Asset evaluation method and system based on multi-dimensional feature fusion

Abstract

The invention relates to the technical field of asset assessment, in particular to an asset assessment method and system based on multi-dimensional feature fusion. The method comprises the following steps: obtaining basic operation data of a lease item, and carrying out operation feature extraction according to the basic operation data of the lease item to obtain first asset evaluation feature data; obtaining user business data and user financial data, and performing asset assessment feature extraction according to the user business data and the user financial data to obtain second asset assessment feature data; performing multi-dimensional fusion according to the first asset assessment feature data and the second asset assessment feature data to obtain asset assessment feature data; and performing asset assessment according to the asset assessment feature data to obtain asset assessment data. According to the method, data of different sources and dimensions are comprehensively collected and fused, so that more accurate, dynamic and meticulous asset assessment can be provided.

Description

Technical Field

[0001] The present invention relates to the technical field of asset evaluation, and in particular to an asset evaluation method and system based on multi-dimensional feature fusion. Background Art

[0002] With the rapid development of modern society, various types of assets (especially leased assets) have become an important part of daily life and business activities for enterprises and individuals. In particular, the operating conditions and efficiency of leased assets such as new energy vehicles directly affect the economic benefits and asset value of leasing companies. Therefore, accurately assessing the asset value and risks of leased assets (such as new energy vehicles) and related users is crucial for asset management and risk control of leasing companies and financial institutions.

[0003] Traditional asset valuation methods mainly rely on traditional accounting methods based on static data and simple financial indicator analysis, such as book value, depreciation value, etc. These methods can provide basic asset evaluation in some cases, but often ignore the dynamic and multi-dimensional characteristic information in the asset operation process. For example, in the operation of leased assets (such as new energy vehicles), factors such as frequency of use, maintenance, failure history, user's ability to pay, and credit status have a significant impact on the future value and risk of assets. These factors are often not fully captured by traditional valuation methods. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an asset evaluation method and system based on multi-dimensional feature fusion to solve at least one of the above technical problems.

[0005] The present application provides an asset evaluation method based on multi-dimensional feature fusion, the method comprising: S1. Obtaining basic operating data of the leased property, and extracting operating characteristics based on the basic operating data of the leased property to obtain first asset evaluation characteristic data; S2, obtaining user business data and user financial data, and extracting asset evaluation features based on the user business data and user financial data to obtain second asset evaluation feature data; S3, performing multi-dimensional fusion on the first asset evaluation feature data and the second asset evaluation feature data to obtain asset evaluation feature data; S4. Perform asset evaluation based on the asset evaluation characteristic data to obtain asset evaluation data.

[0006] In the present invention, the first asset evaluation characteristic data (based on the operating data of the leased object) can reveal the operating status, usage frequency, maintenance status, etc. of the leased object, and the second asset evaluation characteristic data (based on the user's business and financial data) can reveal the user's repayment ability, operating efficiency, etc. Combining these two types of data can form a comprehensive evaluation view, effectively avoiding the limitations of a single perspective. Through the multi-dimensional fusion of the first and second asset evaluation characteristic data, weighted or selective fusion can be performed according to actual needs. For example, for certain high-value or high-risk assets, the weight of operating data can be increased, while for other cases, more reliance can be placed on financial data, making the method applicable to different types of assets and business models. By fusing data from different sources, the defects of a single data source (such as incomplete operating data, lagging financial data, etc.) can be overcome, thereby improving the scientificity and comprehensiveness of the evaluation. For example, the operating status of the leased object (such as high-frequency use or frequent failures) can have a significant impact on the asset value, and these factors are difficult to reflect in pure financial data.

[0007] Optionally, S1 includes: S11. Obtaining basic operating data of leased property; S12, performing segmented analysis on the basic operation data of the leased property to obtain segmented operation data of the leased property; S13, performing event recognition on the leased property operation segment data to obtain leased property operation event data; S14, extracting operation event features from the leased property operation event data to obtain leased property operation event feature data; S15. Obtaining the climate condition data of the leased property corresponding to the basic operation data of the leased property; S16. Perform multimodal feature extraction based on the climate condition data of the leased property and the characteristic data of the operation events of the leased property to obtain first asset evaluation characteristic data.

[0008] The operational segmented data of the leased property in the present invention can reveal its operating mode, event identification and feature extraction can provide in-depth understanding of its special behaviors in operation (such as failures, maintenance, etc.), and the impact of climate conditions on the leased property (such as bad weather may affect operational efficiency) provides more environmental variables for evaluation. Comprehensive evaluation of this information can accurately evaluate the health status, use value and future potential of the leased property. By identifying and extracting features of leased property operation events, key changes and abnormal behaviors of leased property can be captured in a dynamic environment. The event-based data analysis method enables the evaluation to be based not only on static data (such as historical usage records and financial data), but also on dynamic monitoring and prediction of state changes and potential risks of leased property. Severe climate conditions (such as high temperature, severe cold, heavy rain, etc.) have a direct impact on the service life, maintenance frequency and operational efficiency of leased property. Therefore, combined with climate condition data, the real performance of leased property in different environments can be better reflected, especially when it comes to long-term leases or cross-regional operations, the importance of climate factors is particularly prominent. Climate conditions and operational events are not independent, but have a mutually influential relationship. Severe climate conditions lead to more abnormal events (such as failures, delays, etc.) in the operation of leased property. Through multimodal feature extraction, the interaction between climate and operations can be captured, further improving the accuracy and practicality of asset assessment.

[0009] Optionally, the segmented analysis includes: Perform time series segmentation analysis on the basic operating data of the leased property to obtain the first segmented operating data of the leased property; Performing geographical location segmentation analysis on the basic operating data of the leased property to obtain the second leased property operating segmentation data; Perform state switching segment analysis on the basic operation data of the leased property to obtain the third leased property operation segment data; Perform clustering calculations on the first leased property operation segment data, the second leased property operation segment data, and the third leased property operation segment data to obtain first leased property operation segment clustering data, second leased property operation segment clustering data, and third leased property operation segment clustering data, respectively; A graph is constructed based on the first leased property operation segment clustering data, the second leased property operation segment clustering data, and the third leased property operation segment clustering data to obtain leased property operation segment data.

[0010] In the present invention, time series analysis can capture the operational changes of leased items in different time periods; geographic location analysis considers the impact of different regions on the usage patterns of leased items; and state switching analysis can reveal the performance of leased items under different operating conditions. Multi-dimensional data fusion can comprehensively and accurately evaluate the usage status and potential value of leased items. Time series analysis can track the usage frequency and failure conditions of leased items, geographic location segmentation can reveal the usage efficiency and risks of leased items in different regions, and state switching analysis can identify whether leased items have abnormalities or failures. Combined with cluster analysis, these dynamic information can help managers identify trends, make risk predictions, and make strategic adjustments for specific operating modes. By performing cluster analysis on segmented data, operational data can be divided into different categories (such as normal operation, potential risks, high risks, etc.), allowing managers to identify which leased items perform abnormally under certain specific conditions. In different regions, the usage of leased items varies due to differences in climate, market demand, and user behavior. Geographic location segmentation can reveal these differences, and time series analysis and state switching help understand the operating modes and state changes in different regions.

[0011] Optionally, the time series segmentation analysis includes: Perform time-characteristic enhancement processing on the basic operation data of the leased property to obtain time-enhanced feature data; Performing multi-scale pattern recognition on the time-enhanced feature data to obtain pattern recognition feature data; Performing dynamic segmentation point detection on the pattern recognition feature data to obtain dynamic segmentation point data; Perform spatiotemporal segmentation fusion on the dynamic segmentation point data to obtain spatiotemporal segmentation data; Generating segmentation features for the spatiotemporal segmentation data to obtain segmentation feature data; Hierarchical grouping is performed according to the segment characteristic data to obtain the first leased property operation segment data.

[0012] The generation of time-enhanced feature data in the present invention can enhance the expressiveness of data at different time scales, thereby modeling the operating mode of the leased property in a more detailed and accurate manner. For example, by capturing periodic fluctuations and trend changes, the future performance of the leased property can be better predicted, and its value fluctuations and risks can be assessed. Different leased properties may present different usage patterns at different time scales. For example, in daily operations, there are short-term peaks and troughs, while on a long-term scale, a stable operating trend may be presented. Multi-scale pattern recognition can help identify the operating modes at these different time scales and provide clearer data features for further analysis. The detection of dynamic segmentation points can automatically identify important turning points in the operation process, such as equipment failures, changes in usage patterns, or environmental changes, which are usually important bases for evaluating the value or operating status of leased properties. By accurately identifying these key nodes, the operation of the leased property can be better analyzed in depth, and a scientific basis can be provided for future operational adjustments. Leased properties in different regions may exhibit different operating modes due to different geographical locations and climatic conditions. Through the fusion of spatiotemporal segments, these spatial factors and changes in operating time can be taken into account at the same time, helping to identify which time periods and geographical locations have the best or worst performance for the leased assets, thereby optimizing the operating strategy. By generating independent feature data for each segment, the operating characteristics, risk status, failure frequency, etc. of each segment can be described in detail. Hierarchical grouping can organize segmented data into different hierarchical structures based on different operating stages, event types, geographical locations, and other dimensions of the leased assets. For example, operating data in high-risk areas may be extracted separately as a group for in-depth analysis, while data in low-risk areas can be aggregated into a group.

