historical income features, and asset maintenance cost features.

[0080] In this embodiment, the asset quality feature is an indicator reflecting the physical state and health of the asset (such as equipment service life, proportion of crack area); the asset historical income feature refers to the income data generated by the asset in the past (such as rental income, equipment usage rate); the asset maintenance cost feature refers to the cost and time data of maintenance activities (such as repair cost, downtime duration).

[0081] Specifically, the extraction logic of the asset quality feature is: directly extract the difference between the equipment purchase date and the current date to obtain the service life; then obtain the physical state score, that is, predict the equipment health (0 - 100 points) based on sensor data (such as vibration frequency, temperature) through a pre-trained model (such as LSTM). The historical income feature is calculated based on the rental yield and vacancy rate, where the rental yield = annual rental income / asset valuation × 100%; the vacancy rate = vacant area / total rentable area × 100%. The maintenance cost feature considers the single repair cost and the proportion of the annual maintenance budget, where the single repair cost = repair cost / number of repairs; the proportion of the annual maintenance budget = total annual maintenance cost / original asset value × 100%.

[0082] S22: Extract second asset features according to the second asset dataset, where the second asset features include asset market value features, asset liquidity features, asset risk level features, and asset technological advancement features. In this embodiment, the asset market value feature reflects the fair value of the asset in the market (such as the transaction price of similar surrounding assets, the appraisal price of second-hand houses); the asset liquidity feature refers to the ease of asset realization (such as the transaction cycle, the market demand heat); the asset risk level feature refers to the quantitative assessment of potential asset risks (such as the default probability, the risk of technological obsolescence); the asset technological advancement feature refers to the leading degree of the technology adopted by the asset in the industry (such as the energy-saving efficiency, the intelligent level).

[0083] Specifically, the extraction logic of the second asset features is as follows: The market value feature is considered based on the comparable transaction method and the income method. Among them, the comparable transaction method counts the average transaction price per square meter of similar surrounding office buildings in the past 3 years (such as 80,000 yuan per square meter); the income method calculates the asset valuation based on the rental income and the capitalization rate (income method valuation = annual rent × 12 / capitalization rate). The liquidity feature considers the transaction cycle and the market demand index. Among them, the transaction cycle counts the average transaction cycle of similar assets in the past 1 year (such as 60 days for residential properties, 90 days for commercial properties); the market demand index is evaluated through the search engine keyword heat (such as "office building for sale", "factory building for lease"). The risk level feature is considered based on the default probability and the risk of technological obsolescence. Among them, the default probability is predicted based on historical default data (such as the rent overdue rate > 30%) combined with the credit scoring model; the risk of technological obsolescence is used to compare the equipment technical parameters with the latest industry standards (such as the air-conditioning energy efficiency level is lower than the national first-level energy efficiency). The technological advancement feature considers the energy-saving efficiency and the intelligent level. Among them, the energy-saving efficiency calculates the PUE value (data center energy efficiency index) through the energy consumption monitoring data; the intelligent level counts the coverage rate of Internet of Things devices (such as smart meters, automated fire protection systems).

[0084] Exemplarily, the pledge assessment content of a commercial plaza includes: the market value feature is: the average transaction price of surrounding shops is 150,000 yuan per square meter; the liquidity feature is: the average transaction cycle in the past 1 year is 85 days; the technological advancement feature is: equipped with a smart parking system and a photovoltaic roof (the energy-saving efficiency is increased by 30%). This application covers multi-dimensional features such as the physical state, economic value, and legal ownership of the asset; the dynamic nature combines real-time sensor data and historical trends to predict the future performance of the asset.

[0085] It should be understood that the sequence numbers of the steps in the above embodiments do not mean the order of execution is prior or posterior, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of this application.

[0086] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0087] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A digital system for asset management based on AI, characterized in that, Including: A data acquisition module for obtaining multi-modal asset data of target assets based on AI; A data preprocessing module that cleans and processes the collected multi-modal asset data, including rule engine filtering, master data matching, outlier correction, and data balancing processing, to obtain preprocessed multi-modal asset data; A data storage module that stores structured data using a preset first database and stores unstructured data using a preset second database; A data storage architecture module that constructs a hybrid index with several types of retrieval indexes and adopts a three-level storage architecture: in-memory storage for hot data, SSD storage for warm data, and disk storage for cold data; An asset health assessment module that uses a predictive maintenance algorithm to perform failure probability prediction, maintenance priority ranking, and life cycle curve analysis, and generates an asset health assessment report.

2. The digital system for asset management based on AI according to claim 1, characterized in that, The first database is a Hyperbase database; unstructured data is stored based on a preset hierarchical storage strategy, and the preset hierarchical storage strategy includes that for files exceeding the preset file storage threshold, the attribute information of the files is stored in the first database, the ontology data of the files is stored in the second database, and the number of copies of the ontology data is reduced to a preset multiple of the ontology data.