[0013] Optionally, the geographic location segmentation analysis includes: Perform multi-source geographic data fusion on the basic operation data of leased assets to obtain geographic fusion data of leased assets; Dynamically mark the geographical location of the leased object geographical fusion data to obtain the geographically marked data of the leased object; Performing multi-dimensional geographic region mapping on the geographical tag data of the leased object to obtain multi-dimensional region mapping data; Perform regional clustering optimization based on multi-dimensional regional mapping data to obtain preliminary regional clustering data; Dynamically adjust the boundaries according to the preliminary regional clustering data to obtain optimized regional clustering data; Extract operational behavior features based on optimized regional clustering data to obtain regional behavior feature data; Generate geographic semantic segmentation for regional behavior feature data to obtain geographic semantic segmentation data for leased items; Operation segmentation labels are generated according to the geographic semantic segmentation data of the leased object to obtain second operation segmentation data of the leased object.

[0014] The fusion of multi-source data in the present invention enables the geographic location data to not only consider the real-time location of the leased object, but also to be deeply integrated with environmental data (such as weather and traffic conditions), thereby obtaining more comprehensive geographic information. Dynamic marking can timely update the location status of the leased object at different times and in different environments. For example, when the leased object operates across regions, its current specific location can be determined by dynamic marking technology, and combined with local operating environment data (such as climate, traffic conditions, etc.), the operating status of the leased object can be reflected in real time. Multidimensional geographic area mapping not only considers the distribution of geographic space, but also combines factors such as environment, climate, and traffic to form a multi-dimensional regional model. Regional clustering optimization can identify areas with similar operating behaviors and make optimization adjustments, which helps to identify potential efficient areas or risk areas. During the operation of leased objects, the performance of certain areas changes (such as increased demand or climate change), and the dynamic boundary adjustment technology can automatically identify and adjust the clustering results of the area. The extraction of regional behavior feature data can reflect the operating mode, risk status, abnormal behavior, etc. of the leased object in a specific area. For example, in some areas, due to poor traffic or bad climate, frequent failures of leased objects occur, and the behavior feature extraction can capture these abnormal patterns. Geographic semantic segmentation can combine operational behavior with geographic location semantic information. For example, different segmentation labels can be generated for operational behavior under different climate conditions, such as "high-risk area", "efficient operation area", etc. After the operational segmentation labels are generated, the operational status of each leased item in a specific area can be clearly understood, such as "efficient operation area", "risk warning area", etc.

[0015] Optionally, the state switching segment analysis includes: Mark the status characteristics according to the basic operation data of the leased object to obtain the status marking data of the leased object; Perform state switching detection according to the leased object state mark data to obtain the leased object state switching point data; Performing a switching event context analysis on the leased item state switching point data to obtain leased item switching context data; Generate a state fragment according to the leased object switching context data to obtain leased object state fragment data; Extracting behavior state features from the leased item status fragment data to obtain behavior state feature data of the leased item; Perform segmented semantic optimization on the behavior status feature data of the leased object to obtain optimized segmented semantic data; The leased object operation segment data is generated according to the optimized segmented semantic data and the leased object status fragment data to obtain the third leased object operation segment data.

[0016] The state feature marking technology in the present invention can mark the specific state of the leased object, such as "charging", "driving", "fault" or "to be repaired", based on real-time data (such as battery power, fault state, working state, etc.). During the operation of the leased object, there will be a switch from the "charging" state to the "driving" state, or from the "normal operation" to the "fault" state. State switching detection can capture these changes in time and help analyze the transition duration, frequency and conditions between different states. In the process of state switching, context analysis is performed in combination with external factors (such as climate conditions, traffic conditions, equipment health status, etc.) to reveal why the leased object switches state at a certain moment. For example, when the battery power is low, it switches from the "waiting to charge" state to the "charging" state. This switch is triggered by long-term high-load operation or other factors. The state segment generation is not limited to simply cutting the state switching points, but can also be reasonably divided in combination with the operating history data of the leased object (such as driving distance, charging times, etc.). The behavior state feature extraction technology can extract key indicators in each state segment, such as battery efficiency, charging time, driving mileage, failure rate, etc. Segmented semantic optimization not only gives each state segment a clear semantic label (such as "efficient operation area" and "fault warning area"), but also provides the system with a clear behavioral analysis report. For example, when the battery power is low, the "inefficient operation" state will be triggered and marked as "pending maintenance" or "needing optimization". Operational segmented label generation can correspond each state segment of the leased object to a specific operation mode or risk scenario. For example, when the leased object is running at high load for a long time, a "high load operation" label will be generated, and when there are frequent failures, a "frequent failure" label will be generated.

[0017] Optionally, S2 includes: S21. Obtain user business data and user financial data; S22. Extract business risk characteristics based on user business data to obtain business risk characteristic data; S23, calculating the credit capacity according to the business risk characteristic data to obtain credit capacity data; S24, predicting the user's repayment intention based on the credit ability data and the user's financial data to obtain repayment intention prediction data; S25. Perform financial health assessment based on the business risk characteristic data and the user's financial data to obtain financial health data; S26. Generate asset evaluation characteristics using the financial health data and repayment intention prediction data to obtain second asset evaluation characteristic data.

[0018] The business risk feature extraction technology in the present invention can deeply analyze the user's operating behavior (such as payment habits, historical overdue records, debt status, etc.), evaluate whether the user has risks in a certain period of time, and thus provide more accurate data for evaluating the user's debt repayment ability. Through the repayment intention prediction technology, the user's future repayment behavior can be accurately predicted, and the user's future repayment ability and intention can be evaluated, thereby providing forward-looking data support for asset evaluation. Financial health assessment covers various aspects such as income, expenditure, debt, and assets, and can evaluate key indicators such as the user's cash flow, debt repayment ability, and asset accumulation. It helps to identify users with greater financial pressure, take risk control measures in advance, and reduce bad debt risks. The integration of financial health and repayment intention data can provide a series of information including the user's current debt repayment ability, future repayment intention, potential financial problems, etc. The feature data can be used as the basis for asset evaluation, providing more comprehensive decision support for the evaluator and avoiding the limitations of a single data source. Traditional asset evaluation is usually based on static data and lacks prediction of future changes in user behavior. By introducing dynamic prediction and real-time risk assessment, asset evaluation is made more accurate and flexible. For example, as the user's repayment intention changes, the evaluation data can be automatically updated to help adjust the asset management strategy in a timely manner.

[0019] Optionally, S3 includes: S31, performing multi-dimensional feature combination according to the first asset evaluation feature data and the second asset evaluation feature data to obtain feature matrix data; S32, performing feature conflict resolution based on feature matrix data to obtain preliminary asset assessment feature data; S33. Perform feature alignment and fusion on the preliminary asset assessment feature data to obtain asset assessment feature data.

[0020] The multi-dimensional feature combination in the present invention can integrate data from different sources (such as leased property operation data, user business data, financial data, etc.), and depict a more complete user behavior and asset status from multiple dimensions. In the asset evaluation process, data from different sources have problems such as information duplication, inconsistent definitions, and data mismatches. Through feature conflict resolution, these data conflicts can be effectively handled, so that data from different sources can be smoothly combined, errors or inconsistencies can be eliminated, and the reliability of the evaluation can be ensured. Different data sources often use different formats and structures. The feature alignment and fusion steps can convert them into a unified format and standard, so that different features can be effectively combined in the same evaluation framework. This not only improves the efficiency of data utilization, but also reduces errors caused by improper data processing and enhances the accuracy of the evaluation.

[0021] Optionally, S4 includes: S41, predicting asset value based on asset evaluation characteristic data to obtain asset value prediction data; S42, performing asset risk assessment based on the asset assessment characteristic data to obtain asset risk assessment data; S43. Integrate the asset value prediction data and the asset risk assessment data to obtain asset assessment data.

[0022] The asset value prediction in the present invention focuses on the future profit potential of the asset, while the asset risk assessment reveals the risk factors faced by the asset. Combining the two, it is possible to fully understand the risk and return of the asset, so that the evaluation results are more accurate and ensure that investors or managers make wise decisions. In asset management, decision makers need to consider the future value and potential risks of assets. Asset value prediction provides decision makers with potential value-added space for assets, while asset risk assessment reveals the risks faced during investment or operation. The accuracy of asset evaluation depends not only on the current evaluation data, but also on the prediction of future changes and the assessment of possible risks. The combination of asset value prediction and asset risk assessment enables the evaluation results to dynamically reflect market changes and provide decision makers with flexible adjustment space.

[0023] Optionally, the present application further provides an asset evaluation system based on multidimensional feature fusion, which is used to execute the asset evaluation method based on multidimensional feature fusion as described above, and the asset evaluation system based on multidimensional feature fusion includes: An operation feature extraction module is used to obtain basic operation data of the leased object, and extract operation features based on the basic operation data of the leased object to obtain first asset evaluation feature data; An asset evaluation feature extraction module is used to obtain user business data and user financial data, and extract asset evaluation features based on the user business data and user financial data to obtain second asset evaluation feature data; An asset evaluation feature fusion module, used for performing multi-dimensional fusion according to the first asset evaluation feature data and the second asset evaluation feature data to obtain asset evaluation feature data; The asset evaluation module is used to perform asset evaluation based on asset evaluation characteristic data to obtain asset evaluation data.

[0024] The purpose of the present invention is that the evaluation of assets is not only based on the basic operating data of the leased property (such as mileage, number of charging times, etc.), but also integrates the business behavior data and financial status data of the user, which can fully capture the potential value of the asset and the risks it faces, and better predict the future performance of the asset, thereby effectively avoiding the deviation or error caused by the evaluation of a single indicator. Traditional asset evaluation methods often rely on static data and cannot respond to market changes or fluctuations in operating status in a timely manner. By adopting multi-dimensional feature fusion, this method can dynamically adjust the evaluation model according to data of different dimensions to cope with the external market environment, user behavior changes and fluctuations in the operating status of the leased property. Multi-dimensional feature fusion makes asset evaluation more refined through multi-level data processing and analysis. For example, the operating data of the leased property can reveal subtle differences in operation through feature extraction, such as different users' usage preferences for leased properties, and the demand for leased properties in different geographical regions. By combining the user's financial data with their business data, the present invention can more finely evaluate the risk status of the asset. By combining asset value prediction and asset risk assessment, the evaluation results are made more transparent and explainable. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings: Figure 1 A flowchart showing a method for asset evaluation based on multi-dimensional feature fusion according to an embodiment of the present invention is provided; Figure 2 A flowchart showing a method for extracting operation characteristics according to an embodiment of the present invention is shown; Figure 3 A flowchart showing a method for extracting asset evaluation features according to an embodiment of the present invention is provided; Figure 4 A flowchart showing a method for fusing asset evaluation features according to an embodiment of the present invention is provided; Figure 5 A flowchart showing the steps of an asset evaluation method according to an embodiment is shown; The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0026] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0027] In addition, the drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0028] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0029] The operation data of the electric vehicle includes: 30,000 kilometers of driving, 1,200 charging times, and 4 maintenance records. The driving status data of the electric vehicle is collected through sensor data, GPS data, and the vehicle system. Through time series analysis, the operation data of the electric vehicle is segmented by time period, for example, the mileage and charging times of each month are recorded. Assuming the segmented analysis is performed by month, the segmented data obtained are as follows: Month 1: 1,500 kilometers of driving, 100 charging times; Month 2: 2,000 kilometers of driving, 120 charging times; Month 3: 2,200 kilometers of driving, 130 charging times. When analyzing the operation data, the event recognition algorithm is used to identify key events. For example, in the second month, a battery failure event occurred, which was recorded as "battery failure event" and marked as an operation event. Feature extraction is performed on the identified operation events (such as "battery failure event"), including the frequency of event occurrence, duration, time period of occurrence, etc. The event features are as follows: Battery failure event: frequency of occurrence 1 time / month, average duration 3 hours. According to the climate data (such as temperature, humidity, etc.) of the area where the electric vehicle is located, the electric vehicle is operated in a warm climate zone with an average annual temperature of 25°C and humidity of 60%. The climate condition data is combined with the characteristic data of the operation event to extract the characteristics that affect the asset value. After the combined analysis, the characteristic data obtained are as follows: the coefficient of temperature on battery performance: -0.5, the negative impact coefficient of battery failure events on value: -0.3.

[0030] The user's business data includes: monthly rent payment, usage frequency, etc.; financial data includes: user's monthly income of 10,000 yuan, credit score of 720, and debt ratio of 0.4. Based on the user's leasing behavior and payment history, business risk characteristics are extracted. For example, the user has not delayed rent payments in the past six months, so the risk assessment is low. The characteristic data is as follows: Risk score: 0.2 (low risk) According to the credit scoring model (weighted calculation based on various financial indicators), the user's credit ability is calculated. According to the calculation, the user's credit ability score is 85 / 100. Based on the user's credit score and financial status, the user's repayment intention is predicted using a machine learning model or regression algorithm. The prediction result is: Repayment intention score: 0.8 (high repayment intention).

[0031] The user's financial health status (such as income-to-expenditure ratio, debt status, etc.) is used to calculate the user's financial health index. Assuming the calculation result is: Financial Health Index: 75 (good), the user's asset evaluation feature data is generated based on the financial health data and repayment intention data. Assuming the feature data is as follows: User Financial Health Feature: Good, User Repayment Intention Feature: High.

[0032] Combine the operational characteristics of the leased property (such as battery failure rate, temperature impact) and the user's financial data (such as credit score, financial health index) to obtain feature matrix data. For example: [0.2, -0.5, -0.3, 85, 75, 0.8]. Use the algorithm to resolve conflicts in the feature matrix, for example, the relationship between the impact of temperature on the battery and the failure rate is processed to obtain the merged feature data. For example: preliminary evaluation feature: -0.2 (adjusted temperature impact). Align and fuse all feature data according to certain rules to obtain asset evaluation feature data. For example, evaluation feature: 0.7 (comprehensive feature after fusion).

[0033] Use a prediction model (such as linear regression, neural network, etc.) to predict the market value of electric vehicles based on the evaluation feature data. For example, the prediction result is: Predicted asset value: 15,0000 yuan. According to the asset evaluation feature data, perform risk assessment and output the risk coefficient of the asset. For example, the risk assessment result is: Risk assessment: 0.1 (low risk). Integrate the asset value with the risk assessment result to obtain the asset evaluation result. For example, the asset evaluation data is: Asset evaluation data: asset value 15,0000 yuan, risk coefficient 0.1.

[0034] See also Figures 1 to 5 The present application provides an asset evaluation method based on multi-dimensional feature fusion, the method comprising: S1. Obtaining basic operating data of the leased property, and extracting operating characteristics based on the basic operating data of the leased property to obtain first asset evaluation characteristic data; In one embodiment, various basic operating data of the leased property are collected, including frequency of use, maintenance history, service life, equipment status, operating environment data, etc. The data is generated by sensors, equipment management systems or manual records. By analyzing the historical operating data of the leased property, key operating characteristics are extracted. For example, the average failure time of the equipment, the working efficiency of the equipment, the failure rate, the maintenance cycle, the load rate of the equipment, etc. The data is statistically analyzed to calculate the cumulative usage time, repair cycle, and equipment usage cost of the leased property. The extracted operating characteristic data (such as equipment failure rate, frequency of use, maintenance records, etc.) are organized into a structured data set to form the first asset evaluation characteristic data. The data set not only contains the basic status of the equipment, but also can include the market value and maintenance cost of the equipment.

[0035] S2, obtaining user business data and user financial data, and extracting asset evaluation features based on the user business data and user financial data to obtain second asset evaluation feature data; In one embodiment, business data related to the user and the leased property is collected, such as the user's lease contract data, lease term, payment method, payment history, default record, etc. The user's financial health data is collected, including income status, debt ratio, financial statements, cash flow, balance sheet, etc. Features related to asset evaluation are extracted from the user's business data, such as the user's lease payment history, business activity, default record, etc. The user's financial health status is extracted from the financial data, such as income stability, debt ratio, profitability, debt repayment ability, etc. The extracted business features and financial features are integrated into the second asset evaluation feature data. The data set contains information such as the user's credit status and repayment ability, which is used to analyze the risk and value of the asset.

[0036] S3, performing multi-dimensional fusion on the first asset evaluation feature data and the second asset evaluation feature data to obtain asset evaluation feature data; In one embodiment, the first asset evaluation feature data (such as equipment failure rate, work efficiency, etc.) is merged with the second asset evaluation feature data (such as user credit, debt repayment ability, etc.). The operational characteristics of the leased object and the financial characteristics of the user are combined to form a comprehensive evaluation data set. The feature matrix is ​​spliced, that is, the various features extracted from steps S1 and S2 are spliced ​​into a high-dimensional feature vector. For example, if the first data set contains 3 features (failure rate, maintenance history, and usage frequency) and the second data set contains 4 features (user credit score, repayment ability, debt ratio, and income stability), then the spliced ​​feature matrix will contain 7 features. During the feature fusion process, conflicts or redundancies between certain features may occur (for example, the failure frequency of the equipment may be highly correlated with the maintenance history). Through methods such as correlation analysis and principal component analysis (PCA), these redundancies are eliminated and the most representative features are retained. Features from different sources are aligned (to ensure consistency in time or other dimensions) and fused in a weighted manner. The weights can be set according to the relevance or importance of each feature to ensure that the contribution of each feature to the final evaluation result is reasonable.

[0037] S4. Perform asset evaluation based on the asset evaluation characteristic data to obtain asset evaluation data.