3. The digital system for asset management based on AI according to claim 1, characterized in that, The cleaning and processing of the collected multi-modal asset data includes: Constructing a data rule library for data rule editing, where the data rule library includes regular expressions, synonym replacement, and null value filling strategies, and supports graphical rule editing; Filtering outliers of real-time stream data from the collected multi-modal data through a real-time stream computing engine component; using a master data matching engine to perform cross-system data verification on batch data to complete the dynamic cleaning operation of multi-modal asset data; A preset data balancer detects the node load information of each load node in the data distribution of the digital system, and performs node load balancing scheduling operations according to the node load information in combination with the consistent Hash algorithm; Based on a preset graphical monitoring console, display the cleaning progress and error rate of the collected multi-modal data.

4. The digital system for asset management based on AI according to claim 1, characterized in that, The use of a predictive maintenance algorithm to perform failure probability prediction, maintenance priority ranking, and life cycle curve analysis, and generate an asset health assessment report, includes: Obtaining the device operation parameters, maintenance records, and environmental monitoring data within the asset management scope; and extracting the structured fields of the asset devices and the unstructured features including the asset health status, where the structured fields include the asset model, service life, and maintenance history; the unstructured features include image texture features and text sentiment tendencies, to determine the current asset device health data; Based on the current asset device health data, combining a hybrid architecture of a convolutional neural network and a long short-term memory network to predict the failure probability of the asset device, obtaining failure probability prediction information; combining the device criticality coefficient and the remaining life prediction to perform maintenance priority ranking, obtaining maintenance priority information; Based on the current asset device health data, failure probability prediction information, and maintenance priority information, generate a device health decay curve; Generate an asset health assessment report based on a preset health monitoring cycle instruction.

5. The digital system for asset management based on AI according to claim 1, characterized in that, The several retrieval indexes include a global index based on the B+ tree structure, a local index based on the LSM tree structure, and a full-text index based on the inverted index and the BM25 algorithm.

6. A data processing method for asset management based on AI, characterized in that, Applied to the data processing method for asset management based on AI as described in any one of claims 1-5, the method includes: Preprocessing the asset data set obtained by the data acquisition module through the data preprocessing module to obtain a first asset data set and a second asset data set; Extracting asset features related to asset management according to the first asset data set and the second asset data set, wherein the asset features include first asset features and second asset features; Calculating asset management indicators representing asset value and management efficiency according to the asset features, wherein the asset management indicators include first asset management indicators and second asset management indicators; Classifying and managing the first asset data set and the second asset data set according to the asset management indicators; Obtaining the classification management result, and generating an asset management strategy according to the classification management result.

7. The data processing method for asset management based on AI according to claim 6, wherein The preprocessing the asset data set obtained by the data acquisition module through the data preprocessing module to obtain a first asset data set and a second asset data set includes: The data preprocessing module collects asset information from different data sources of the data acquisition module to form an initial asset data set; Performing data cleaning and formatting on the initial asset data set to obtain a preprocessed asset data set; Dividing the preprocessed asset data set into a first asset data set and a second asset data set according to the asset type and business requirements.

8. The data processing method for asset management based on AI according to claim 6 or 7, characterized in that The calculating asset management indicators representing asset value and management efficiency according to the asset features includes: Defining preset management correlation factors, and screening the first asset features according to the preset management correlation factors to obtain first key asset features, wherein the first key asset features include asset quality features, asset utilization rate features, and asset depreciation features; Calculating first asset management indicators according to the first key asset features; Screening second asset features according to the preset management correlation factors to obtain second key asset features, wherein the second key asset features include asset income features, asset risk features, and asset liquidity features; Calculating second asset management indicators according to the second key asset features.

9. The data processing method for asset management based on AI according to claim 6 or 7, characterized in that The classifying and managing the first asset data set and the second asset data set according to the asset management indicators includes: Identifying high-efficiency assets and low-efficiency assets in the first asset data set according to the first asset management indicators, and classifying and managing the high-efficiency assets and the low-efficiency assets; Identifying high-risk assets and low-risk assets in the second asset data set according to the second asset management indicators, and classifying and managing the high-risk assets and the low-risk assets; and / or The extracting asset features related to asset management according to the first asset data set and the second asset data set includes: Extract first asset features according to the first asset dataset, where the first asset features include asset quality features, asset historical return features, and asset maintenance cost features; Extract second asset features according to the second asset dataset, where the second asset features include asset market value features, asset liquidity features, asset risk level features, and asset technological advancement features.

10. The data processing method for asset management based on AI according to claim 8, wherein The first asset management metrics include first asset efficiency metrics and first asset value metrics. Calculating the first asset management metrics according to the first key asset features specifically includes: Calculate the first asset efficiency metrics according to the asset quality features and the asset utilization rate features; calculate the first asset value metrics according to the asset historical return features and the asset depreciation features; The second asset management metrics include second asset return metrics and second asset risk metrics. Calculating the second asset management metrics according to the second key asset features specifically includes: Calculate the second asset return metrics according to the asset return features and the asset liquidity features; calculate the second asset risk metrics according to the asset risk features and the asset technological advancement features.

Images