[0038] In one embodiment, multiple regression analysis or machine learning models are used to predict the value of the fused asset evaluation feature data. For example, the current value or future value of the asset is predicted through data such as the operating characteristics of the leased object, the credit score of the user, and the financial health status. The predicted asset value reflects the market value of the leased object under current conditions. The fused feature data is evaluated for asset risk using models such as logistic regression and decision trees. For example, based on the user's credit status, repayment history, and the operating status of the leased object, the risks faced by the asset (such as default risk, market volatility risk, etc.) are evaluated. The risk assessment results are presented in the form of risk scores. For example, a risk score of 0.2 indicates low risk and 0.8 indicates high risk. The results of asset value prediction and asset risk assessment are integrated to obtain asset evaluation data. The integration can be achieved by methods such as weighted average, and the weights can be adjusted according to the type of asset, market conditions, or corporate strategy. The generated asset evaluation data will include the predicted value of the asset, risk score, and other relevant evaluation results, providing decision makers with a comprehensive basis for asset evaluation.

[0039] Optionally, S1 includes: S11. Obtaining basic operating data of leased property; In one embodiment, basic operation data of the leased item is obtained from the equipment management system, sensor data, log records or other operation monitoring platforms. Such data include but are not limited to the equipment's operating time, fault records, maintenance history, usage frequency, power consumption, etc. The data can be collected in real time by sensor equipment, or exported through manual records and equipment management platforms.

[0040] S12, performing segmented analysis on the basic operation data of the leased property to obtain segmented operation data of the leased property; In one embodiment, the operating data of the leased object is divided into several time periods according to time. For example, the data is divided into time periods of different granularities such as days, weeks, and months according to the operating time of the equipment. The segmentation points are marked by detecting the time points where major changes occur in the data (such as failures, maintenance records, etc.). For data related to the location of the leased object (such as mobile equipment, transportation tools, etc.), segmentation is performed according to the geographical location changes of the equipment. For example, the use of the equipment in different geographical locations is recorded, and the data is segmented according to the region or operation scenario. The state switching of the equipment (such as operation, stop, maintenance, failure, etc.) is another important segmentation criterion. The data is cut according to the state change points of the equipment to obtain segmented data of the state switching. The time segmentation data, geographical segmentation data, and state switching data are clustered by a clustering algorithm (such as K-means or DBSCAN), and the segmented data is further refined so that each segment of data presents similarity under a certain dimension. The clustering process can select appropriate features for clustering according to the working mode or failure mode of the equipment. All segmented data (time period, geographic segment, status segment) are integrated to form the operational segmented data of the leased property. Each segment represents the operational characteristics of the leased property at a specific time, location or status.

[0041] S13, performing event recognition on the leased property operation segment data to obtain leased property operation event data; In one embodiment, the meaning of "event" is defined. For example, equipment failure, overload operation, downtime maintenance, etc. can be regarded as important operational events. Operational events can not only be equipment failures, but also equipment reaching a certain performance threshold (such as excessive temperature, abnormal power consumption, etc.). According to the sensor data of the device, thresholds or rules are set to detect specific events. For example, if the operating temperature of the equipment exceeds the safety threshold, or the power consumption suddenly increases, it can be identified as an "overheating" or "high power consumption" event, and the occurrence of the event can be automatically identified by anomaly detection or time series analysis methods based on statistical methods. For each segment of data, if a defined operational event occurs within the segment, the event is marked, and the event type, start time, end time and other attributes related to the event (such as the impact of the event, duration, etc.) are specified. Aggregating the identified event data to form an operational event data set for the leased property will help assess the health status and potential risks of the leased property.

[0042] S14, extracting operation event features from the leased property operation event data to obtain leased property operation event feature data; In one embodiment, meaningful features are extracted from identified operational events. For example, for "equipment failure" events, features such as the frequency of occurrence of the failure, the type of failure, the duration of the failure, and the length of time it takes to repair the failure are extracted; for "overload" events, features such as the maximum value of the equipment overload, the duration, and the frequency of overload occurrence are extracted. Time-related features are extracted based on information such as the time of occurrence of the event, the interval between events, and the state of the event (such as the performance of the equipment before and after the failure). For example, features such as changes in equipment load before and after the event, and equipment recovery time. For events related to equipment performance degradation (such as aging, wear, etc.), features are extracted by statistically analyzing the changing trends of the equipment's performance indicators (such as output power, efficiency, rotation speed, etc.). All extracted operational event features are organized into a structured data set to form operational event feature data for the leased property.

[0043] S15. Obtaining the climate condition data of the leased property corresponding to the basic operation data of the leased property; In one embodiment, climate condition data related to the leased property is obtained, such as temperature, humidity, precipitation, wind speed, etc. These data come from weather forecast services, on-site climate monitoring equipment, or historical climate data records. The climate condition data is synchronized with the basic operating data of the leased property to ensure that the operating data and climate condition data of the leased property can be effectively compared and analyzed at the same time point.

[0044] S16. Perform multimodal feature extraction based on the climate condition data of the leased property and the characteristic data of the operation events of the leased property to obtain first asset evaluation characteristic data.

[0045] In one embodiment, new features are extracted by combining climate condition data and operational event feature data. For example, the failure rate of equipment in high temperature or high humidity environments, changes in the operating efficiency of equipment under specific climate conditions, etc. The impact of climate conditions on the operation of the leased property is combined with the operational event features, and fused through weighted average, principal component analysis, etc. to generate new fused features. The fused features can describe the comprehensive performance of the leased property under different environmental conditions. The multimodal features (including operational event features and climate condition impacts) are organized into the first asset evaluation feature data. The data set will contain the operational performance of the leased property under different climate conditions and provide a basis for subsequent asset evaluation.

[0046] Optionally, the segmented analysis includes: Perform time series segmentation analysis on the basic operating data of the leased property to obtain the first segmented operating data of the leased property; In one embodiment, the time series data of the leased object, such as the working time, temperature, power, etc. of the equipment, is obtained. The data is arranged in chronological order, and the time interval can be seconds, minutes, hours, etc., which is determined according to the granularity of specific data collection. In order to enhance the expressiveness of the time series data, additional time characteristics are added through feature engineering. For example, features such as "day of the week", "hour", and "holiday mark" are added to each timestamp to help the model identify periodic or non-periodic patterns. By analyzing the trend changes, mutation points, and outliers of the time series data, thresholds are set to divide time periods. The time points with large data fluctuations are detected through a sliding window as the splitting points (such as when a fault occurs, when the operating status of the equipment changes significantly). According to the preset rules and the detected splitting points, the data is cut into multiple time periods according to the time interval. For example, the operating data of the equipment is cut by day, week, and month to generate multiple subsets, each of which represents an independent time period. Through the above processing, each segment of data is marked and organized to form the first leased object operation segmented data. Each segmented data contains information such as the status and performance of the equipment during the period.

[0047] Performing geographical location segmentation analysis on the basic operating data of the leased property to obtain the second leased property operating segmentation data; In one embodiment, geographic location information related to the leased item is obtained, including GPS coordinates (longitude, latitude), the operating area of ​​the equipment (such as factory area, city block, etc.), and the use of the equipment in different geographic locations. The space is divided into multiple areas according to the actual operating range of the equipment. For example, the area where the equipment is located is divided into several grids according to the GPS coordinates, and each grid represents a geographical area. For fixed equipment, the area is divided according to the installation location of the equipment. The data is marked according to the location information of the equipment. When the location of the equipment changes, a new data segmentation is triggered. The specific segmentation rules are determined according to the movement trajectory of the equipment, regional changes, etc. For example, if the equipment enters a new geographical area, the system will automatically divide the data into different regional segments. Through the above-mentioned geographical segmentation processing, the operating data in each geographical area is divided into different segments. The generated segmented data can describe the working status of the equipment in different locations, including information such as load and environmental conditions during operation.

[0048] Perform state switching segment analysis on the basic operation data of the leased property to obtain the third leased property operation segment data; In one embodiment, different working states of the equipment (such as running, stopping, maintenance, fault, etc.) are defined according to the use requirements and maintenance standards of the equipment. The state needs to be extracted from the operation data, relying on the sensor data of the equipment (such as temperature, load, current, etc.) and external events (such as shutdown maintenance, fault repair, etc.). According to the operation data of the equipment, the change point of the equipment state is detected. The state switching is manifested as a significant change in certain characteristic values. For example, when the equipment switches from the running state to the shutdown state, it is manifested as a sharp drop in power consumption or a sudden drop in the temperature of the equipment. Use the threshold method or time series analysis method to identify these switching points. Each time the state of the equipment changes, the state switching point is automatically marked, and the data is divided into different state segments. The data in each segment will correspond to the operation data of the equipment in a certain state, such as data in the operation of the equipment, data in the shutdown maintenance, etc. The data of each state segment is integrated to form the third lease operation segment data. The segmented data will help analyze the performance of the equipment in different states and its changing trends.

[0049] Perform clustering calculations on the first leased property operation segment data, the second leased property operation segment data, and the third leased property operation segment data to obtain first leased property operation segment clustering data, second leased property operation segment clustering data, and third leased property operation segment clustering data, respectively; In one embodiment, meaningful features are extracted from each segmented data, and the features include time, geography and state-related attributes such as the operating efficiency, failure rate, and load change of the equipment. The three types of segmented data of time, geography and state are merged to form a data set containing multi-dimensional features. Each segmented data point will include time features, geographical features, state features, etc. A clustering algorithm is applied to perform cluster analysis on these multi-dimensional data. The clustering goal is to automatically cluster similar segmented data into one category based on the performance of the equipment at different times, places, and states. Commonly used methods here include K-means clustering, DBSCAN or hierarchical clustering, etc. According to the clustering results, each segmented data is classified. For example, the first leased operation segment clustering data will cluster similar time period data together, the second leased operation segment clustering data will cluster operation data with similar performance in the same area, and the third leased operation segment clustering data will cluster operation data in the same state.

[0050] A graph is constructed based on the first leased property operation segment clustering data, the second leased property operation segment clustering data, and the third leased property operation segment clustering data to obtain leased property operation segment data.

[0051] In one embodiment, each segmented data is regarded as a node in the graph according to the clustering results. Different segmented data nodes are connected by similarity or correlation. For example, if the state changes of the equipment in two time periods are similar, they can be connected by an edge. The nodes in the graph represent different operational segments, and the weight of the edge can be calculated based on the similarity between the nodes. The similarity can be measured by distance measurement, correlation measurement, etc. For example, if two segments are very close in time or overlap in geographical location, the edge weight between them will be larger. After the graph is constructed, the graph structure can be further processed by the graph optimization method to remove noise data and optimize the connection relationship between nodes, so that the graph is more compact and reflects the operation law of the leased object. After the graph is constructed, a complete leased operation segmented data graph can be obtained, in which each node represents an operation segment and each edge represents the relationship between different segments.

[0052] Optionally, the time series segmentation analysis includes: Perform time-characteristic enhancement processing on the basic operation data of the leased property to obtain time-enhanced feature data; In one embodiment, time series operational data of the leased property is extracted, such as key indicators such as equipment power, temperature, speed, and load at each time point. The timestamp of the data is a key attribute for such analysis, and is usually collected in seconds, minutes, or hours. In order to enhance the representation ability of time series data, a series of time characteristics are added to the data. For example, time dimension features such as "hour", "day of the week", and "whether it is a holiday" are added to capture the pattern of periodic changes. In addition, the statistical features of the sliding window (such as mean, standard deviation, maximum value, minimum value, etc.) are introduced to further enhance the time sensitivity of the data. Signal processing methods are used to smooth the data, remove noise data, and enhance trends and seasonal changes, which is accomplished by calculating moving averages, exponential weighted averages, and other methods. The above-mentioned processed features are added to the original time series data to generate an enhanced feature data set, and the enhanced features help capture the periodic changes in equipment operation.

[0053] Performing multi-scale pattern recognition on the time-enhanced feature data to obtain pattern recognition feature data; In one embodiment, a suitable multi-scale range is set according to the operating law of the equipment. For example, there may be short-cycle behaviors (such as high-frequency vibration of the equipment) and long-cycle behaviors (such as seasonal changes or annual maintenance cycles). Patterns of different scales are identified in the data. Multi-scale window segmentation is performed on the time series. For example, for short-cycle fluctuations, a shorter window (such as hourly units) is selected, and for long-cycle changes, a longer window (such as monthly units) is selected. In each window, the key statistics of the data (such as mean, variance, maximum value, minimum value, etc.) are calculated to extract its pattern. By analyzing the data at different scales, through threshold judgment or mapping by a preset parameter library, a statistically significant fluctuation pattern is identified. For example, some equipment will show similar load fluctuations in each working cycle, but have different performances in different seasons. Different time patterns are identified by calculating the pattern similarity or variation within the window. Based on the data pattern recognition results of different scales, through spectrum analysis and periodicity recognition, or, change point detection and outlier analysis, representative features (capable of reflecting the data pattern recognition results) such as periodic changes, sudden anomalies, etc. are extracted.

[0054] Performing dynamic segmentation point detection on the pattern recognition feature data to obtain dynamic segmentation point data; In one embodiment, it is defined which changes or features will become segmentation points. For example, a sharp change in the load, temperature or other key indicators of the equipment (such as sudden failure, equipment switching state, etc.) can be used as a potential segmentation point. Based on the pattern recognition feature, the degree of change of the data is calculated, that is, the change trend of the characteristic data over time, and the magnitude of the change is determined by calculating the rate of change of the data (such as the first-order difference) or calculating its standard deviation, variance and other statistics through a sliding window. The segmentation point is dynamically determined by setting a threshold or detecting significant fluctuations in the degree of change. For example, when the operating state of the equipment changes dramatically, or when a certain characteristic value exceeds a predetermined threshold, this time point can be marked as a segmentation point. In order to avoid over-segmentation or ignoring important change points, a smoothing algorithm (such as median filtering or Kalman filtering) is used to optimize the detected segmentation points, eliminate noise interference, and ensure the effectiveness and accuracy of the segmentation points. Through dynamic segmentation point detection, a segmentation point sequence is generated to represent each state switching point or behavior change point during the operation of the equipment.

[0055] Perform spatiotemporal segmentation fusion on the dynamic segmentation point data to obtain spatiotemporal segmentation data; In one embodiment, the behavior changes of the device in different geographical locations are an important supplement to the changes in the time series. By combining the geographical location information (such as the GPS coordinates of the device), each time segment point is combined with the corresponding geographical location data to construct a spatiotemporal feature. Use the segment point detection results and geographical information to construct a spatiotemporal data set. For example, the time features corresponding to each segment point are fused with the spatial features (such as the area or location where the device is located) to form spatiotemporal feature data. According to the spatiotemporal data, the correlation between different time and space segments is analyzed. By fusing the spatiotemporal segments, the errors caused by single-dimensional data are reduced, and the model's adaptability to complex spatiotemporal changes is enhanced. Based on the fused spatiotemporal data, spatiotemporal segmented data is generated. These data can more accurately reflect the operating status of the device under different time and space conditions.

[0056] Generating segmentation features for the spatiotemporal segmentation data to obtain segmentation feature data; In one embodiment, the most representative features are selected from the spatiotemporal segmented data. For example, the load change rate, temperature change, fault occurrence frequency and other features of the equipment are selected as the feature values ​​of each segment. In each spatiotemporal segment, the statistics of each key feature, such as the mean, standard deviation, maximum value, minimum value, etc., are calculated, the correlation between the features is calculated, and the key indicators that can represent each segment are further extracted. Multiple segment features are aggregated to form a unified feature set, for example, time features, geographic features, state features, etc. are weighted averaged or weighted summed to obtain segment features. Through the above-mentioned feature calculation and aggregation, feature data of each segment is generated, and the feature data provides a detailed description for hierarchical grouping and cluster analysis.

[0057] Hierarchical grouping is performed according to the segment characteristic data to obtain the first leased property operation segment data.

[0058] In one embodiment, hierarchical clustering analysis is performed on the segmented feature data, and different operating segments are grouped into the same group based on the similarity of the data. At this time, hierarchical clustering is used to help automatically identify segments with similar operating modes, avoiding artificial setting of the number of segments. The segmentation results are optimized by calculating the differences within each group and the similarities between groups. The similarity of the data within the group is maintained as much as possible to avoid excessive overlap or over-division. Through hierarchical grouping, the first leased item operation segmentation data is obtained. The segmented data will have similar time, geographic, and status characteristics, which are used for asset evaluation, risk prediction and other analyses.

[0059] Optionally, the geographic location segmentation analysis includes: Perform multi-source geographic data fusion on the basic operation data of leased assets to obtain geographic fusion data of leased assets; In one embodiment, the geographic data of the leased property is obtained from multiple data sources (such as GPS, geographic information system (GIS), sensors, etc.). The data includes but is not limited to the location coordinates of the equipment, environmental conditions, regional characteristics, etc. The operational data of the leased property is fused with the geographic data. Specifically, each operating time point of the leased property is matched with the corresponding geographic coordinate information. The "geographic fusion data" obtained after fusion includes the geographic location (longitude, latitude), timestamp and equipment status of each operating data point. If there is missing geographic data, it is supplemented by interpolation, extrapolation or the use of other sensor data to ensure the integrity of the data. The obtained geographic fusion data set includes detailed information on multiple dimensions such as time, space, and equipment status, providing a basis for subsequent segmented analysis.

[0060] Dynamically mark the geographical location of the leased object geographical fusion data to obtain the geographically marked data of the leased object; In one embodiment, the geographical location of the leased property is dynamically marked based on the geographical coordinates of each operating data point. For example, the location of the equipment is mapped to a specific geographical area (such as a city, block or facility location) using latitude and longitude information, and geographical area marking is performed by introducing a map API (such as Google Maps, Baidu Maps, etc.). The location of the equipment may change over time, so the geographical location marking is dynamic. By regularly updating the geographical location of the equipment, its location is reflected in real time. For further processing and analysis, the geographical tag data is standardized into a unified coordinate system to ensure data consistency. For geographical data from different sources, a unified coordinate system is used for conversion to ensure the accuracy of the marking. The resulting geographical tag data set includes the precise location of the equipment at each time point, the geographical area label, and related time information.

[0061] Performing multi-dimensional geographic region mapping on the geographical tag data of the leased object to obtain multi-dimensional region mapping data; In one embodiment, the region is divided into multiple levels of geographical regions based on geographic tag data. For example, it can be divided from multiple dimensions such as global, provincial, municipal, district, street, etc. According to the geographical location of the equipment, it is mapped to these levels. By combining different geographical level information, a multi-dimensional regional mapping is constructed. For example, at the city level, it can be further subdivided into residential areas, commercial areas, industrial areas, etc.; at the regional level, it can be refined into specific buildings, workshops, transportation routes, etc. In each geographical area, the spatial characteristics of the area are extracted, such as traffic density, weather conditions, terrain characteristics, etc. Through these multi-dimensional mappings, a data set containing different geographical locations of the leased objects and their corresponding spatial characteristics is generated. The multi-dimensional regional mapping data will provide information for regional clustering and behavioral feature extraction.

[0062] Perform regional clustering optimization based on multi-dimensional regional mapping data to obtain preliminary regional clustering data; In one embodiment, features related to regional operations, such as geographic location, operating hours, equipment density, etc., are selected as inputs for cluster analysis. Based on multidimensional regional mapping data, clustering methods (such as K-means, DBSCAN, etc.) are used to group geographic regions. The goal of clustering is to classify regions with similar operating characteristics into the same category. For example, equipment operation in some areas is affected by traffic congestion, while other areas are affected by weather or facility density. Different geographic regions are divided into multiple cluster groups based on the similarity of regional characteristics. Each cluster group represents a class of regions with similar operating characteristics. The generated preliminary regional clustering data will include a cluster label for each geographic region, as well as the characteristic data of the region.

[0063] Dynamically adjust the boundaries according to the preliminary regional clustering data to obtain optimized regional clustering data; In one embodiment, since the morphology of the geographical area is dynamic, the boundaries of the preliminary clustering will no longer adapt to the actual situation due to changes in time or external environment. Therefore, the boundaries of the region are dynamically adjusted. For example, considering factors such as seasonal changes, holidays or emergencies, the cluster boundaries are adjusted to better reflect the operating conditions of the region. Boundary adjustments can be made based on the trend of data changes. For example, if equipment in certain areas frequently interrupts operation, while other areas operate stably, the areas with more interruptions can be separated from a cluster, or the boundaries of the clusters can be redefined. By adjusting the boundaries of the preliminary clustering results, the heterogeneity between regions is better reflected, and the degree of aggregation between similar regions is improved. The optimized regional clustering data obtained is more in line with the actual geographical and operational laws than the preliminary clustering data, providing more accurate regional divisions for analysis.

[0064] Extract operational behavior features based on optimized regional clustering data to obtain regional behavior feature data; In one embodiment, the operational behavior characteristics of each region are extracted from the optimized regional clustering data. For example, the frequency of use, failure rate, operating efficiency, etc. of the equipment in each region are analyzed. Other regional-related features such as weather, traffic, personnel density, etc. can also be extracted. Statistical analysis is performed on the equipment operation behavior in each regional cluster, such as calculating indicators such as the average equipment operating time, average failure rate, and load of equipment in the region for each region. The operating modes of different regions are identified by extracting behavioral characteristics. Some regions have a high frequency of equipment use, while other regions have less equipment density, but higher equipment operating efficiency. The behavioral characteristics will serve as the basis for analyzing the operating conditions of different regions and provide support for geographic semantic segmentation.

[0065] Generate geographic semantic segmentation for regional behavior feature data to obtain geographic semantic segmentation data for leased items; In one embodiment, the rules for geographic semantic segmentation are defined based on the extracted regional behavioral features. For example, several semantic segments are created based on the workload, failure frequency or efficiency of regional equipment, such as "efficient area", "inefficient area", "fault-prone area", etc. The regions are segmented according to the rules using the previously extracted behavioral feature data. For example, if the equipment in certain areas fails frequently, it is marked as a "fault area"; if the equipment in certain areas has a high operating efficiency, it is marked as an "efficient area", and a corresponding semantic label is assigned to each area. Detailed geographic semantic segmentation data for leased items is generated by analyzing the regional behavioral features.

[0066] Operation segmentation labels are generated according to the geographic semantic segmentation data of the leased object to obtain second operation segmentation data of the leased object.

[0067] In one embodiment, an operational segmentation label is generated for each geographic area based on the geographic semantic segmentation data and the operational behavior characteristics of the leased property. The label will reflect the operational status of the leased property in different geographic locations, such as "high load", "low load", "frequent failures", etc. The label is further optimized based on the actual operation of the equipment. For example, if the load of equipment in certain areas exceeds a predetermined threshold, the labels of these areas need to be adjusted from "low load" to "high load". The generated second leased property operational segmentation data will contain the operational segmentation label for each geographic area, as well as the operational characteristic data of the area, to help further asset evaluation and optimization.

[0068] Optionally, the state switching segment analysis includes: Mark the status characteristics according to the basic operation data of the leased object to obtain the status marking data of the leased object; In one embodiment, the operational data of the leased object is collected, and the data includes various operational indicators of the leased object, such as temperature, pressure, load, speed, operating time, etc. Different status characteristics are defined according to the various operational indicators of the leased object. The status characteristics are set based on thresholds, for example, normal state: when the operational indicators of the leased object are within the normal range, it is defined as "normal". Abnormal state: when certain indicators exceed the preset threshold, it is defined as "abnormal". Fault state: when certain indicators trigger a fault alarm, it is defined as "fault". According to the defined status characteristics, the operational data at each time point is marked. For example, if the temperature at a certain moment exceeds the set upper limit, the state corresponding to the time point is "abnormal", and the obtained state marking data contains the state category information of the leased object at each time point, forming a time-series state data sequence.

[0069] Perform state switching detection according to the leased object state mark data to obtain the leased object state switching point data; In one embodiment, it is necessary to set a standard for state switching. For example, when the leased item switches from a "normal" state to an "abnormal" state, or from an "abnormal" state to a "faulty" state, it is considered a state switch. By comparing the state marks of adjacent time points, the moment of state change is detected. If the state at the current moment is inconsistent with the state at the previous moment, it is considered that a state switch has occurred. Sometimes, multiple state changes in a short period of time do not represent actual state switches (for example, false switches caused by instantaneous fluctuations). Therefore, it is necessary to set a minimum switching interval time to avoid redundant switching points. Through the detection process, all state switching points are calibrated, and these switching points will serve as basic data for analysis.

[0070] Performing a switching event context analysis on the leased item state switching point data to obtain leased item switching context data; In one embodiment, each state switch is not only a change of state, but also related to specific operating conditions and background environment. The definition of "context" can consider the following: the device state value before and after the switch occurs (such as temperature, pressure, etc.). External environmental factors (such as weather, workload, etc.). Historical information such as the use time of the device and the accumulated running time. For each state switching point, the context of the state switch is constructed by extracting the operating data at that moment and a certain time range before and after. For example, if the device switches from the "normal" state to the "abnormal" state at a certain moment, the data points before and after the moment (such as device load, temperature, etc.) are extracted and analyzed together with the external conditions at that moment (such as temperature, humidity, etc.). Through context analysis, the commonalities and rules of different switching events are identified. For example, it is found that a certain device is more likely to switch states under high load, or has a higher failure rate in a high temperature environment. Based on the context analysis of each switching point, a data set containing the switching point and its corresponding context information is generated to provide detailed background information for the generation of state fragments.

[0071] Generate a state fragment according to the leased object switching context data to obtain leased object state fragment data; In one embodiment, a state segment refers to a state in which the leased item is always in a certain state (such as "normal" or "faulty") within a certain time window. Therefore, the state data is segmented and continuous identical state values ​​are grouped into one segment. The state data is divided into multiple segments in chronological order according to the detected state switching points and the switching context data. For example, if the device is always in the "normal" state between time periods T1 and T2, a "normal" state segment can be generated; if it is in the "faulty" state between T3 and T4, a "faulty" state segment is generated. The boundaries of the state segments can be dynamically adjusted according to the context data. For example, if it is found that the device is in the "faulty" state for more than a certain period of time, the segment can be marked as a long-term fault segment, otherwise, it is a short-term fault segment. According to the above rules, the start time, end time, state type and context information of each state segment are obtained.

[0072] Extracting behavior state features from the leased item status fragment data to obtain behavior state feature data of the leased item; In one embodiment, for each state segment, behavior-related features are extracted. For example, the duration of the state segment. The rate of change of various indicators before and after the state change. Influencing factors of the external environment in the state segment (such as temperature, humidity, workload, etc.). The operating efficiency and failure rate of the device in this state. The behavioral characteristics of each state segment are calculated. For example, if the average operating efficiency of the device in the "normal" state is 90%, then this behavioral feature can reflect the performance of the device in the normal state. Through the above calculations, the behavioral state characteristic data of each state segment is obtained.

[0073] Perform segmented semantic optimization on the behavior status feature data of the leased object to obtain optimized segmented semantic data; In one embodiment, semantic labels of state segments are defined based on behavioral state characteristics. For example, high-efficiency state: the device is in a "normal" state, lasts for a long time and has high operating efficiency. Fault warning state: the device is in an "abnormal" state, lasts for a long time and multiple warning signals appear. High-risk state: the device is in a "fault" state, the fault recovery time is long or faults occur frequently. The state segments are optimized based on the behavioral state characteristics to improve the accuracy of the labels. For example, if a device has a similar high fault frequency in multiple state segments, its semantic label is optimized to "high fault risk state". Generate a corresponding semantic label for each state segment to ensure that each state segment can be clearly identified and understood, and obtain optimized segmented semantic data, including semantic labels for the time range, type and behavioral characteristics of each state segment.

[0074] The leased object operation segment data is generated according to the optimized segmented semantic data and the leased object status fragment data to obtain the third leased object operation segment data.

[0075] In one embodiment, the optimized segmented semantic data is integrated with the original state segment data to form leased operation segment data, which includes not only the time information of each state segment, but also the state type, behavior characteristics, semantic labels, etc. of the segment. In combination with the operational background and specific needs of the leased object, more detailed labels are generated for each operational segment. For example, the "normal" state of the equipment will be refined into "high efficiency normal" or "low efficiency normal".

[0076] Optionally, S2 includes: S21. Obtain user business data and user financial data; In one embodiment, the user's business data and financial data are collected. The business data includes the user's business type, transaction history, contract terms, payment cycle, etc.; the financial data includes the user's income, expenditure, assets, liabilities, credit record, etc.

[0077] S22. Extract business risk characteristics based on user business data to obtain business risk characteristic data; In one embodiment, some potential risk factors are defined based on business data. For example, the user's transaction frequency, contract breach record, payment history, volatility of transaction amount, stability of business cycle, etc. all affect their business risk. By analyzing the user's transaction history, the transaction frequency, amount fluctuation, transaction interval time and other characteristics are extracted to obtain transaction behavior characteristic data. Check whether the user has a record of not paying on time or defaulting, mark the historical number of defaults, default amount and other information, and obtain default history characteristic data. Based on the business data, determine whether there is abnormal behavior, such as frequent changes in transaction accounts, frequent changes in business addresses, etc., and obtain credit abnormality characteristic data. By performing weighted calculation on these characteristics, a business risk score is obtained, which indicates the risk level of the user in the business, and the business risk score characteristic data is obtained. Combined with the above characteristics, the weighted values ​​of different risk factors are weighted or combined to generate a set of business risk characteristic data.

[0078] S23, calculating the credit capacity according to the business risk characteristic data to obtain credit capacity data; In one embodiment, a credit capacity assessment model is constructed based on business risk characteristics and financial status. The model can calculate the user's credit capacity based on historical data and taking into account various factors (such as business risk characteristics, income, expenditure, liabilities, etc.). Multidimensional data such as business risk characteristics and financial status are weighted. Each feature is standardized so that data of different units and dimensions can be effectively combined. Based on the weighted feature data, a credit scoring model (such as linear regression, decision tree, etc.) is used for calculation. The output of the model is the user's credit capacity score, which reflects his ability to perform on time. The user's credit capacity score is obtained through model calculation, and this score will be used for subsequent repayment intention prediction and financial health assessment.

[0079] S24, predicting the user's repayment intention based on the credit ability data and the user's financial data to obtain repayment intention prediction data; In one embodiment, repayment intention refers to the possibility of a user repaying a debt on time. Its influencing factors include not only credit ability score, but also closely related to financial status, spending plan, income fluctuations and other factors. Credit ability is an important feature for predicting repayment intention, and users with high credit scores will show a strong repayment intention. Financial status includes indicators such as the user's monthly income, debt-to-asset ratio, and working capital. These financial information directly affects the user's repayment ability and intention. Analyze the user's daily spending behavior. If the expenditure ratio is too large, it will affect their repayment intention. Use historical data to train the prediction model, considering the above characteristics (such as credit score, financial health status, income and expenditure status, etc.). Predict the user's repayment intention through machine learning methods (such as classification algorithms) or regression algorithms. The model output is the user's repayment intention score, indicating whether he will repay on time. Generate repayment intention data from the prediction results to indicate the user's repayment possibility and provide data support for further financial health assessment and asset assessment.

[0080] S25. Perform financial health assessment based on the business risk characteristic data and the user's financial data to obtain financial health data; In one embodiment, financial health refers to whether the user's financial situation is sound and whether there is sufficient liquidity and capital for daily operations and to deal with emergencies. Financial health indicators include asset-liability ratio, cash flow, investment return rate, debt repayment capacity, etc. Calculate the ratio of the user's assets to liabilities to determine whether its asset-liability structure is healthy. Analyze the user's cash inflows and outflows, especially the ratio of available cash flow to debt repayment. The user's liquidity indicators, such as savings, accounts receivable, outstanding debts, etc., affect their ability to deal with emergencies. By constructing a financial health scoring model, the above indicators are weighted and calculated to output a financial health score to represent the user's financial health status. Based on the output results of the evaluation model, the user's financial health data is generated.

[0081] S26. Generate asset evaluation characteristics using the financial health data and repayment intention prediction data to obtain second asset evaluation characteristic data.

[0082] In one embodiment, the user's financial health data and repayment intention prediction data are merged. The financial health score is combined with the repayment intention score to obtain an asset evaluation feature data set. These features are weighted according to actual needs. For example, the weights of the financial health score and the repayment intention score can be adjusted according to actual evaluation needs to ensure that the evaluation results reflect the user's true situation. The above features are integrated to form an asset evaluation feature matrix, which includes the user's business risk, credit ability, financial health status and repayment intention.

[0083] Optionally, S3 includes: S31, performing multi-dimensional feature combination according to the first asset evaluation feature data and the second asset evaluation feature data to obtain feature matrix data; In one embodiment, features from different sources are combined to construct a complete feature matrix. The rows of the feature matrix represent each data instance (such as each leased user), and the columns represent different feature items. The feature matrix contains the operational features of the leased property (such as working status, number of failures), climate conditions (such as temperature, humidity effects), financial health features (such as debt ratio, cash flow status), etc.

[0084] S32, performing feature conflict resolution based on feature matrix data to obtain preliminary asset assessment feature data; In one embodiment, in the feature matrix, there are conflicts or redundancies between certain features. For example, the financial health feature and the repayment intention prediction feature contain some similar information, resulting in duplication or contradiction of certain features. By analyzing the correlation between the features, identify which features are in conflict. For example, there is a high correlation between the credit score and the business risk score, and some financial health data has been covered by the repayment intention prediction model. Use statistical methods such as correlation coefficient and information gain to evaluate the relationship between the features and determine which features are redundant or conflicting. For conflicting or redundant features, two methods can be used to handle them: for features with high correlation, choose to delete one of the features to avoid multicollinearity between features. For features with less conflict, the conflicting features can be merged into a new feature by weighted average, principal component analysis, etc., to retain more information. After the above processing, a preliminary asset evaluation feature data set without significant conflict is obtained. The data set has been simplified and redundant or contradictory information has been reduced.

[0085] S33. Perform feature alignment and fusion on the preliminary asset assessment feature data to obtain asset assessment feature data.

[0086] In one embodiment, there are differences in the units, dimensions and dimensions of features between different data sources (such as leased property operation data and user financial data). For example, the operation data of leased property is recorded on a daily basis, while the financial data is summarized on a monthly basis. Therefore, it is necessary to align different features. Interpolate data of different time scales so that each feature is consistent in the time dimension. For example, linearly interpolate financial data on a daily basis, or average leased property operation data on a monthly basis. The aligned data needs to be further fused. The fusion method includes: for multiple similar features (such as operating status and environmental conditions), merge them by weighted average to obtain a comprehensive indicator. Alternatively, for features from different fields (such as leased property operation data and user financial data), combine them together by splicing to form a unified feature set. Alternatively, for some complex relationships, use nonlinear fusion methods, such as multi-layer perceptron (MLP) in neural networks to fuse features to capture the complex relationships between different features.

[0087] During the fusion process, the value ranges of features may differ, which will affect the fusion results. Therefore, after feature fusion, the feature data is normalized or standardized to ensure that all features are on the same scale to avoid excessive influence of certain features on the results. Through feature alignment and fusion data, a unified asset evaluation feature data set is obtained. The data set contains comprehensive information on leased property operation data and user financial data, providing comprehensive input for asset evaluation.

[0088] Optionally, S4 includes: S41, predicting asset value based on asset evaluation characteristic data to obtain asset value prediction data; In one embodiment, the asset evaluation feature data has been processed by multi-dimensional fusion, conflict resolution and alignment, and includes comprehensive lease operation features, user financial features and related environmental impact data. For asset value prediction, a regression model is selected for prediction, such as linear regression, support vector regression (SVR), decision tree regression, etc., or other deep learning models, such as neural network regression models. The asset evaluation feature data is input into the model through a training set generated based on historical data. The model learns the relationship between assets and these features through historical data, and then predicts the value of future assets. For regression models, mean square error (MSE) or root mean square error (RMSE) can be used as a loss function for optimization. After training, the model predicts the input feature data and outputs the predicted asset value. For example, the future market value, residual value or future profitability of the leased object is predicted. The asset value prediction data obtained is the expected market value, residual value or future income of each leased object or asset. These data provide a basis for the next step of risk assessment and decision-making.

[0089] S42, performing asset risk assessment based on the asset assessment characteristic data to obtain asset risk assessment data; In one embodiment, asset risk assessment involves predicting various risks that assets will face in the future, such as market risk, operational risk, financial risk, etc. A corresponding risk assessment model is established based on asset assessment feature data. Risk is assessed by statistical models (such as Logistic regression, decision tree, random forest) or machine learning models (such as support vector machine, K-nearest neighbor). From the asset assessment feature data, features related to risk assessment are selected. For example, the operating status of the leased property, historical failure rate, financial status of the user, credit score, etc. The features may also include other risk factors that affect the security and long-term value of the asset, such as the impact of natural disasters (climate condition data) on the asset, the possibility of asset damage, the risk of user default, etc. The model is trained using the asset risk conditions in historical data. The loss function can select the logarithmic loss in logistic regression, or the common loss function of other classification algorithms. After training, the model will predict the risk level of each asset and output the probability of a certain risk level, such as low risk, medium risk, and high risk. Each asset will be assessed as a different risk level (such as low, medium, and high), and a corresponding risk score will be given to reflect the potential risk of the asset in the future.

[0090] S43. Integrate the asset value prediction data and the asset risk assessment data to obtain asset assessment data.

[0091] In one embodiment, the asset evaluation data needs to integrate the value prediction and risk assessment of the asset, and the two data points provide a comprehensive asset status. The asset value prediction data provides the potential market value or residual value of the asset, while the asset risk assessment data provides the potential risk level faced by the asset. The asset value prediction and risk assessment are combined (direct data combination, expert engine-based data mapping, or weighted calculation) to obtain a comprehensive score that reflects the comprehensive status of the asset in the current economic environment.

[0092] Optionally, the present application further provides an asset evaluation system based on multidimensional feature fusion, which is used to execute the asset evaluation method based on multidimensional feature fusion as described above, and the asset evaluation system based on multidimensional feature fusion includes: An operation feature extraction module is used to obtain basic operation data of the leased object, and extract operation features based on the basic operation data of the leased object to obtain first asset evaluation feature data; An asset evaluation feature extraction module is used to obtain user business data and user financial data, and extract asset evaluation features based on the user business data and user financial data to obtain second asset evaluation feature data; An asset evaluation feature fusion module, used for performing multi-dimensional fusion according to the first asset evaluation feature data and the second asset evaluation feature data to obtain asset evaluation feature data; The asset evaluation module is used to perform asset evaluation based on asset evaluation characteristic data to obtain asset evaluation data.

[0093] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0094] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An asset evaluation method based on multi-dimensional feature fusion, characterized in that: The method comprises: S1. Obtaining basic operating data of the leased property, and extracting operating characteristics based on the basic operating data of the leased property to obtain first asset evaluation characteristic data; S2. Obtain user business data and user financial data, and extract asset evaluation features based on the user business data and user financial data to obtain second asset evaluation feature data; S3, performing multi-dimensional fusion on the first asset evaluation feature data and the second asset evaluation feature data to obtain asset evaluation feature data; S4. Perform asset evaluation based on the asset evaluation characteristic data to obtain asset evaluation data.

2. The method according to claim 1, characterized in that S1 includes: Obtain basic operating data of leased assets; Conduct segmented analysis on the basic operating data of the leased property to obtain segmented operating data of the leased property; Perform event recognition on the segmented data of leased property operation to obtain leased property operation event data; Extracting operation event features from the leased property operation event data to obtain the leased property operation event feature data; Obtain the leased property climate condition data corresponding to the leased property basic operation data; Multimodal feature extraction is performed based on the climate condition data of the leased object and the characteristic data of the operation events of the leased object to obtain the first asset evaluation characteristic data.

3. The method according to claim 2, characterized in that The segmented analysis includes: Perform time series segmentation analysis on the basic operating data of the leased property to obtain the first segmented operating data of the leased property; Performing geographical location segmentation analysis on the basic operating data of the leased property to obtain the second leased property operating segmentation data; Perform state switching segment analysis on the basic operation data of the leased property to obtain the third leased property operation segment data; Perform clustering calculations on the first leased property operation segment data, the second leased property operation segment data, and the third leased property operation segment data to obtain first leased property operation segment clustering data, second leased property operation segment clustering data, and third leased property operation segment clustering data, respectively; A graph is constructed based on the first leased property operation segment clustering data, the second leased property operation segment clustering data, and the third leased property operation segment clustering data to obtain leased property operation segment data.

4. The method according to claim 3, characterized in that The time series segmentation analysis includes: Perform time-characteristic enhancement processing on the basic operation data of the leased property to obtain time-enhanced feature data; Performing multi-scale pattern recognition on the time-enhanced feature data to obtain pattern recognition feature data; Performing dynamic segmentation point detection on the pattern recognition feature data to obtain dynamic segmentation point data; Perform spatiotemporal segmentation fusion on the dynamic segmentation point data to obtain spatiotemporal segmentation data; Generating segmentation features for the spatiotemporal segmentation data to obtain segmentation feature data; Hierarchical grouping is performed according to the segment characteristic data to obtain the first leased property operation segment data.

5. The method according to claim 3, characterized in that: The geographical location segmentation analysis includes: Perform multi-source geographic data fusion on the basic operation data of leased assets to obtain geographic fusion data of leased assets; Dynamically mark the geographical location of the leased object geographical fusion data to obtain the geographically marked data of the leased object; Perform multi-dimensional geographic region mapping on the geographical tag data of the leased object to obtain multi-dimensional region mapping data; Perform regional clustering optimization based on multi-dimensional regional mapping data to obtain preliminary regional clustering data; Dynamically adjust the boundaries according to the preliminary regional clustering data to obtain optimized regional clustering data; Extract operational behavior features based on optimized regional clustering data to obtain regional behavior feature data; Generate geographic semantic segmentation for regional behavior feature data to obtain geographic semantic segmentation data for leased items; Operation segmentation labels are generated according to the geographic semantic segmentation data of the leased object to obtain second operation segmentation data of the leased object.

6. The method according to claim 3, characterized in that The state switching segment analysis includes: Mark the status characteristics according to the basic operation data of the leased object to obtain the status marking data of the leased object; Perform state switching detection according to the leased object state mark data to obtain the leased object state switching point data; Performing a switching event context analysis on the leased item state switching point data to obtain leased item switching context data; Generate a state fragment according to the leased object switching context data to obtain leased object state fragment data; Extracting behavior state features from the leased item status fragment data to obtain behavior state feature data of the leased item; Perform segmented semantic optimization on the behavior status feature data of the leased object to obtain optimized segmented semantic data; The leased object operation segment data is generated according to the optimized segmented semantic data and the leased object status fragment data to obtain the third leased object operation segment data.

7. The method according to claim 1, characterized in that S2 include: Obtain user business data and user financial data; Extract business risk features based on user business data to obtain business risk feature data; Calculate credit capacity based on business risk characteristic data to obtain credit capacity data; Predicting the user's repayment intention based on the credit capacity data and the user's financial data to obtain repayment intention prediction data; Conduct financial health assessment based on business risk feature data and user financial data to obtain financial health data; The financial health data and the repayment intention prediction data are used to generate asset evaluation features to obtain second asset evaluation feature data.

8. The method according to claim 1, characterized in that S3 include: Perform multi-dimensional feature combination according to the first asset evaluation feature data and the second asset evaluation feature data to obtain feature matrix data; According to the characteristic matrix data, characteristic conflicts are resolved to obtain preliminary asset evaluation characteristic data; The preliminary asset evaluation feature data is aligned and fused to obtain the asset evaluation feature data.

9. The method according to claim 1, characterized in that: S4 includes: Predict asset value based on asset evaluation characteristic data to obtain asset value prediction data; Perform asset risk assessment based on asset assessment characteristic data to obtain asset risk assessment data; Asset valuation data is obtained by integrating asset value prediction data and asset risk assessment data.

10. An asset evaluation system based on multi-dimensional feature fusion, characterized in that: Used to execute the asset evaluation method based on multi-dimensional feature fusion as claimed in claim 1, the asset evaluation system based on multi-dimensional feature fusion comprises: An operation feature extraction module is used to obtain basic operation data of the leased object, and extract operation features based on the basic operation data of the leased object to obtain first asset evaluation feature data; An asset evaluation feature extraction module is used to obtain user business data and user financial data, and extract asset evaluation features based on the user business data and user financial data to obtain second asset evaluation feature data; An asset evaluation feature fusion module, used for performing multi-dimensional fusion according to the first asset evaluation feature data and the second asset evaluation feature data to obtain asset evaluation feature data; The asset evaluation module is used to perform asset evaluation based on asset evaluation characteristic data to obtain asset evaluation data.

Images